cpu.fx

graph():
    %arg0_1 : [#users=1] = placeholder[target=arg0_1]
    %arg1_1 : [#users=1] = placeholder[target=arg1_1]
    %arg2_1 : [#users=32] = placeholder[target=arg2_1]
    %arg3_1 : [#users=1] = placeholder[target=arg3_1]
    %cat : [#users=1] = call_function[target=torch.ops.aten.cat](args = ([%arg0_1, %arg0_1],), kwargs = {})
    %expand : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%arg1_1, [2]), kwargs = {})
    %arange : [#users=1] = call_function[target=torch.ops.aten.arange](args = (0, 160), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %mul : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%arange, -9.210340371976184), kwargs = {})
    %div : [#users=1] = call_function[target=torch.ops.aten.div](args = (%mul, 160), kwargs = {})
    %exp : [#users=1] = call_function[target=torch.ops.aten.exp](args = (%div,), kwargs = {})
    %slice_1 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%expand, 0, 0, 9223372036854775807), kwargs = {})
    %unsqueeze : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_1, 1), kwargs = {})
    %unsqueeze_1 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%exp, 0), kwargs = {})
    %slice_2 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%unsqueeze_1, 1, 0, 9223372036854775807), kwargs = {})
    %mul_1 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%unsqueeze, %slice_2), kwargs = {})
    %mul_2 : [#users=2] = call_function[target=torch.ops.aten.mul](args = (%mul_1, 1), kwargs = {})
    %sin : [#users=1] = call_function[target=torch.ops.aten.sin](args = (%mul_2,), kwargs = {})
    %cos : [#users=1] = call_function[target=torch.ops.aten.cos](args = (%mul_2,), kwargs = {})
    %cat_1 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%sin, %cos], -1), kwargs = {})
    %slice_3 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%cat_1, 0, 0, 9223372036854775807), kwargs = {})
    %slice_4 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_3, 1, 160, 9223372036854775807), kwargs = {})
    %slice_5 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%cat_1, 0, 0, 9223372036854775807), kwargs = {})
    %slice_6 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_5, 1, 0, 160), kwargs = {})
    %cat_2 : [#users=1] = call_function[target=torch.ops.aten.cat](args = ([%slice_4, %slice_6], -1), kwargs = {})
    %_param_constant0 : [#users=1] = get_attr[target=_param_constant0]
    %t : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant0,), kwargs = {})
    %_param_constant1 : [#users=1] = get_attr[target=_param_constant1]
    %addmm : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant1, %cat_2, %t), kwargs = {})
    %silu : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm,), kwargs = {})
    %_param_constant2 : [#users=1] = get_attr[target=_param_constant2]
    %t_1 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant2,), kwargs = {})
    %_param_constant3 : [#users=1] = get_attr[target=_param_constant3]
    %addmm_1 : [#users=22] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant3, %silu, %t_1), kwargs = {})
    %_param_constant4 : [#users=1] = get_attr[target=_param_constant4]
    %_param_constant5 : [#users=1] = get_attr[target=_param_constant5]
    %convolution : [#users=3] = call_function[target=torch.ops.aten.convolution](args = (%cat, %_param_constant4, %_param_constant5, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %view : [#users=2] = call_function[target=torch.ops.aten.view](args = (%convolution, [2, 32, 10, 4096]), kwargs = {})
    %var_mean : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem : [#users=1] = call_function[target=operator.getitem](args = (%var_mean, 0), kwargs = {})
    %getitem_1 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean, 1), kwargs = {})
    %add : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem, 1e-05), kwargs = {})
    %rsqrt : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add,), kwargs = {})
    %sub : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view, %getitem_1), kwargs = {})
    %mul_3 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub, %rsqrt), kwargs = {})
    %view_1 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_3, [2, 320, 64, 64]), kwargs = {})
    %_param_constant6 : [#users=1] = get_attr[target=_param_constant6]
    %unsqueeze_2 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant6, 0), kwargs = {})
    %unsqueeze_3 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_2, 2), kwargs = {})
    %unsqueeze_4 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_3, 3), kwargs = {})
    %_param_constant7 : [#users=1] = get_attr[target=_param_constant7]
    %unsqueeze_5 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant7, 0), kwargs = {})
    %unsqueeze_6 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_5, 2), kwargs = {})
    %unsqueeze_7 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_6, 3), kwargs = {})
    %mul_4 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_1, %unsqueeze_7), kwargs = {})
    %add_1 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_4, %unsqueeze_4), kwargs = {})
    %squeeze : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_1, 3), kwargs = {})
    %squeeze_1 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze, 2), kwargs = {})
    %squeeze_2 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt, 3), kwargs = {})
    %squeeze_3 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_2, 2), kwargs = {})
    %detach : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_1,), kwargs = {})
    %detach_1 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_3,), kwargs = {})
    %silu_1 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_1,), kwargs = {})
    %_param_constant8 : [#users=1] = get_attr[target=_param_constant8]
    %_param_constant9 : [#users=1] = get_attr[target=_param_constant9]
    %convolution_1 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_1, %_param_constant8, %_param_constant9, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_2 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant10 : [#users=1] = get_attr[target=_param_constant10]
    %t_2 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant10,), kwargs = {})
    %_param_constant11 : [#users=1] = get_attr[target=_param_constant11]
    %addmm_2 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant11, %silu_2, %t_2), kwargs = {})
    %slice_7 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_2, 0, 0, 9223372036854775807), kwargs = {})
    %slice_8 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_7, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_8 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_8, 2), kwargs = {})
    %unsqueeze_9 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_8, 3), kwargs = {})
    %add_2 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_1, %unsqueeze_9), kwargs = {})
    %view_2 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_2, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_1 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_2, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_2 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_1, 0), kwargs = {})
    %getitem_3 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_1, 1), kwargs = {})
    %add_3 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_2, 1e-05), kwargs = {})
    %rsqrt_1 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_3,), kwargs = {})
    %sub_1 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_2, %getitem_3), kwargs = {})
    %mul_5 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_1, %rsqrt_1), kwargs = {})
    %view_3 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_5, [2, 320, 64, 64]), kwargs = {})
    %_param_constant12 : [#users=1] = get_attr[target=_param_constant12]
    %unsqueeze_10 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant12, 0), kwargs = {})
    %unsqueeze_11 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_10, 2), kwargs = {})
    %unsqueeze_12 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_11, 3), kwargs = {})
    %_param_constant13 : [#users=1] = get_attr[target=_param_constant13]
    %unsqueeze_13 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant13, 0), kwargs = {})
    %unsqueeze_14 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_13, 2), kwargs = {})
    %unsqueeze_15 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_14, 3), kwargs = {})
    %mul_6 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_3, %unsqueeze_15), kwargs = {})
    %add_4 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_6, %unsqueeze_12), kwargs = {})
    %squeeze_4 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_3, 3), kwargs = {})
    %squeeze_5 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_4, 2), kwargs = {})
    %squeeze_6 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_1, 3), kwargs = {})
    %squeeze_7 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_6, 2), kwargs = {})
    %detach_2 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_5,), kwargs = {})
    %detach_3 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_7,), kwargs = {})
    %silu_3 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_4,), kwargs = {})
    %_param_constant14 : [#users=1] = get_attr[target=_param_constant14]
    %_param_constant15 : [#users=1] = get_attr[target=_param_constant15]
    %convolution_2 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_3, %_param_constant14, %_param_constant15, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_5 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution, %convolution_2), kwargs = {})
    %div_1 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_5, 1.0), kwargs = {})
    %view_4 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_1, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_2 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_4, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_4 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_2, 0), kwargs = {})
    %getitem_5 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_2, 1), kwargs = {})
    %add_6 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_4, 1e-06), kwargs = {})
    %rsqrt_2 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_6,), kwargs = {})
    %sub_2 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_4, %getitem_5), kwargs = {})
    %mul_7 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_2, %rsqrt_2), kwargs = {})
    %view_5 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_7, [2, 320, 64, 64]), kwargs = {})
    %_param_constant16 : [#users=1] = get_attr[target=_param_constant16]
    %unsqueeze_16 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant16, 0), kwargs = {})
    %unsqueeze_17 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_16, 2), kwargs = {})
    %unsqueeze_18 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_17, 3), kwargs = {})
    %_param_constant17 : [#users=1] = get_attr[target=_param_constant17]
    %unsqueeze_19 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant17, 0), kwargs = {})
    %unsqueeze_20 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_19, 2), kwargs = {})
    %unsqueeze_21 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_20, 3), kwargs = {})
    %mul_8 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_5, %unsqueeze_21), kwargs = {})
    %add_7 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_8, %unsqueeze_18), kwargs = {})
    %squeeze_8 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_5, 3), kwargs = {})
    %squeeze_9 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_8, 2), kwargs = {})
    %squeeze_10 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_2, 3), kwargs = {})
    %squeeze_11 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_10, 2), kwargs = {})
    %detach_4 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_9,), kwargs = {})
    %detach_5 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_11,), kwargs = {})
    %permute : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_7, [0, 2, 3, 1]), kwargs = {})
    %view_6 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute, [2, 4096, 320]), kwargs = {})
    %_param_constant18 : [#users=1] = get_attr[target=_param_constant18]
    %t_3 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant18,), kwargs = {})
    %expand_1 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_6, [2, 4096, 320]), kwargs = {})
    %view_7 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_1, [2, 4096, 320]), kwargs = {})
    %expand_2 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_3, [2, 320, 320]), kwargs = {})
    %view_8 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_2, [2, 320, 320]), kwargs = {})
    %bmm : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_7, %view_8), kwargs = {})
    %_unsafe_view : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm, [2, 4096, 320]), kwargs = {})
    %_param_constant19 : [#users=1] = get_attr[target=_param_constant19]
    %add_8 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view, %_param_constant19), kwargs = {})
    %var_mean_3 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_8, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_6 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_3, 0), kwargs = {})
    %getitem_7 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_3, 1), kwargs = {})
    %add_9 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_6, 1e-05), kwargs = {})
    %rsqrt_3 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_9,), kwargs = {})
    %sub_3 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_8, %getitem_7), kwargs = {})
    %mul_9 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_3, %rsqrt_3), kwargs = {})
    %_param_constant20 : [#users=1] = get_attr[target=_param_constant20]
    %mul_10 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_9, %_param_constant20), kwargs = {})
    %_param_constant21 : [#users=1] = get_attr[target=_param_constant21]
    %add_10 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_10, %_param_constant21), kwargs = {})
    %_param_constant22 : [#users=1] = get_attr[target=_param_constant22]
    %t_4 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant22,), kwargs = {})
    %view_9 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_10, [8192, 320]), kwargs = {})
    %mm : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_9, %t_4), kwargs = {})
    %_unsafe_view_1 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm, [2, 4096, 320]), kwargs = {})
    %view_10 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_1, [2, 4096, 5, 64]), kwargs = {})
    %permute_1 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_10, [0, 2, 1, 3]), kwargs = {})
    %clone : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_1,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_2 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone, [10, 4096, 64]), kwargs = {})
    %_param_constant23 : [#users=1] = get_attr[target=_param_constant23]
    %t_5 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant23,), kwargs = {})
    %view_11 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_10, [8192, 320]), kwargs = {})
    %mm_1 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_11, %t_5), kwargs = {})
    %_unsafe_view_3 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_1, [2, 4096, 320]), kwargs = {})
    %_param_constant24 : [#users=1] = get_attr[target=_param_constant24]
    %t_6 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant24,), kwargs = {})
    %view_12 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_10, [8192, 320]), kwargs = {})
    %mm_2 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_12, %t_6), kwargs = {})
    %_unsafe_view_4 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_2, [2, 4096, 320]), kwargs = {})
    %view_13 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_3, [2, 4096, 5, 64]), kwargs = {})
    %permute_2 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_13, [0, 2, 1, 3]), kwargs = {})
    %clone_1 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_2,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_5 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_1, [10, 4096, 64]), kwargs = {})
    %view_14 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_4, [2, 4096, 5, 64]), kwargs = {})
    %permute_3 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_14, [0, 2, 1, 3]), kwargs = {})
    %clone_2 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_3,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_6 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_2, [10, 4096, 64]), kwargs = {})
    %empty : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([10, 4096, 4096],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_5, -1, -2), kwargs = {})
    %baddbmm : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty, %_unsafe_view_2, %transpose), kwargs = {beta: 0, alpha: 0.125})
    %_softmax : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm, -1, False), kwargs = {})
    %detach_6 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax,), kwargs = {})
    %bmm_1 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax, %_unsafe_view_6), kwargs = {})
    %view_15 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_1, [2, 5, 4096, 64]), kwargs = {})
    %permute_4 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_15, [0, 2, 1, 3]), kwargs = {})
    %clone_3 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_4,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_7 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_3, [2, 4096, 320]), kwargs = {})
    %_param_constant25 : [#users=1] = get_attr[target=_param_constant25]
    %t_7 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant25,), kwargs = {})
    %view_16 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_7, [8192, 320]), kwargs = {})
    %_param_constant26 : [#users=1] = get_attr[target=_param_constant26]
    %addmm_3 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant26, %view_16, %t_7), kwargs = {})
    %view_17 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_3, [2, 4096, 320]), kwargs = {})
    %add_11 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_17, %add_8), kwargs = {})
    %var_mean_4 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_11, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_8 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_4, 0), kwargs = {})
    %getitem_9 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_4, 1), kwargs = {})
    %add_12 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_8, 1e-05), kwargs = {})
    %rsqrt_4 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_12,), kwargs = {})
    %sub_4 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_11, %getitem_9), kwargs = {})
    %mul_11 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_4, %rsqrt_4), kwargs = {})
    %_param_constant27 : [#users=1] = get_attr[target=_param_constant27]
    %mul_12 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_11, %_param_constant27), kwargs = {})
    %_param_constant28 : [#users=1] = get_attr[target=_param_constant28]
    %add_13 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_12, %_param_constant28), kwargs = {})
    %_param_constant29 : [#users=1] = get_attr[target=_param_constant29]
    %t_8 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant29,), kwargs = {})
    %view_18 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_13, [8192, 320]), kwargs = {})
    %mm_3 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_18, %t_8), kwargs = {})
    %_unsafe_view_8 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_3, [2, 4096, 320]), kwargs = {})
    %view_19 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_8, [2, 4096, 5, 64]), kwargs = {})
    %permute_5 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_19, [0, 2, 1, 3]), kwargs = {})
    %clone_4 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_5,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_9 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_4, [10, 4096, 64]), kwargs = {})
    %_param_constant30 : [#users=1] = get_attr[target=_param_constant30]
    %t_9 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant30,), kwargs = {})
    %view_20 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_4 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_20, %t_9), kwargs = {})
    %_unsafe_view_10 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_4, [2, 64, 320]), kwargs = {})
    %_param_constant31 : [#users=1] = get_attr[target=_param_constant31]
    %t_10 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant31,), kwargs = {})
    %view_21 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_5 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_21, %t_10), kwargs = {})
    %_unsafe_view_11 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_5, [2, 64, 320]), kwargs = {})
    %view_22 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_10, [2, 64, 5, 64]), kwargs = {})
    %permute_6 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_22, [0, 2, 1, 3]), kwargs = {})
    %clone_5 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_6,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_12 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_5, [10, 64, 64]), kwargs = {})
    %view_23 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_11, [2, 64, 5, 64]), kwargs = {})
    %permute_7 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_23, [0, 2, 1, 3]), kwargs = {})
    %clone_6 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_7,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_13 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_6, [10, 64, 64]), kwargs = {})
    %empty_1 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([10, 4096, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_1 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_12, -1, -2), kwargs = {})
    %baddbmm_1 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_1, %_unsafe_view_9, %transpose_1), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_1 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_1, -1, False), kwargs = {})
    %detach_7 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_1,), kwargs = {})
    %bmm_2 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_1, %_unsafe_view_13), kwargs = {})
    %view_24 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_2, [2, 5, 4096, 64]), kwargs = {})
    %permute_8 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_24, [0, 2, 1, 3]), kwargs = {})
    %clone_7 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_8,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_14 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_7, [2, 4096, 320]), kwargs = {})
    %_param_constant32 : [#users=1] = get_attr[target=_param_constant32]
    %t_11 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant32,), kwargs = {})
    %view_25 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_14, [8192, 320]), kwargs = {})
    %_param_constant33 : [#users=1] = get_attr[target=_param_constant33]
    %addmm_4 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant33, %view_25, %t_11), kwargs = {})
    %view_26 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_4, [2, 4096, 320]), kwargs = {})
    %add_14 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_26, %add_11), kwargs = {})
    %var_mean_5 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_14, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_10 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_5, 0), kwargs = {})
    %getitem_11 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_5, 1), kwargs = {})
    %add_15 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_10, 1e-05), kwargs = {})
    %rsqrt_5 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_15,), kwargs = {})
    %sub_5 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_14, %getitem_11), kwargs = {})
    %mul_13 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_5, %rsqrt_5), kwargs = {})
    %_param_constant34 : [#users=1] = get_attr[target=_param_constant34]
    %mul_14 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_13, %_param_constant34), kwargs = {})
    %_param_constant35 : [#users=1] = get_attr[target=_param_constant35]
    %add_16 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_14, %_param_constant35), kwargs = {})
    %_param_constant36 : [#users=1] = get_attr[target=_param_constant36]
    %t_12 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant36,), kwargs = {})
    %view_27 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_16, [8192, 320]), kwargs = {})
    %_param_constant37 : [#users=1] = get_attr[target=_param_constant37]
    %addmm_5 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant37, %view_27, %t_12), kwargs = {})
    %view_28 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_5, [2, 4096, 2560]), kwargs = {})
    %slice_9 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_28, -1, 0, 1280), kwargs = {})
    %slice_10 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_28, -1, 1280, 2560), kwargs = {})
    %gelu : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_10,), kwargs = {})
    %mul_15 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_9, %gelu), kwargs = {})
    %_param_constant38 : [#users=1] = get_attr[target=_param_constant38]
    %t_13 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant38,), kwargs = {})
    %view_29 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_15, [8192, 1280]), kwargs = {})
    %_param_constant39 : [#users=1] = get_attr[target=_param_constant39]
    %addmm_6 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant39, %view_29, %t_13), kwargs = {})
    %view_30 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_6, [2, 4096, 320]), kwargs = {})
    %add_17 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_30, %add_14), kwargs = {})
    %_param_constant40 : [#users=1] = get_attr[target=_param_constant40]
    %t_14 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant40,), kwargs = {})
    %view_31 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_17, [8192, 320]), kwargs = {})
    %_param_constant41 : [#users=1] = get_attr[target=_param_constant41]
    %addmm_7 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant41, %view_31, %t_14), kwargs = {})
    %view_32 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_7, [2, 4096, 320]), kwargs = {})
    %view_33 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_32, [2, 64, 64, 320]), kwargs = {})
    %permute_9 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_33, [0, 3, 1, 2]), kwargs = {})
    %clone_8 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_9,), kwargs = {memory_format: torch.contiguous_format})
    %add_18 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%clone_8, %div_1), kwargs = {})
    %view_34 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_18, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_6 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_34, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_12 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_6, 0), kwargs = {})
    %getitem_13 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_6, 1), kwargs = {})
    %add_19 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_12, 1e-05), kwargs = {})
    %rsqrt_6 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_19,), kwargs = {})
    %sub_6 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_34, %getitem_13), kwargs = {})
    %mul_16 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_6, %rsqrt_6), kwargs = {})
    %view_35 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_16, [2, 320, 64, 64]), kwargs = {})
    %_param_constant42 : [#users=1] = get_attr[target=_param_constant42]
    %unsqueeze_22 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant42, 0), kwargs = {})
    %unsqueeze_23 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_22, 2), kwargs = {})
    %unsqueeze_24 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_23, 3), kwargs = {})
    %_param_constant43 : [#users=1] = get_attr[target=_param_constant43]
    %unsqueeze_25 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant43, 0), kwargs = {})
    %unsqueeze_26 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_25, 2), kwargs = {})
    %unsqueeze_27 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_26, 3), kwargs = {})
    %mul_17 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_35, %unsqueeze_27), kwargs = {})
    %add_20 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_17, %unsqueeze_24), kwargs = {})
    %squeeze_12 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_13, 3), kwargs = {})
    %squeeze_13 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_12, 2), kwargs = {})
    %squeeze_14 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_6, 3), kwargs = {})
    %squeeze_15 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_14, 2), kwargs = {})
    %detach_8 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_13,), kwargs = {})
    %detach_9 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_15,), kwargs = {})
    %silu_4 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_20,), kwargs = {})
    %_param_constant44 : [#users=1] = get_attr[target=_param_constant44]
    %_param_constant45 : [#users=1] = get_attr[target=_param_constant45]
    %convolution_3 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_4, %_param_constant44, %_param_constant45, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_5 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant46 : [#users=1] = get_attr[target=_param_constant46]
    %t_15 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant46,), kwargs = {})
    %_param_constant47 : [#users=1] = get_attr[target=_param_constant47]
    %addmm_8 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant47, %silu_5, %t_15), kwargs = {})
    %slice_11 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_8, 0, 0, 9223372036854775807), kwargs = {})
    %slice_12 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_11, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_28 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_12, 2), kwargs = {})
    %unsqueeze_29 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_28, 3), kwargs = {})
    %add_21 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_3, %unsqueeze_29), kwargs = {})
    %view_36 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_21, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_7 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_36, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_14 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_7, 0), kwargs = {})
    %getitem_15 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_7, 1), kwargs = {})
    %add_22 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_14, 1e-05), kwargs = {})
    %rsqrt_7 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_22,), kwargs = {})
    %sub_7 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_36, %getitem_15), kwargs = {})
    %mul_18 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_7, %rsqrt_7), kwargs = {})
    %view_37 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_18, [2, 320, 64, 64]), kwargs = {})
    %_param_constant48 : [#users=1] = get_attr[target=_param_constant48]
    %unsqueeze_30 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant48, 0), kwargs = {})
    %unsqueeze_31 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_30, 2), kwargs = {})
    %unsqueeze_32 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_31, 3), kwargs = {})
    %_param_constant49 : [#users=1] = get_attr[target=_param_constant49]
    %unsqueeze_33 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant49, 0), kwargs = {})
    %unsqueeze_34 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_33, 2), kwargs = {})
    %unsqueeze_35 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_34, 3), kwargs = {})
    %mul_19 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_37, %unsqueeze_35), kwargs = {})
    %add_23 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_19, %unsqueeze_32), kwargs = {})
    %squeeze_16 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_15, 3), kwargs = {})
    %squeeze_17 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_16, 2), kwargs = {})
    %squeeze_18 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_7, 3), kwargs = {})
    %squeeze_19 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_18, 2), kwargs = {})
    %detach_10 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_17,), kwargs = {})
    %detach_11 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_19,), kwargs = {})
    %silu_6 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_23,), kwargs = {})
    %_param_constant50 : [#users=1] = get_attr[target=_param_constant50]
    %_param_constant51 : [#users=1] = get_attr[target=_param_constant51]
    %convolution_4 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_6, %_param_constant50, %_param_constant51, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_24 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%add_18, %convolution_4), kwargs = {})
    %div_2 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_24, 1.0), kwargs = {})
    %view_38 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_2, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_8 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_38, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_16 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_8, 0), kwargs = {})
    %getitem_17 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_8, 1), kwargs = {})
    %add_25 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_16, 1e-06), kwargs = {})
    %rsqrt_8 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_25,), kwargs = {})
    %sub_8 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_38, %getitem_17), kwargs = {})
    %mul_20 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_8, %rsqrt_8), kwargs = {})
    %view_39 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_20, [2, 320, 64, 64]), kwargs = {})
    %_param_constant52 : [#users=1] = get_attr[target=_param_constant52]
    %unsqueeze_36 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant52, 0), kwargs = {})
    %unsqueeze_37 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_36, 2), kwargs = {})
    %unsqueeze_38 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_37, 3), kwargs = {})
    %_param_constant53 : [#users=1] = get_attr[target=_param_constant53]
    %unsqueeze_39 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant53, 0), kwargs = {})
    %unsqueeze_40 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_39, 2), kwargs = {})
    %unsqueeze_41 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_40, 3), kwargs = {})
    %mul_21 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_39, %unsqueeze_41), kwargs = {})
    %add_26 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_21, %unsqueeze_38), kwargs = {})
    %squeeze_20 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_17, 3), kwargs = {})
    %squeeze_21 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_20, 2), kwargs = {})
    %squeeze_22 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_8, 3), kwargs = {})
    %squeeze_23 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_22, 2), kwargs = {})
    %detach_12 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_21,), kwargs = {})
    %detach_13 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_23,), kwargs = {})
    %permute_10 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_26, [0, 2, 3, 1]), kwargs = {})
    %view_40 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_10, [2, 4096, 320]), kwargs = {})
    %_param_constant54 : [#users=1] = get_attr[target=_param_constant54]
    %t_16 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant54,), kwargs = {})
    %expand_3 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_40, [2, 4096, 320]), kwargs = {})
    %view_41 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_3, [2, 4096, 320]), kwargs = {})
    %expand_4 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_16, [2, 320, 320]), kwargs = {})
    %view_42 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_4, [2, 320, 320]), kwargs = {})
    %bmm_3 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_41, %view_42), kwargs = {})
    %_unsafe_view_15 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_3, [2, 4096, 320]), kwargs = {})
    %_param_constant55 : [#users=1] = get_attr[target=_param_constant55]
    %add_27 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_15, %_param_constant55), kwargs = {})
    %var_mean_9 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_27, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_18 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_9, 0), kwargs = {})
    %getitem_19 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_9, 1), kwargs = {})
    %add_28 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_18, 1e-05), kwargs = {})
    %rsqrt_9 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_28,), kwargs = {})
    %sub_9 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_27, %getitem_19), kwargs = {})
    %mul_22 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_9, %rsqrt_9), kwargs = {})
    %_param_constant56 : [#users=1] = get_attr[target=_param_constant56]
    %mul_23 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_22, %_param_constant56), kwargs = {})
    %_param_constant57 : [#users=1] = get_attr[target=_param_constant57]
    %add_29 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_23, %_param_constant57), kwargs = {})
    %_param_constant58 : [#users=1] = get_attr[target=_param_constant58]
    %t_17 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant58,), kwargs = {})
    %view_43 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_29, [8192, 320]), kwargs = {})
    %mm_6 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_43, %t_17), kwargs = {})
    %_unsafe_view_16 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_6, [2, 4096, 320]), kwargs = {})
    %view_44 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_16, [2, 4096, 5, 64]), kwargs = {})
    %permute_11 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_44, [0, 2, 1, 3]), kwargs = {})
    %clone_9 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_11,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_17 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_9, [10, 4096, 64]), kwargs = {})
    %_param_constant59 : [#users=1] = get_attr[target=_param_constant59]
    %t_18 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant59,), kwargs = {})
    %view_45 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_29, [8192, 320]), kwargs = {})
    %mm_7 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_45, %t_18), kwargs = {})
    %_unsafe_view_18 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_7, [2, 4096, 320]), kwargs = {})
    %_param_constant60 : [#users=1] = get_attr[target=_param_constant60]
    %t_19 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant60,), kwargs = {})
    %view_46 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_29, [8192, 320]), kwargs = {})
    %mm_8 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_46, %t_19), kwargs = {})
    %_unsafe_view_19 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_8, [2, 4096, 320]), kwargs = {})
    %view_47 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_18, [2, 4096, 5, 64]), kwargs = {})
    %permute_12 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_47, [0, 2, 1, 3]), kwargs = {})
    %clone_10 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_12,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_20 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_10, [10, 4096, 64]), kwargs = {})
    %view_48 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_19, [2, 4096, 5, 64]), kwargs = {})
    %permute_13 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_48, [0, 2, 1, 3]), kwargs = {})
    %clone_11 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_13,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_21 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_11, [10, 4096, 64]), kwargs = {})
    %empty_2 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([10, 4096, 4096],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_2 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_20, -1, -2), kwargs = {})
    %baddbmm_2 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_2, %_unsafe_view_17, %transpose_2), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_2 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_2, -1, False), kwargs = {})
    %detach_14 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_2,), kwargs = {})
    %bmm_4 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_2, %_unsafe_view_21), kwargs = {})
    %view_49 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_4, [2, 5, 4096, 64]), kwargs = {})
    %permute_14 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_49, [0, 2, 1, 3]), kwargs = {})
    %clone_12 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_14,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_22 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_12, [2, 4096, 320]), kwargs = {})
    %_param_constant61 : [#users=1] = get_attr[target=_param_constant61]
    %t_20 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant61,), kwargs = {})
    %view_50 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_22, [8192, 320]), kwargs = {})
    %_param_constant62 : [#users=1] = get_attr[target=_param_constant62]
    %addmm_9 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant62, %view_50, %t_20), kwargs = {})
    %view_51 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_9, [2, 4096, 320]), kwargs = {})
    %add_30 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_51, %add_27), kwargs = {})
    %var_mean_10 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_30, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_20 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_10, 0), kwargs = {})
    %getitem_21 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_10, 1), kwargs = {})
    %add_31 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_20, 1e-05), kwargs = {})
    %rsqrt_10 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_31,), kwargs = {})
    %sub_10 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_30, %getitem_21), kwargs = {})
    %mul_24 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_10, %rsqrt_10), kwargs = {})
    %_param_constant63 : [#users=1] = get_attr[target=_param_constant63]
    %mul_25 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_24, %_param_constant63), kwargs = {})
    %_param_constant64 : [#users=1] = get_attr[target=_param_constant64]
    %add_32 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_25, %_param_constant64), kwargs = {})
    %_param_constant65 : [#users=1] = get_attr[target=_param_constant65]
    %t_21 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant65,), kwargs = {})
    %view_52 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_32, [8192, 320]), kwargs = {})
    %mm_9 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_52, %t_21), kwargs = {})
    %_unsafe_view_23 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_9, [2, 4096, 320]), kwargs = {})
    %view_53 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_23, [2, 4096, 5, 64]), kwargs = {})
    %permute_15 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_53, [0, 2, 1, 3]), kwargs = {})
    %clone_13 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_15,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_24 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_13, [10, 4096, 64]), kwargs = {})
    %_param_constant66 : [#users=1] = get_attr[target=_param_constant66]
    %t_22 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant66,), kwargs = {})
    %view_54 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_10 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_54, %t_22), kwargs = {})
    %_unsafe_view_25 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_10, [2, 64, 320]), kwargs = {})
    %_param_constant67 : [#users=1] = get_attr[target=_param_constant67]
    %t_23 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant67,), kwargs = {})
    %view_55 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_11 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_55, %t_23), kwargs = {})
    %_unsafe_view_26 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_11, [2, 64, 320]), kwargs = {})
    %view_56 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_25, [2, 64, 5, 64]), kwargs = {})
    %permute_16 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_56, [0, 2, 1, 3]), kwargs = {})
    %clone_14 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_16,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_27 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_14, [10, 64, 64]), kwargs = {})
    %view_57 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_26, [2, 64, 5, 64]), kwargs = {})
    %permute_17 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_57, [0, 2, 1, 3]), kwargs = {})
    %clone_15 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_17,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_28 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_15, [10, 64, 64]), kwargs = {})
    %empty_3 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([10, 4096, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_3 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_27, -1, -2), kwargs = {})
    %baddbmm_3 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_3, %_unsafe_view_24, %transpose_3), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_3 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_3, -1, False), kwargs = {})
    %detach_15 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_3,), kwargs = {})
    %bmm_5 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_3, %_unsafe_view_28), kwargs = {})
    %view_58 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_5, [2, 5, 4096, 64]), kwargs = {})
    %permute_18 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_58, [0, 2, 1, 3]), kwargs = {})
    %clone_16 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_18,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_29 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_16, [2, 4096, 320]), kwargs = {})
    %_param_constant68 : [#users=1] = get_attr[target=_param_constant68]
    %t_24 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant68,), kwargs = {})
    %view_59 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_29, [8192, 320]), kwargs = {})
    %_param_constant69 : [#users=1] = get_attr[target=_param_constant69]
    %addmm_10 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant69, %view_59, %t_24), kwargs = {})
    %view_60 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_10, [2, 4096, 320]), kwargs = {})
    %add_33 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_60, %add_30), kwargs = {})
    %var_mean_11 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_33, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_22 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_11, 0), kwargs = {})
    %getitem_23 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_11, 1), kwargs = {})
    %add_34 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_22, 1e-05), kwargs = {})
    %rsqrt_11 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_34,), kwargs = {})
    %sub_11 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_33, %getitem_23), kwargs = {})
    %mul_26 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_11, %rsqrt_11), kwargs = {})
    %_param_constant70 : [#users=1] = get_attr[target=_param_constant70]
    %mul_27 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_26, %_param_constant70), kwargs = {})
    %_param_constant71 : [#users=1] = get_attr[target=_param_constant71]
    %add_35 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_27, %_param_constant71), kwargs = {})
    %_param_constant72 : [#users=1] = get_attr[target=_param_constant72]
    %t_25 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant72,), kwargs = {})
    %view_61 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_35, [8192, 320]), kwargs = {})
    %_param_constant73 : [#users=1] = get_attr[target=_param_constant73]
    %addmm_11 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant73, %view_61, %t_25), kwargs = {})
    %view_62 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_11, [2, 4096, 2560]), kwargs = {})
    %slice_13 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_62, -1, 0, 1280), kwargs = {})
    %slice_14 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_62, -1, 1280, 2560), kwargs = {})
    %gelu_1 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_14,), kwargs = {})
    %mul_28 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_13, %gelu_1), kwargs = {})
    %_param_constant74 : [#users=1] = get_attr[target=_param_constant74]
    %t_26 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant74,), kwargs = {})
    %view_63 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_28, [8192, 1280]), kwargs = {})
    %_param_constant75 : [#users=1] = get_attr[target=_param_constant75]
    %addmm_12 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant75, %view_63, %t_26), kwargs = {})
    %view_64 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_12, [2, 4096, 320]), kwargs = {})
    %add_36 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_64, %add_33), kwargs = {})
    %_param_constant76 : [#users=1] = get_attr[target=_param_constant76]
    %t_27 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant76,), kwargs = {})
    %view_65 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_36, [8192, 320]), kwargs = {})
    %_param_constant77 : [#users=1] = get_attr[target=_param_constant77]
    %addmm_13 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant77, %view_65, %t_27), kwargs = {})
    %view_66 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_13, [2, 4096, 320]), kwargs = {})
    %view_67 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_66, [2, 64, 64, 320]), kwargs = {})
    %permute_19 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_67, [0, 3, 1, 2]), kwargs = {})
    %clone_17 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_19,), kwargs = {memory_format: torch.contiguous_format})
    %add_37 : [#users=2] = call_function[target=torch.ops.aten.add](args = (%clone_17, %div_2), kwargs = {})
    %_param_constant78 : [#users=1] = get_attr[target=_param_constant78]
    %_param_constant79 : [#users=1] = get_attr[target=_param_constant79]
    %convolution_5 : [#users=3] = call_function[target=torch.ops.aten.convolution](args = (%add_37, %_param_constant78, %_param_constant79, [2, 2], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %view_68 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%convolution_5, [2, 32, 10, 1024]), kwargs = {})
    %var_mean_12 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_68, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_24 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_12, 0), kwargs = {})
    %getitem_25 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_12, 1), kwargs = {})
    %add_38 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_24, 1e-05), kwargs = {})
    %rsqrt_12 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_38,), kwargs = {})
    %sub_12 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_68, %getitem_25), kwargs = {})
    %mul_29 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_12, %rsqrt_12), kwargs = {})
    %view_69 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_29, [2, 320, 32, 32]), kwargs = {})
    %_param_constant80 : [#users=1] = get_attr[target=_param_constant80]
    %unsqueeze_42 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant80, 0), kwargs = {})
    %unsqueeze_43 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_42, 2), kwargs = {})
    %unsqueeze_44 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_43, 3), kwargs = {})
    %_param_constant81 : [#users=1] = get_attr[target=_param_constant81]
    %unsqueeze_45 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant81, 0), kwargs = {})
    %unsqueeze_46 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_45, 2), kwargs = {})
    %unsqueeze_47 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_46, 3), kwargs = {})
    %mul_30 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_69, %unsqueeze_47), kwargs = {})
    %add_39 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_30, %unsqueeze_44), kwargs = {})
    %squeeze_24 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_25, 3), kwargs = {})
    %squeeze_25 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_24, 2), kwargs = {})
    %squeeze_26 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_12, 3), kwargs = {})
    %squeeze_27 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_26, 2), kwargs = {})
    %detach_16 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_25,), kwargs = {})
    %detach_17 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_27,), kwargs = {})
    %silu_7 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_39,), kwargs = {})
    %_param_constant82 : [#users=1] = get_attr[target=_param_constant82]
    %_param_constant83 : [#users=1] = get_attr[target=_param_constant83]
    %convolution_6 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_7, %_param_constant82, %_param_constant83, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_8 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant84 : [#users=1] = get_attr[target=_param_constant84]
    %t_28 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant84,), kwargs = {})
    %_param_constant85 : [#users=1] = get_attr[target=_param_constant85]
    %addmm_14 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant85, %silu_8, %t_28), kwargs = {})
    %slice_15 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_14, 0, 0, 9223372036854775807), kwargs = {})
    %slice_16 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_15, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_48 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_16, 2), kwargs = {})
    %unsqueeze_49 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_48, 3), kwargs = {})
    %add_40 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_6, %unsqueeze_49), kwargs = {})
    %view_70 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_40, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_13 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_70, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_26 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_13, 0), kwargs = {})
    %getitem_27 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_13, 1), kwargs = {})
    %add_41 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_26, 1e-05), kwargs = {})
    %rsqrt_13 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_41,), kwargs = {})
    %sub_13 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_70, %getitem_27), kwargs = {})
    %mul_31 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_13, %rsqrt_13), kwargs = {})
    %view_71 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_31, [2, 640, 32, 32]), kwargs = {})
    %_param_constant86 : [#users=1] = get_attr[target=_param_constant86]
    %unsqueeze_50 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant86, 0), kwargs = {})
    %unsqueeze_51 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_50, 2), kwargs = {})
    %unsqueeze_52 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_51, 3), kwargs = {})
    %_param_constant87 : [#users=1] = get_attr[target=_param_constant87]
    %unsqueeze_53 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant87, 0), kwargs = {})
    %unsqueeze_54 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_53, 2), kwargs = {})
    %unsqueeze_55 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_54, 3), kwargs = {})
    %mul_32 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_71, %unsqueeze_55), kwargs = {})
    %add_42 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_32, %unsqueeze_52), kwargs = {})
    %squeeze_28 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_27, 3), kwargs = {})
    %squeeze_29 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_28, 2), kwargs = {})
    %squeeze_30 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_13, 3), kwargs = {})
    %squeeze_31 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_30, 2), kwargs = {})
    %detach_18 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_29,), kwargs = {})
    %detach_19 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_31,), kwargs = {})
    %silu_9 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_42,), kwargs = {})
    %_param_constant88 : [#users=1] = get_attr[target=_param_constant88]
    %_param_constant89 : [#users=1] = get_attr[target=_param_constant89]
    %convolution_7 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_9, %_param_constant88, %_param_constant89, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant90 : [#users=1] = get_attr[target=_param_constant90]
    %_param_constant91 : [#users=1] = get_attr[target=_param_constant91]
    %convolution_8 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%convolution_5, %_param_constant90, %_param_constant91, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_43 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_8, %convolution_7), kwargs = {})
    %div_3 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_43, 1.0), kwargs = {})
    %view_72 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_3, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_14 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_72, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_28 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_14, 0), kwargs = {})
    %getitem_29 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_14, 1), kwargs = {})
    %add_44 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_28, 1e-06), kwargs = {})
    %rsqrt_14 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_44,), kwargs = {})
    %sub_14 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_72, %getitem_29), kwargs = {})
    %mul_33 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_14, %rsqrt_14), kwargs = {})
    %view_73 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_33, [2, 640, 32, 32]), kwargs = {})
    %_param_constant92 : [#users=1] = get_attr[target=_param_constant92]
    %unsqueeze_56 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant92, 0), kwargs = {})
    %unsqueeze_57 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_56, 2), kwargs = {})
    %unsqueeze_58 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_57, 3), kwargs = {})
    %_param_constant93 : [#users=1] = get_attr[target=_param_constant93]
    %unsqueeze_59 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant93, 0), kwargs = {})
    %unsqueeze_60 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_59, 2), kwargs = {})
    %unsqueeze_61 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_60, 3), kwargs = {})
    %mul_34 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_73, %unsqueeze_61), kwargs = {})
    %add_45 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_34, %unsqueeze_58), kwargs = {})
    %squeeze_32 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_29, 3), kwargs = {})
    %squeeze_33 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_32, 2), kwargs = {})
    %squeeze_34 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_14, 3), kwargs = {})
    %squeeze_35 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_34, 2), kwargs = {})
    %detach_20 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_33,), kwargs = {})
    %detach_21 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_35,), kwargs = {})
    %permute_20 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_45, [0, 2, 3, 1]), kwargs = {})
    %view_74 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_20, [2, 1024, 640]), kwargs = {})
    %_param_constant94 : [#users=1] = get_attr[target=_param_constant94]
    %t_29 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant94,), kwargs = {})
    %expand_5 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_74, [2, 1024, 640]), kwargs = {})
    %view_75 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_5, [2, 1024, 640]), kwargs = {})
    %expand_6 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_29, [2, 640, 640]), kwargs = {})
    %view_76 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_6, [2, 640, 640]), kwargs = {})
    %bmm_6 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_75, %view_76), kwargs = {})
    %_unsafe_view_30 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_6, [2, 1024, 640]), kwargs = {})
    %_param_constant95 : [#users=1] = get_attr[target=_param_constant95]
    %add_46 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_30, %_param_constant95), kwargs = {})
    %var_mean_15 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_46, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_30 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_15, 0), kwargs = {})
    %getitem_31 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_15, 1), kwargs = {})
    %add_47 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_30, 1e-05), kwargs = {})
    %rsqrt_15 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_47,), kwargs = {})
    %sub_15 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_46, %getitem_31), kwargs = {})
    %mul_35 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_15, %rsqrt_15), kwargs = {})
    %_param_constant96 : [#users=1] = get_attr[target=_param_constant96]
    %mul_36 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_35, %_param_constant96), kwargs = {})
    %_param_constant97 : [#users=1] = get_attr[target=_param_constant97]
    %add_48 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_36, %_param_constant97), kwargs = {})
    %_param_constant98 : [#users=1] = get_attr[target=_param_constant98]
    %t_30 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant98,), kwargs = {})
    %view_77 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_48, [2048, 640]), kwargs = {})
    %mm_12 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_77, %t_30), kwargs = {})
    %_unsafe_view_31 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_12, [2, 1024, 640]), kwargs = {})
    %view_78 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_31, [2, 1024, 10, 64]), kwargs = {})
    %permute_21 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_78, [0, 2, 1, 3]), kwargs = {})
    %clone_18 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_21,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_32 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_18, [20, 1024, 64]), kwargs = {})
    %_param_constant99 : [#users=1] = get_attr[target=_param_constant99]
    %t_31 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant99,), kwargs = {})
    %view_79 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_48, [2048, 640]), kwargs = {})
    %mm_13 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_79, %t_31), kwargs = {})
    %_unsafe_view_33 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_13, [2, 1024, 640]), kwargs = {})
    %_param_constant100 : [#users=1] = get_attr[target=_param_constant100]
    %t_32 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant100,), kwargs = {})
    %view_80 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_48, [2048, 640]), kwargs = {})
    %mm_14 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_80, %t_32), kwargs = {})
    %_unsafe_view_34 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_14, [2, 1024, 640]), kwargs = {})
    %view_81 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_33, [2, 1024, 10, 64]), kwargs = {})
    %permute_22 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_81, [0, 2, 1, 3]), kwargs = {})
    %clone_19 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_22,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_35 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_19, [20, 1024, 64]), kwargs = {})
    %view_82 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_34, [2, 1024, 10, 64]), kwargs = {})
    %permute_23 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_82, [0, 2, 1, 3]), kwargs = {})
    %clone_20 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_23,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_36 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_20, [20, 1024, 64]), kwargs = {})
    %empty_4 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([20, 1024, 1024],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_4 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_35, -1, -2), kwargs = {})
    %baddbmm_4 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_4, %_unsafe_view_32, %transpose_4), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_4 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_4, -1, False), kwargs = {})
    %detach_22 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_4,), kwargs = {})
    %bmm_7 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_4, %_unsafe_view_36), kwargs = {})
    %view_83 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_7, [2, 10, 1024, 64]), kwargs = {})
    %permute_24 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_83, [0, 2, 1, 3]), kwargs = {})
    %clone_21 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_24,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_37 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_21, [2, 1024, 640]), kwargs = {})
    %_param_constant101 : [#users=1] = get_attr[target=_param_constant101]
    %t_33 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant101,), kwargs = {})
    %view_84 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_37, [2048, 640]), kwargs = {})
    %_param_constant102 : [#users=1] = get_attr[target=_param_constant102]
    %addmm_15 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant102, %view_84, %t_33), kwargs = {})
    %view_85 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_15, [2, 1024, 640]), kwargs = {})
    %add_49 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_85, %add_46), kwargs = {})
    %var_mean_16 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_49, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_32 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_16, 0), kwargs = {})
    %getitem_33 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_16, 1), kwargs = {})
    %add_50 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_32, 1e-05), kwargs = {})
    %rsqrt_16 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_50,), kwargs = {})
    %sub_16 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_49, %getitem_33), kwargs = {})
    %mul_37 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_16, %rsqrt_16), kwargs = {})
    %_param_constant103 : [#users=1] = get_attr[target=_param_constant103]
    %mul_38 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_37, %_param_constant103), kwargs = {})
    %_param_constant104 : [#users=1] = get_attr[target=_param_constant104]
    %add_51 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_38, %_param_constant104), kwargs = {})
    %_param_constant105 : [#users=1] = get_attr[target=_param_constant105]
    %t_34 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant105,), kwargs = {})
    %view_86 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_51, [2048, 640]), kwargs = {})
    %mm_15 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_86, %t_34), kwargs = {})
    %_unsafe_view_38 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_15, [2, 1024, 640]), kwargs = {})
    %view_87 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_38, [2, 1024, 10, 64]), kwargs = {})
    %permute_25 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_87, [0, 2, 1, 3]), kwargs = {})
    %clone_22 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_25,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_39 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_22, [20, 1024, 64]), kwargs = {})
    %_param_constant106 : [#users=1] = get_attr[target=_param_constant106]
    %t_35 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant106,), kwargs = {})
    %view_88 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_16 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_88, %t_35), kwargs = {})
    %_unsafe_view_40 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_16, [2, 64, 640]), kwargs = {})
    %_param_constant107 : [#users=1] = get_attr[target=_param_constant107]
    %t_36 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant107,), kwargs = {})
    %view_89 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_17 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_89, %t_36), kwargs = {})
    %_unsafe_view_41 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_17, [2, 64, 640]), kwargs = {})
    %view_90 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_40, [2, 64, 10, 64]), kwargs = {})
    %permute_26 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_90, [0, 2, 1, 3]), kwargs = {})
    %clone_23 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_26,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_42 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_23, [20, 64, 64]), kwargs = {})
    %view_91 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_41, [2, 64, 10, 64]), kwargs = {})
    %permute_27 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_91, [0, 2, 1, 3]), kwargs = {})
    %clone_24 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_27,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_43 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_24, [20, 64, 64]), kwargs = {})
    %empty_5 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([20, 1024, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_5 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_42, -1, -2), kwargs = {})
    %baddbmm_5 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_5, %_unsafe_view_39, %transpose_5), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_5 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_5, -1, False), kwargs = {})
    %detach_23 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_5,), kwargs = {})
    %bmm_8 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_5, %_unsafe_view_43), kwargs = {})
    %view_92 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_8, [2, 10, 1024, 64]), kwargs = {})
    %permute_28 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_92, [0, 2, 1, 3]), kwargs = {})
    %clone_25 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_28,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_44 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_25, [2, 1024, 640]), kwargs = {})
    %_param_constant108 : [#users=1] = get_attr[target=_param_constant108]
    %t_37 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant108,), kwargs = {})
    %view_93 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_44, [2048, 640]), kwargs = {})
    %_param_constant109 : [#users=1] = get_attr[target=_param_constant109]
    %addmm_16 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant109, %view_93, %t_37), kwargs = {})
    %view_94 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_16, [2, 1024, 640]), kwargs = {})
    %add_52 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_94, %add_49), kwargs = {})
    %var_mean_17 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_52, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_34 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_17, 0), kwargs = {})
    %getitem_35 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_17, 1), kwargs = {})
    %add_53 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_34, 1e-05), kwargs = {})
    %rsqrt_17 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_53,), kwargs = {})
    %sub_17 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_52, %getitem_35), kwargs = {})
    %mul_39 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_17, %rsqrt_17), kwargs = {})
    %_param_constant110 : [#users=1] = get_attr[target=_param_constant110]
    %mul_40 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_39, %_param_constant110), kwargs = {})
    %_param_constant111 : [#users=1] = get_attr[target=_param_constant111]
    %add_54 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_40, %_param_constant111), kwargs = {})
    %_param_constant112 : [#users=1] = get_attr[target=_param_constant112]
    %t_38 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant112,), kwargs = {})
    %view_95 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_54, [2048, 640]), kwargs = {})
    %_param_constant113 : [#users=1] = get_attr[target=_param_constant113]
    %addmm_17 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant113, %view_95, %t_38), kwargs = {})
    %view_96 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_17, [2, 1024, 5120]), kwargs = {})
    %slice_17 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_96, -1, 0, 2560), kwargs = {})
    %slice_18 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_96, -1, 2560, 5120), kwargs = {})
    %gelu_2 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_18,), kwargs = {})
    %mul_41 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_17, %gelu_2), kwargs = {})
    %_param_constant114 : [#users=1] = get_attr[target=_param_constant114]
    %t_39 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant114,), kwargs = {})
    %view_97 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_41, [2048, 2560]), kwargs = {})
    %_param_constant115 : [#users=1] = get_attr[target=_param_constant115]
    %addmm_18 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant115, %view_97, %t_39), kwargs = {})
    %view_98 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_18, [2, 1024, 640]), kwargs = {})
    %add_55 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_98, %add_52), kwargs = {})
    %_param_constant116 : [#users=1] = get_attr[target=_param_constant116]
    %t_40 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant116,), kwargs = {})
    %view_99 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_55, [2048, 640]), kwargs = {})
    %_param_constant117 : [#users=1] = get_attr[target=_param_constant117]
    %addmm_19 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant117, %view_99, %t_40), kwargs = {})
    %view_100 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_19, [2, 1024, 640]), kwargs = {})
    %view_101 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_100, [2, 32, 32, 640]), kwargs = {})
    %permute_29 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_101, [0, 3, 1, 2]), kwargs = {})
    %clone_26 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_29,), kwargs = {memory_format: torch.contiguous_format})
    %add_56 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%clone_26, %div_3), kwargs = {})
    %view_102 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_56, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_18 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_102, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_36 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_18, 0), kwargs = {})
    %getitem_37 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_18, 1), kwargs = {})
    %add_57 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_36, 1e-05), kwargs = {})
    %rsqrt_18 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_57,), kwargs = {})
    %sub_18 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_102, %getitem_37), kwargs = {})
    %mul_42 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_18, %rsqrt_18), kwargs = {})
    %view_103 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_42, [2, 640, 32, 32]), kwargs = {})
    %_param_constant118 : [#users=1] = get_attr[target=_param_constant118]
    %unsqueeze_62 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant118, 0), kwargs = {})
    %unsqueeze_63 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_62, 2), kwargs = {})
    %unsqueeze_64 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_63, 3), kwargs = {})
    %_param_constant119 : [#users=1] = get_attr[target=_param_constant119]
    %unsqueeze_65 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant119, 0), kwargs = {})
    %unsqueeze_66 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_65, 2), kwargs = {})
    %unsqueeze_67 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_66, 3), kwargs = {})
    %mul_43 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_103, %unsqueeze_67), kwargs = {})
    %add_58 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_43, %unsqueeze_64), kwargs = {})
    %squeeze_36 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_37, 3), kwargs = {})
    %squeeze_37 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_36, 2), kwargs = {})
    %squeeze_38 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_18, 3), kwargs = {})
    %squeeze_39 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_38, 2), kwargs = {})
    %detach_24 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_37,), kwargs = {})
    %detach_25 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_39,), kwargs = {})
    %silu_10 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_58,), kwargs = {})
    %_param_constant120 : [#users=1] = get_attr[target=_param_constant120]
    %_param_constant121 : [#users=1] = get_attr[target=_param_constant121]
    %convolution_9 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_10, %_param_constant120, %_param_constant121, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_11 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant122 : [#users=1] = get_attr[target=_param_constant122]
    %t_41 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant122,), kwargs = {})
    %_param_constant123 : [#users=1] = get_attr[target=_param_constant123]
    %addmm_20 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant123, %silu_11, %t_41), kwargs = {})
    %slice_19 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_20, 0, 0, 9223372036854775807), kwargs = {})
    %slice_20 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_19, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_68 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_20, 2), kwargs = {})
    %unsqueeze_69 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_68, 3), kwargs = {})
    %add_59 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_9, %unsqueeze_69), kwargs = {})
    %view_104 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_59, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_19 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_104, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_38 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_19, 0), kwargs = {})
    %getitem_39 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_19, 1), kwargs = {})
    %add_60 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_38, 1e-05), kwargs = {})
    %rsqrt_19 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_60,), kwargs = {})
    %sub_19 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_104, %getitem_39), kwargs = {})
    %mul_44 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_19, %rsqrt_19), kwargs = {})
    %view_105 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_44, [2, 640, 32, 32]), kwargs = {})
    %_param_constant124 : [#users=1] = get_attr[target=_param_constant124]
    %unsqueeze_70 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant124, 0), kwargs = {})
    %unsqueeze_71 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_70, 2), kwargs = {})
    %unsqueeze_72 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_71, 3), kwargs = {})
    %_param_constant125 : [#users=1] = get_attr[target=_param_constant125]
    %unsqueeze_73 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant125, 0), kwargs = {})
    %unsqueeze_74 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_73, 2), kwargs = {})
    %unsqueeze_75 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_74, 3), kwargs = {})
    %mul_45 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_105, %unsqueeze_75), kwargs = {})
    %add_61 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_45, %unsqueeze_72), kwargs = {})
    %squeeze_40 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_39, 3), kwargs = {})
    %squeeze_41 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_40, 2), kwargs = {})
    %squeeze_42 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_19, 3), kwargs = {})
    %squeeze_43 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_42, 2), kwargs = {})
    %detach_26 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_41,), kwargs = {})
    %detach_27 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_43,), kwargs = {})
    %silu_12 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_61,), kwargs = {})
    %_param_constant126 : [#users=1] = get_attr[target=_param_constant126]
    %_param_constant127 : [#users=1] = get_attr[target=_param_constant127]
    %convolution_10 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_12, %_param_constant126, %_param_constant127, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_62 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%add_56, %convolution_10), kwargs = {})
    %div_4 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_62, 1.0), kwargs = {})
    %view_106 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_4, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_20 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_106, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_40 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_20, 0), kwargs = {})
    %getitem_41 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_20, 1), kwargs = {})
    %add_63 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_40, 1e-06), kwargs = {})
    %rsqrt_20 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_63,), kwargs = {})
    %sub_20 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_106, %getitem_41), kwargs = {})
    %mul_46 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_20, %rsqrt_20), kwargs = {})
    %view_107 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_46, [2, 640, 32, 32]), kwargs = {})
    %_param_constant128 : [#users=1] = get_attr[target=_param_constant128]
    %unsqueeze_76 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant128, 0), kwargs = {})
    %unsqueeze_77 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_76, 2), kwargs = {})
    %unsqueeze_78 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_77, 3), kwargs = {})
    %_param_constant129 : [#users=1] = get_attr[target=_param_constant129]
    %unsqueeze_79 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant129, 0), kwargs = {})
    %unsqueeze_80 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_79, 2), kwargs = {})
    %unsqueeze_81 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_80, 3), kwargs = {})
    %mul_47 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_107, %unsqueeze_81), kwargs = {})
    %add_64 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_47, %unsqueeze_78), kwargs = {})
    %squeeze_44 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_41, 3), kwargs = {})
    %squeeze_45 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_44, 2), kwargs = {})
    %squeeze_46 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_20, 3), kwargs = {})
    %squeeze_47 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_46, 2), kwargs = {})
    %detach_28 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_45,), kwargs = {})
    %detach_29 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_47,), kwargs = {})
    %permute_30 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_64, [0, 2, 3, 1]), kwargs = {})
    %view_108 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_30, [2, 1024, 640]), kwargs = {})
    %_param_constant130 : [#users=1] = get_attr[target=_param_constant130]
    %t_42 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant130,), kwargs = {})
    %expand_7 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_108, [2, 1024, 640]), kwargs = {})
    %view_109 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_7, [2, 1024, 640]), kwargs = {})
    %expand_8 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_42, [2, 640, 640]), kwargs = {})
    %view_110 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_8, [2, 640, 640]), kwargs = {})
    %bmm_9 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_109, %view_110), kwargs = {})
    %_unsafe_view_45 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_9, [2, 1024, 640]), kwargs = {})
    %_param_constant131 : [#users=1] = get_attr[target=_param_constant131]
    %add_65 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_45, %_param_constant131), kwargs = {})
    %var_mean_21 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_65, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_42 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_21, 0), kwargs = {})
    %getitem_43 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_21, 1), kwargs = {})
    %add_66 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_42, 1e-05), kwargs = {})
    %rsqrt_21 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_66,), kwargs = {})
    %sub_21 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_65, %getitem_43), kwargs = {})
    %mul_48 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_21, %rsqrt_21), kwargs = {})
    %_param_constant132 : [#users=1] = get_attr[target=_param_constant132]
    %mul_49 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_48, %_param_constant132), kwargs = {})
    %_param_constant133 : [#users=1] = get_attr[target=_param_constant133]
    %add_67 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_49, %_param_constant133), kwargs = {})
    %_param_constant134 : [#users=1] = get_attr[target=_param_constant134]
    %t_43 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant134,), kwargs = {})
    %view_111 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_67, [2048, 640]), kwargs = {})
    %mm_18 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_111, %t_43), kwargs = {})
    %_unsafe_view_46 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_18, [2, 1024, 640]), kwargs = {})
    %view_112 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_46, [2, 1024, 10, 64]), kwargs = {})
    %permute_31 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_112, [0, 2, 1, 3]), kwargs = {})
    %clone_27 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_31,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_47 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_27, [20, 1024, 64]), kwargs = {})
    %_param_constant135 : [#users=1] = get_attr[target=_param_constant135]
    %t_44 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant135,), kwargs = {})
    %view_113 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_67, [2048, 640]), kwargs = {})
    %mm_19 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_113, %t_44), kwargs = {})
    %_unsafe_view_48 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_19, [2, 1024, 640]), kwargs = {})
    %_param_constant136 : [#users=1] = get_attr[target=_param_constant136]
    %t_45 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant136,), kwargs = {})
    %view_114 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_67, [2048, 640]), kwargs = {})
    %mm_20 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_114, %t_45), kwargs = {})
    %_unsafe_view_49 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_20, [2, 1024, 640]), kwargs = {})
    %view_115 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_48, [2, 1024, 10, 64]), kwargs = {})
    %permute_32 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_115, [0, 2, 1, 3]), kwargs = {})
    %clone_28 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_32,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_50 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_28, [20, 1024, 64]), kwargs = {})
    %view_116 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_49, [2, 1024, 10, 64]), kwargs = {})
    %permute_33 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_116, [0, 2, 1, 3]), kwargs = {})
    %clone_29 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_33,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_51 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_29, [20, 1024, 64]), kwargs = {})
    %empty_6 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([20, 1024, 1024],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_6 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_50, -1, -2), kwargs = {})
    %baddbmm_6 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_6, %_unsafe_view_47, %transpose_6), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_6 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_6, -1, False), kwargs = {})
    %detach_30 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_6,), kwargs = {})
    %bmm_10 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_6, %_unsafe_view_51), kwargs = {})
    %view_117 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_10, [2, 10, 1024, 64]), kwargs = {})
    %permute_34 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_117, [0, 2, 1, 3]), kwargs = {})
    %clone_30 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_34,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_52 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_30, [2, 1024, 640]), kwargs = {})
    %_param_constant137 : [#users=1] = get_attr[target=_param_constant137]
    %t_46 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant137,), kwargs = {})
    %view_118 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_52, [2048, 640]), kwargs = {})
    %_param_constant138 : [#users=1] = get_attr[target=_param_constant138]
    %addmm_21 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant138, %view_118, %t_46), kwargs = {})
    %view_119 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_21, [2, 1024, 640]), kwargs = {})
    %add_68 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_119, %add_65), kwargs = {})
    %var_mean_22 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_68, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_44 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_22, 0), kwargs = {})
    %getitem_45 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_22, 1), kwargs = {})
    %add_69 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_44, 1e-05), kwargs = {})
    %rsqrt_22 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_69,), kwargs = {})
    %sub_22 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_68, %getitem_45), kwargs = {})
    %mul_50 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_22, %rsqrt_22), kwargs = {})
    %_param_constant139 : [#users=1] = get_attr[target=_param_constant139]
    %mul_51 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_50, %_param_constant139), kwargs = {})
    %_param_constant140 : [#users=1] = get_attr[target=_param_constant140]
    %add_70 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_51, %_param_constant140), kwargs = {})
    %_param_constant141 : [#users=1] = get_attr[target=_param_constant141]
    %t_47 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant141,), kwargs = {})
    %view_120 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_70, [2048, 640]), kwargs = {})
    %mm_21 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_120, %t_47), kwargs = {})
    %_unsafe_view_53 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_21, [2, 1024, 640]), kwargs = {})
    %view_121 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_53, [2, 1024, 10, 64]), kwargs = {})
    %permute_35 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_121, [0, 2, 1, 3]), kwargs = {})
    %clone_31 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_35,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_54 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_31, [20, 1024, 64]), kwargs = {})
    %_param_constant142 : [#users=1] = get_attr[target=_param_constant142]
    %t_48 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant142,), kwargs = {})
    %view_122 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_22 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_122, %t_48), kwargs = {})
    %_unsafe_view_55 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_22, [2, 64, 640]), kwargs = {})
    %_param_constant143 : [#users=1] = get_attr[target=_param_constant143]
    %t_49 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant143,), kwargs = {})
    %view_123 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_23 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_123, %t_49), kwargs = {})
    %_unsafe_view_56 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_23, [2, 64, 640]), kwargs = {})
    %view_124 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_55, [2, 64, 10, 64]), kwargs = {})
    %permute_36 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_124, [0, 2, 1, 3]), kwargs = {})
    %clone_32 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_36,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_57 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_32, [20, 64, 64]), kwargs = {})
    %view_125 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_56, [2, 64, 10, 64]), kwargs = {})
    %permute_37 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_125, [0, 2, 1, 3]), kwargs = {})
    %clone_33 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_37,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_58 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_33, [20, 64, 64]), kwargs = {})
    %empty_7 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([20, 1024, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_7 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_57, -1, -2), kwargs = {})
    %baddbmm_7 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_7, %_unsafe_view_54, %transpose_7), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_7 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_7, -1, False), kwargs = {})
    %detach_31 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_7,), kwargs = {})
    %bmm_11 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_7, %_unsafe_view_58), kwargs = {})
    %view_126 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_11, [2, 10, 1024, 64]), kwargs = {})
    %permute_38 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_126, [0, 2, 1, 3]), kwargs = {})
    %clone_34 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_38,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_59 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_34, [2, 1024, 640]), kwargs = {})
    %_param_constant144 : [#users=1] = get_attr[target=_param_constant144]
    %t_50 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant144,), kwargs = {})
    %view_127 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_59, [2048, 640]), kwargs = {})
    %_param_constant145 : [#users=1] = get_attr[target=_param_constant145]
    %addmm_22 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant145, %view_127, %t_50), kwargs = {})
    %view_128 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_22, [2, 1024, 640]), kwargs = {})
    %add_71 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_128, %add_68), kwargs = {})
    %var_mean_23 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_71, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_46 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_23, 0), kwargs = {})
    %getitem_47 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_23, 1), kwargs = {})
    %add_72 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_46, 1e-05), kwargs = {})
    %rsqrt_23 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_72,), kwargs = {})
    %sub_23 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_71, %getitem_47), kwargs = {})
    %mul_52 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_23, %rsqrt_23), kwargs = {})
    %_param_constant146 : [#users=1] = get_attr[target=_param_constant146]
    %mul_53 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_52, %_param_constant146), kwargs = {})
    %_param_constant147 : [#users=1] = get_attr[target=_param_constant147]
    %add_73 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_53, %_param_constant147), kwargs = {})
    %_param_constant148 : [#users=1] = get_attr[target=_param_constant148]
    %t_51 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant148,), kwargs = {})
    %view_129 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_73, [2048, 640]), kwargs = {})
    %_param_constant149 : [#users=1] = get_attr[target=_param_constant149]
    %addmm_23 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant149, %view_129, %t_51), kwargs = {})
    %view_130 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_23, [2, 1024, 5120]), kwargs = {})
    %slice_21 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_130, -1, 0, 2560), kwargs = {})
    %slice_22 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_130, -1, 2560, 5120), kwargs = {})
    %gelu_3 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_22,), kwargs = {})
    %mul_54 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_21, %gelu_3), kwargs = {})
    %_param_constant150 : [#users=1] = get_attr[target=_param_constant150]
    %t_52 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant150,), kwargs = {})
    %view_131 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_54, [2048, 2560]), kwargs = {})
    %_param_constant151 : [#users=1] = get_attr[target=_param_constant151]
    %addmm_24 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant151, %view_131, %t_52), kwargs = {})
    %view_132 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_24, [2, 1024, 640]), kwargs = {})
    %add_74 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_132, %add_71), kwargs = {})
    %_param_constant152 : [#users=1] = get_attr[target=_param_constant152]
    %t_53 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant152,), kwargs = {})
    %view_133 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_74, [2048, 640]), kwargs = {})
    %_param_constant153 : [#users=1] = get_attr[target=_param_constant153]
    %addmm_25 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant153, %view_133, %t_53), kwargs = {})
    %view_134 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_25, [2, 1024, 640]), kwargs = {})
    %view_135 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_134, [2, 32, 32, 640]), kwargs = {})
    %permute_39 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_135, [0, 3, 1, 2]), kwargs = {})
    %clone_35 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_39,), kwargs = {memory_format: torch.contiguous_format})
    %add_75 : [#users=2] = call_function[target=torch.ops.aten.add](args = (%clone_35, %div_4), kwargs = {})
    %_param_constant154 : [#users=1] = get_attr[target=_param_constant154]
    %_param_constant155 : [#users=1] = get_attr[target=_param_constant155]
    %convolution_11 : [#users=3] = call_function[target=torch.ops.aten.convolution](args = (%add_75, %_param_constant154, %_param_constant155, [2, 2], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %view_136 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%convolution_11, [2, 32, 20, 256]), kwargs = {})
    %var_mean_24 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_136, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_48 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_24, 0), kwargs = {})
    %getitem_49 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_24, 1), kwargs = {})
    %add_76 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_48, 1e-05), kwargs = {})
    %rsqrt_24 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_76,), kwargs = {})
    %sub_24 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_136, %getitem_49), kwargs = {})
    %mul_55 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_24, %rsqrt_24), kwargs = {})
    %view_137 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_55, [2, 640, 16, 16]), kwargs = {})
    %_param_constant156 : [#users=1] = get_attr[target=_param_constant156]
    %unsqueeze_82 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant156, 0), kwargs = {})
    %unsqueeze_83 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_82, 2), kwargs = {})
    %unsqueeze_84 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_83, 3), kwargs = {})
    %_param_constant157 : [#users=1] = get_attr[target=_param_constant157]
    %unsqueeze_85 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant157, 0), kwargs = {})
    %unsqueeze_86 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_85, 2), kwargs = {})
    %unsqueeze_87 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_86, 3), kwargs = {})
    %mul_56 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_137, %unsqueeze_87), kwargs = {})
    %add_77 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_56, %unsqueeze_84), kwargs = {})
    %squeeze_48 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_49, 3), kwargs = {})
    %squeeze_49 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_48, 2), kwargs = {})
    %squeeze_50 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_24, 3), kwargs = {})
    %squeeze_51 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_50, 2), kwargs = {})
    %detach_32 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_49,), kwargs = {})
    %detach_33 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_51,), kwargs = {})
    %silu_13 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_77,), kwargs = {})
    %_param_constant158 : [#users=1] = get_attr[target=_param_constant158]
    %_param_constant159 : [#users=1] = get_attr[target=_param_constant159]
    %convolution_12 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_13, %_param_constant158, %_param_constant159, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_14 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant160 : [#users=1] = get_attr[target=_param_constant160]
    %t_54 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant160,), kwargs = {})
    %_param_constant161 : [#users=1] = get_attr[target=_param_constant161]
    %addmm_26 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant161, %silu_14, %t_54), kwargs = {})
    %slice_23 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_26, 0, 0, 9223372036854775807), kwargs = {})
    %slice_24 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_23, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_88 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_24, 2), kwargs = {})
    %unsqueeze_89 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_88, 3), kwargs = {})
    %add_78 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_12, %unsqueeze_89), kwargs = {})
    %view_138 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_78, [2, 32, 40, 256]), kwargs = {})
    %var_mean_25 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_138, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_50 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_25, 0), kwargs = {})
    %getitem_51 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_25, 1), kwargs = {})
    %add_79 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_50, 1e-05), kwargs = {})
    %rsqrt_25 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_79,), kwargs = {})
    %sub_25 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_138, %getitem_51), kwargs = {})
    %mul_57 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_25, %rsqrt_25), kwargs = {})
    %view_139 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_57, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant162 : [#users=1] = get_attr[target=_param_constant162]
    %unsqueeze_90 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant162, 0), kwargs = {})
    %unsqueeze_91 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_90, 2), kwargs = {})
    %unsqueeze_92 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_91, 3), kwargs = {})
    %_param_constant163 : [#users=1] = get_attr[target=_param_constant163]
    %unsqueeze_93 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant163, 0), kwargs = {})
    %unsqueeze_94 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_93, 2), kwargs = {})
    %unsqueeze_95 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_94, 3), kwargs = {})
    %mul_58 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_139, %unsqueeze_95), kwargs = {})
    %add_80 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_58, %unsqueeze_92), kwargs = {})
    %squeeze_52 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_51, 3), kwargs = {})
    %squeeze_53 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_52, 2), kwargs = {})
    %squeeze_54 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_25, 3), kwargs = {})
    %squeeze_55 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_54, 2), kwargs = {})
    %detach_34 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_53,), kwargs = {})
    %detach_35 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_55,), kwargs = {})
    %silu_15 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_80,), kwargs = {})
    %_param_constant164 : [#users=1] = get_attr[target=_param_constant164]
    %_param_constant165 : [#users=1] = get_attr[target=_param_constant165]
    %convolution_13 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_15, %_param_constant164, %_param_constant165, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant166 : [#users=1] = get_attr[target=_param_constant166]
    %_param_constant167 : [#users=1] = get_attr[target=_param_constant167]
    %convolution_14 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%convolution_11, %_param_constant166, %_param_constant167, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_81 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_14, %convolution_13), kwargs = {})
    %div_5 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_81, 1.0), kwargs = {})
    %view_140 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_5, [2, 32, 40, 256]), kwargs = {})
    %var_mean_26 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_140, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_52 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_26, 0), kwargs = {})
    %getitem_53 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_26, 1), kwargs = {})
    %add_82 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_52, 1e-06), kwargs = {})
    %rsqrt_26 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_82,), kwargs = {})
    %sub_26 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_140, %getitem_53), kwargs = {})
    %mul_59 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_26, %rsqrt_26), kwargs = {})
    %view_141 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_59, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant168 : [#users=1] = get_attr[target=_param_constant168]
    %unsqueeze_96 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant168, 0), kwargs = {})
    %unsqueeze_97 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_96, 2), kwargs = {})
    %unsqueeze_98 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_97, 3), kwargs = {})
    %_param_constant169 : [#users=1] = get_attr[target=_param_constant169]
    %unsqueeze_99 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant169, 0), kwargs = {})
    %unsqueeze_100 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_99, 2), kwargs = {})
    %unsqueeze_101 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_100, 3), kwargs = {})
    %mul_60 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_141, %unsqueeze_101), kwargs = {})
    %add_83 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_60, %unsqueeze_98), kwargs = {})
    %squeeze_56 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_53, 3), kwargs = {})
    %squeeze_57 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_56, 2), kwargs = {})
    %squeeze_58 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_26, 3), kwargs = {})
    %squeeze_59 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_58, 2), kwargs = {})
    %detach_36 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_57,), kwargs = {})
    %detach_37 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_59,), kwargs = {})
    %permute_40 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_83, [0, 2, 3, 1]), kwargs = {})
    %view_142 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_40, [2, 256, 1280]), kwargs = {})
    %_param_constant170 : [#users=1] = get_attr[target=_param_constant170]
    %t_55 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant170,), kwargs = {})
    %expand_9 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_142, [2, 256, 1280]), kwargs = {})
    %view_143 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_9, [2, 256, 1280]), kwargs = {})
    %expand_10 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_55, [2, 1280, 1280]), kwargs = {})
    %view_144 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_10, [2, 1280, 1280]), kwargs = {})
    %bmm_12 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_143, %view_144), kwargs = {})
    %_unsafe_view_60 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_12, [2, 256, 1280]), kwargs = {})
    %_param_constant171 : [#users=1] = get_attr[target=_param_constant171]
    %add_84 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_60, %_param_constant171), kwargs = {})
    %var_mean_27 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_84, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_54 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_27, 0), kwargs = {})
    %getitem_55 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_27, 1), kwargs = {})
    %add_85 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_54, 1e-05), kwargs = {})
    %rsqrt_27 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_85,), kwargs = {})
    %sub_27 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_84, %getitem_55), kwargs = {})
    %mul_61 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_27, %rsqrt_27), kwargs = {})
    %_param_constant172 : [#users=1] = get_attr[target=_param_constant172]
    %mul_62 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_61, %_param_constant172), kwargs = {})
    %_param_constant173 : [#users=1] = get_attr[target=_param_constant173]
    %add_86 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_62, %_param_constant173), kwargs = {})
    %_param_constant174 : [#users=1] = get_attr[target=_param_constant174]
    %t_56 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant174,), kwargs = {})
    %view_145 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_86, [512, 1280]), kwargs = {})
    %mm_24 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_145, %t_56), kwargs = {})
    %_unsafe_view_61 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_24, [2, 256, 1280]), kwargs = {})
    %view_146 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_61, [2, 256, 20, 64]), kwargs = {})
    %permute_41 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_146, [0, 2, 1, 3]), kwargs = {})
    %clone_36 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_41,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_62 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_36, [40, 256, 64]), kwargs = {})
    %_param_constant175 : [#users=1] = get_attr[target=_param_constant175]
    %t_57 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant175,), kwargs = {})
    %view_147 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_86, [512, 1280]), kwargs = {})
    %mm_25 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_147, %t_57), kwargs = {})
    %_unsafe_view_63 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_25, [2, 256, 1280]), kwargs = {})
    %_param_constant176 : [#users=1] = get_attr[target=_param_constant176]
    %t_58 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant176,), kwargs = {})
    %view_148 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_86, [512, 1280]), kwargs = {})
    %mm_26 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_148, %t_58), kwargs = {})
    %_unsafe_view_64 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_26, [2, 256, 1280]), kwargs = {})
    %view_149 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_63, [2, 256, 20, 64]), kwargs = {})
    %permute_42 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_149, [0, 2, 1, 3]), kwargs = {})
    %clone_37 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_42,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_65 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_37, [40, 256, 64]), kwargs = {})
    %view_150 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_64, [2, 256, 20, 64]), kwargs = {})
    %permute_43 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_150, [0, 2, 1, 3]), kwargs = {})
    %clone_38 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_43,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_66 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_38, [40, 256, 64]), kwargs = {})
    %empty_8 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 256, 256],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_8 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_65, -1, -2), kwargs = {})
    %baddbmm_8 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_8, %_unsafe_view_62, %transpose_8), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_8 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_8, -1, False), kwargs = {})
    %detach_38 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_8,), kwargs = {})
    %bmm_13 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_8, %_unsafe_view_66), kwargs = {})
    %view_151 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_13, [2, 20, 256, 64]), kwargs = {})
    %permute_44 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_151, [0, 2, 1, 3]), kwargs = {})
    %clone_39 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_44,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_67 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_39, [2, 256, 1280]), kwargs = {})
    %_param_constant177 : [#users=1] = get_attr[target=_param_constant177]
    %t_59 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant177,), kwargs = {})
    %view_152 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_67, [512, 1280]), kwargs = {})
    %_param_constant178 : [#users=1] = get_attr[target=_param_constant178]
    %addmm_27 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant178, %view_152, %t_59), kwargs = {})
    %view_153 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_27, [2, 256, 1280]), kwargs = {})
    %add_87 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_153, %add_84), kwargs = {})
    %var_mean_28 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_87, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_56 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_28, 0), kwargs = {})
    %getitem_57 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_28, 1), kwargs = {})
    %add_88 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_56, 1e-05), kwargs = {})
    %rsqrt_28 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_88,), kwargs = {})
    %sub_28 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_87, %getitem_57), kwargs = {})
    %mul_63 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_28, %rsqrt_28), kwargs = {})
    %_param_constant179 : [#users=1] = get_attr[target=_param_constant179]
    %mul_64 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_63, %_param_constant179), kwargs = {})
    %_param_constant180 : [#users=1] = get_attr[target=_param_constant180]
    %add_89 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_64, %_param_constant180), kwargs = {})
    %_param_constant181 : [#users=1] = get_attr[target=_param_constant181]
    %t_60 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant181,), kwargs = {})
    %view_154 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_89, [512, 1280]), kwargs = {})
    %mm_27 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_154, %t_60), kwargs = {})
    %_unsafe_view_68 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_27, [2, 256, 1280]), kwargs = {})
    %view_155 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_68, [2, 256, 20, 64]), kwargs = {})
    %permute_45 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_155, [0, 2, 1, 3]), kwargs = {})
    %clone_40 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_45,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_69 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_40, [40, 256, 64]), kwargs = {})
    %_param_constant182 : [#users=1] = get_attr[target=_param_constant182]
    %t_61 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant182,), kwargs = {})
    %view_156 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_28 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_156, %t_61), kwargs = {})
    %_unsafe_view_70 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_28, [2, 64, 1280]), kwargs = {})
    %_param_constant183 : [#users=1] = get_attr[target=_param_constant183]
    %t_62 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant183,), kwargs = {})
    %view_157 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_29 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_157, %t_62), kwargs = {})
    %_unsafe_view_71 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_29, [2, 64, 1280]), kwargs = {})
    %view_158 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_70, [2, 64, 20, 64]), kwargs = {})
    %permute_46 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_158, [0, 2, 1, 3]), kwargs = {})
    %clone_41 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_46,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_72 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_41, [40, 64, 64]), kwargs = {})
    %view_159 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_71, [2, 64, 20, 64]), kwargs = {})
    %permute_47 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_159, [0, 2, 1, 3]), kwargs = {})
    %clone_42 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_47,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_73 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_42, [40, 64, 64]), kwargs = {})
    %empty_9 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 256, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_9 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_72, -1, -2), kwargs = {})
    %baddbmm_9 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_9, %_unsafe_view_69, %transpose_9), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_9 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_9, -1, False), kwargs = {})
    %detach_39 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_9,), kwargs = {})
    %bmm_14 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_9, %_unsafe_view_73), kwargs = {})
    %view_160 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_14, [2, 20, 256, 64]), kwargs = {})
    %permute_48 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_160, [0, 2, 1, 3]), kwargs = {})
    %clone_43 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_48,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_74 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_43, [2, 256, 1280]), kwargs = {})
    %_param_constant184 : [#users=1] = get_attr[target=_param_constant184]
    %t_63 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant184,), kwargs = {})
    %view_161 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_74, [512, 1280]), kwargs = {})
    %_param_constant185 : [#users=1] = get_attr[target=_param_constant185]
    %addmm_28 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant185, %view_161, %t_63), kwargs = {})
    %view_162 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_28, [2, 256, 1280]), kwargs = {})
    %add_90 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_162, %add_87), kwargs = {})
    %var_mean_29 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_90, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_58 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_29, 0), kwargs = {})
    %getitem_59 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_29, 1), kwargs = {})
    %add_91 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_58, 1e-05), kwargs = {})
    %rsqrt_29 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_91,), kwargs = {})
    %sub_29 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_90, %getitem_59), kwargs = {})
    %mul_65 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_29, %rsqrt_29), kwargs = {})
    %_param_constant186 : [#users=1] = get_attr[target=_param_constant186]
    %mul_66 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_65, %_param_constant186), kwargs = {})
    %_param_constant187 : [#users=1] = get_attr[target=_param_constant187]
    %add_92 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_66, %_param_constant187), kwargs = {})
    %_param_constant188 : [#users=1] = get_attr[target=_param_constant188]
    %t_64 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant188,), kwargs = {})
    %view_163 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_92, [512, 1280]), kwargs = {})
    %_param_constant189 : [#users=1] = get_attr[target=_param_constant189]
    %addmm_29 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant189, %view_163, %t_64), kwargs = {})
    %view_164 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_29, [2, 256, 10240]), kwargs = {})
    %slice_25 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_164, -1, 0, 5120), kwargs = {})
    %slice_26 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_164, -1, 5120, 10240), kwargs = {})
    %gelu_4 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_26,), kwargs = {})
    %mul_67 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_25, %gelu_4), kwargs = {})
    %_param_constant190 : [#users=1] = get_attr[target=_param_constant190]
    %t_65 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant190,), kwargs = {})
    %view_165 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_67, [512, 5120]), kwargs = {})
    %_param_constant191 : [#users=1] = get_attr[target=_param_constant191]
    %addmm_30 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant191, %view_165, %t_65), kwargs = {})
    %view_166 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_30, [2, 256, 1280]), kwargs = {})
    %add_93 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_166, %add_90), kwargs = {})
    %_param_constant192 : [#users=1] = get_attr[target=_param_constant192]
    %t_66 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant192,), kwargs = {})
    %view_167 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_93, [512, 1280]), kwargs = {})
    %_param_constant193 : [#users=1] = get_attr[target=_param_constant193]
    %addmm_31 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant193, %view_167, %t_66), kwargs = {})
    %view_168 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_31, [2, 256, 1280]), kwargs = {})
    %view_169 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_168, [2, 16, 16, 1280]), kwargs = {})
    %permute_49 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_169, [0, 3, 1, 2]), kwargs = {})
    %clone_44 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_49,), kwargs = {memory_format: torch.contiguous_format})
    %add_94 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%clone_44, %div_5), kwargs = {})
    %view_170 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_94, [2, 32, 40, 256]), kwargs = {})
    %var_mean_30 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_170, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_60 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_30, 0), kwargs = {})
    %getitem_61 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_30, 1), kwargs = {})
    %add_95 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_60, 1e-05), kwargs = {})
    %rsqrt_30 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_95,), kwargs = {})
    %sub_30 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_170, %getitem_61), kwargs = {})
    %mul_68 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_30, %rsqrt_30), kwargs = {})
    %view_171 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_68, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant194 : [#users=1] = get_attr[target=_param_constant194]
    %unsqueeze_102 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant194, 0), kwargs = {})
    %unsqueeze_103 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_102, 2), kwargs = {})
    %unsqueeze_104 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_103, 3), kwargs = {})
    %_param_constant195 : [#users=1] = get_attr[target=_param_constant195]
    %unsqueeze_105 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant195, 0), kwargs = {})
    %unsqueeze_106 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_105, 2), kwargs = {})
    %unsqueeze_107 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_106, 3), kwargs = {})
    %mul_69 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_171, %unsqueeze_107), kwargs = {})
    %add_96 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_69, %unsqueeze_104), kwargs = {})
    %squeeze_60 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_61, 3), kwargs = {})
    %squeeze_61 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_60, 2), kwargs = {})
    %squeeze_62 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_30, 3), kwargs = {})
    %squeeze_63 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_62, 2), kwargs = {})
    %detach_40 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_61,), kwargs = {})
    %detach_41 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_63,), kwargs = {})
    %silu_16 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_96,), kwargs = {})
    %_param_constant196 : [#users=1] = get_attr[target=_param_constant196]
    %_param_constant197 : [#users=1] = get_attr[target=_param_constant197]
    %convolution_15 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_16, %_param_constant196, %_param_constant197, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_17 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant198 : [#users=1] = get_attr[target=_param_constant198]
    %t_67 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant198,), kwargs = {})
    %_param_constant199 : [#users=1] = get_attr[target=_param_constant199]
    %addmm_32 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant199, %silu_17, %t_67), kwargs = {})
    %slice_27 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_32, 0, 0, 9223372036854775807), kwargs = {})
    %slice_28 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_27, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_108 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_28, 2), kwargs = {})
    %unsqueeze_109 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_108, 3), kwargs = {})
    %add_97 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_15, %unsqueeze_109), kwargs = {})
    %view_172 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_97, [2, 32, 40, 256]), kwargs = {})
    %var_mean_31 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_172, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_62 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_31, 0), kwargs = {})
    %getitem_63 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_31, 1), kwargs = {})
    %add_98 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_62, 1e-05), kwargs = {})
    %rsqrt_31 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_98,), kwargs = {})
    %sub_31 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_172, %getitem_63), kwargs = {})
    %mul_70 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_31, %rsqrt_31), kwargs = {})
    %view_173 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_70, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant200 : [#users=1] = get_attr[target=_param_constant200]
    %unsqueeze_110 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant200, 0), kwargs = {})
    %unsqueeze_111 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_110, 2), kwargs = {})
    %unsqueeze_112 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_111, 3), kwargs = {})
    %_param_constant201 : [#users=1] = get_attr[target=_param_constant201]
    %unsqueeze_113 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant201, 0), kwargs = {})
    %unsqueeze_114 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_113, 2), kwargs = {})
    %unsqueeze_115 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_114, 3), kwargs = {})
    %mul_71 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_173, %unsqueeze_115), kwargs = {})
    %add_99 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_71, %unsqueeze_112), kwargs = {})
    %squeeze_64 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_63, 3), kwargs = {})
    %squeeze_65 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_64, 2), kwargs = {})
    %squeeze_66 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_31, 3), kwargs = {})
    %squeeze_67 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_66, 2), kwargs = {})
    %detach_42 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_65,), kwargs = {})
    %detach_43 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_67,), kwargs = {})
    %silu_18 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_99,), kwargs = {})
    %_param_constant202 : [#users=1] = get_attr[target=_param_constant202]
    %_param_constant203 : [#users=1] = get_attr[target=_param_constant203]
    %convolution_16 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_18, %_param_constant202, %_param_constant203, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_100 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%add_94, %convolution_16), kwargs = {})
    %div_6 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_100, 1.0), kwargs = {})
    %view_174 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_6, [2, 32, 40, 256]), kwargs = {})
    %var_mean_32 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_174, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_64 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_32, 0), kwargs = {})
    %getitem_65 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_32, 1), kwargs = {})
    %add_101 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_64, 1e-06), kwargs = {})
    %rsqrt_32 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_101,), kwargs = {})
    %sub_32 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_174, %getitem_65), kwargs = {})
    %mul_72 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_32, %rsqrt_32), kwargs = {})
    %view_175 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_72, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant204 : [#users=1] = get_attr[target=_param_constant204]
    %unsqueeze_116 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant204, 0), kwargs = {})
    %unsqueeze_117 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_116, 2), kwargs = {})
    %unsqueeze_118 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_117, 3), kwargs = {})
    %_param_constant205 : [#users=1] = get_attr[target=_param_constant205]
    %unsqueeze_119 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant205, 0), kwargs = {})
    %unsqueeze_120 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_119, 2), kwargs = {})
    %unsqueeze_121 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_120, 3), kwargs = {})
    %mul_73 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_175, %unsqueeze_121), kwargs = {})
    %add_102 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_73, %unsqueeze_118), kwargs = {})
    %squeeze_68 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_65, 3), kwargs = {})
    %squeeze_69 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_68, 2), kwargs = {})
    %squeeze_70 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_32, 3), kwargs = {})
    %squeeze_71 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_70, 2), kwargs = {})
    %detach_44 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_69,), kwargs = {})
    %detach_45 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_71,), kwargs = {})
    %permute_50 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_102, [0, 2, 3, 1]), kwargs = {})
    %view_176 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_50, [2, 256, 1280]), kwargs = {})
    %_param_constant206 : [#users=1] = get_attr[target=_param_constant206]
    %t_68 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant206,), kwargs = {})
    %expand_11 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_176, [2, 256, 1280]), kwargs = {})
    %view_177 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_11, [2, 256, 1280]), kwargs = {})
    %expand_12 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_68, [2, 1280, 1280]), kwargs = {})
    %view_178 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_12, [2, 1280, 1280]), kwargs = {})
    %bmm_15 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_177, %view_178), kwargs = {})
    %_unsafe_view_75 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_15, [2, 256, 1280]), kwargs = {})
    %_param_constant207 : [#users=1] = get_attr[target=_param_constant207]
    %add_103 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_75, %_param_constant207), kwargs = {})
    %var_mean_33 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_103, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_66 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_33, 0), kwargs = {})
    %getitem_67 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_33, 1), kwargs = {})
    %add_104 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_66, 1e-05), kwargs = {})
    %rsqrt_33 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_104,), kwargs = {})
    %sub_33 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_103, %getitem_67), kwargs = {})
    %mul_74 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_33, %rsqrt_33), kwargs = {})
    %_param_constant208 : [#users=1] = get_attr[target=_param_constant208]
    %mul_75 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_74, %_param_constant208), kwargs = {})
    %_param_constant209 : [#users=1] = get_attr[target=_param_constant209]
    %add_105 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_75, %_param_constant209), kwargs = {})
    %_param_constant210 : [#users=1] = get_attr[target=_param_constant210]
    %t_69 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant210,), kwargs = {})
    %view_179 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_105, [512, 1280]), kwargs = {})
    %mm_30 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_179, %t_69), kwargs = {})
    %_unsafe_view_76 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_30, [2, 256, 1280]), kwargs = {})
    %view_180 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_76, [2, 256, 20, 64]), kwargs = {})
    %permute_51 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_180, [0, 2, 1, 3]), kwargs = {})
    %clone_45 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_51,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_77 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_45, [40, 256, 64]), kwargs = {})
    %_param_constant211 : [#users=1] = get_attr[target=_param_constant211]
    %t_70 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant211,), kwargs = {})
    %view_181 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_105, [512, 1280]), kwargs = {})
    %mm_31 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_181, %t_70), kwargs = {})
    %_unsafe_view_78 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_31, [2, 256, 1280]), kwargs = {})
    %_param_constant212 : [#users=1] = get_attr[target=_param_constant212]
    %t_71 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant212,), kwargs = {})
    %view_182 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_105, [512, 1280]), kwargs = {})
    %mm_32 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_182, %t_71), kwargs = {})
    %_unsafe_view_79 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_32, [2, 256, 1280]), kwargs = {})
    %view_183 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_78, [2, 256, 20, 64]), kwargs = {})
    %permute_52 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_183, [0, 2, 1, 3]), kwargs = {})
    %clone_46 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_52,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_80 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_46, [40, 256, 64]), kwargs = {})
    %view_184 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_79, [2, 256, 20, 64]), kwargs = {})
    %permute_53 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_184, [0, 2, 1, 3]), kwargs = {})
    %clone_47 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_53,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_81 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_47, [40, 256, 64]), kwargs = {})
    %empty_10 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 256, 256],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_10 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_80, -1, -2), kwargs = {})
    %baddbmm_10 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_10, %_unsafe_view_77, %transpose_10), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_10 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_10, -1, False), kwargs = {})
    %detach_46 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_10,), kwargs = {})
    %bmm_16 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_10, %_unsafe_view_81), kwargs = {})
    %view_185 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_16, [2, 20, 256, 64]), kwargs = {})
    %permute_54 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_185, [0, 2, 1, 3]), kwargs = {})
    %clone_48 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_54,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_82 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_48, [2, 256, 1280]), kwargs = {})
    %_param_constant213 : [#users=1] = get_attr[target=_param_constant213]
    %t_72 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant213,), kwargs = {})
    %view_186 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_82, [512, 1280]), kwargs = {})
    %_param_constant214 : [#users=1] = get_attr[target=_param_constant214]
    %addmm_33 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant214, %view_186, %t_72), kwargs = {})
    %view_187 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_33, [2, 256, 1280]), kwargs = {})
    %add_106 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_187, %add_103), kwargs = {})
    %var_mean_34 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_106, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_68 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_34, 0), kwargs = {})
    %getitem_69 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_34, 1), kwargs = {})
    %add_107 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_68, 1e-05), kwargs = {})
    %rsqrt_34 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_107,), kwargs = {})
    %sub_34 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_106, %getitem_69), kwargs = {})
    %mul_76 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_34, %rsqrt_34), kwargs = {})
    %_param_constant215 : [#users=1] = get_attr[target=_param_constant215]
    %mul_77 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_76, %_param_constant215), kwargs = {})
    %_param_constant216 : [#users=1] = get_attr[target=_param_constant216]
    %add_108 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_77, %_param_constant216), kwargs = {})
    %_param_constant217 : [#users=1] = get_attr[target=_param_constant217]
    %t_73 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant217,), kwargs = {})
    %view_188 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_108, [512, 1280]), kwargs = {})
    %mm_33 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_188, %t_73), kwargs = {})
    %_unsafe_view_83 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_33, [2, 256, 1280]), kwargs = {})
    %view_189 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_83, [2, 256, 20, 64]), kwargs = {})
    %permute_55 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_189, [0, 2, 1, 3]), kwargs = {})
    %clone_49 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_55,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_84 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_49, [40, 256, 64]), kwargs = {})
    %_param_constant218 : [#users=1] = get_attr[target=_param_constant218]
    %t_74 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant218,), kwargs = {})
    %view_190 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_34 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_190, %t_74), kwargs = {})
    %_unsafe_view_85 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_34, [2, 64, 1280]), kwargs = {})
    %_param_constant219 : [#users=1] = get_attr[target=_param_constant219]
    %t_75 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant219,), kwargs = {})
    %view_191 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_35 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_191, %t_75), kwargs = {})
    %_unsafe_view_86 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_35, [2, 64, 1280]), kwargs = {})
    %view_192 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_85, [2, 64, 20, 64]), kwargs = {})
    %permute_56 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_192, [0, 2, 1, 3]), kwargs = {})
    %clone_50 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_56,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_87 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_50, [40, 64, 64]), kwargs = {})
    %view_193 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_86, [2, 64, 20, 64]), kwargs = {})
    %permute_57 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_193, [0, 2, 1, 3]), kwargs = {})
    %clone_51 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_57,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_88 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_51, [40, 64, 64]), kwargs = {})
    %empty_11 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 256, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_11 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_87, -1, -2), kwargs = {})
    %baddbmm_11 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_11, %_unsafe_view_84, %transpose_11), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_11 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_11, -1, False), kwargs = {})
    %detach_47 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_11,), kwargs = {})
    %bmm_17 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_11, %_unsafe_view_88), kwargs = {})
    %view_194 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_17, [2, 20, 256, 64]), kwargs = {})
    %permute_58 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_194, [0, 2, 1, 3]), kwargs = {})
    %clone_52 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_58,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_89 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_52, [2, 256, 1280]), kwargs = {})
    %_param_constant220 : [#users=1] = get_attr[target=_param_constant220]
    %t_76 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant220,), kwargs = {})
    %view_195 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_89, [512, 1280]), kwargs = {})
    %_param_constant221 : [#users=1] = get_attr[target=_param_constant221]
    %addmm_34 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant221, %view_195, %t_76), kwargs = {})
    %view_196 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_34, [2, 256, 1280]), kwargs = {})
    %add_109 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_196, %add_106), kwargs = {})
    %var_mean_35 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_109, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_70 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_35, 0), kwargs = {})
    %getitem_71 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_35, 1), kwargs = {})
    %add_110 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_70, 1e-05), kwargs = {})
    %rsqrt_35 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_110,), kwargs = {})
    %sub_35 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_109, %getitem_71), kwargs = {})
    %mul_78 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_35, %rsqrt_35), kwargs = {})
    %_param_constant222 : [#users=1] = get_attr[target=_param_constant222]
    %mul_79 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_78, %_param_constant222), kwargs = {})
    %_param_constant223 : [#users=1] = get_attr[target=_param_constant223]
    %add_111 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_79, %_param_constant223), kwargs = {})
    %_param_constant224 : [#users=1] = get_attr[target=_param_constant224]
    %t_77 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant224,), kwargs = {})
    %view_197 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_111, [512, 1280]), kwargs = {})
    %_param_constant225 : [#users=1] = get_attr[target=_param_constant225]
    %addmm_35 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant225, %view_197, %t_77), kwargs = {})
    %view_198 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_35, [2, 256, 10240]), kwargs = {})
    %slice_29 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_198, -1, 0, 5120), kwargs = {})
    %slice_30 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_198, -1, 5120, 10240), kwargs = {})
    %gelu_5 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_30,), kwargs = {})
    %mul_80 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_29, %gelu_5), kwargs = {})
    %_param_constant226 : [#users=1] = get_attr[target=_param_constant226]
    %t_78 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant226,), kwargs = {})
    %view_199 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_80, [512, 5120]), kwargs = {})
    %_param_constant227 : [#users=1] = get_attr[target=_param_constant227]
    %addmm_36 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant227, %view_199, %t_78), kwargs = {})
    %view_200 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_36, [2, 256, 1280]), kwargs = {})
    %add_112 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_200, %add_109), kwargs = {})
    %_param_constant228 : [#users=1] = get_attr[target=_param_constant228]
    %t_79 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant228,), kwargs = {})
    %view_201 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_112, [512, 1280]), kwargs = {})
    %_param_constant229 : [#users=1] = get_attr[target=_param_constant229]
    %addmm_37 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant229, %view_201, %t_79), kwargs = {})
    %view_202 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_37, [2, 256, 1280]), kwargs = {})
    %view_203 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_202, [2, 16, 16, 1280]), kwargs = {})
    %permute_59 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_203, [0, 3, 1, 2]), kwargs = {})
    %clone_53 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_59,), kwargs = {memory_format: torch.contiguous_format})
    %add_113 : [#users=2] = call_function[target=torch.ops.aten.add](args = (%clone_53, %div_6), kwargs = {})
    %_param_constant230 : [#users=1] = get_attr[target=_param_constant230]
    %_param_constant231 : [#users=1] = get_attr[target=_param_constant231]
    %convolution_17 : [#users=3] = call_function[target=torch.ops.aten.convolution](args = (%add_113, %_param_constant230, %_param_constant231, [2, 2], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %view_204 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%convolution_17, [2, 32, 40, 64]), kwargs = {})
    %var_mean_36 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_204, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_72 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_36, 0), kwargs = {})
    %getitem_73 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_36, 1), kwargs = {})
    %add_114 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_72, 1e-05), kwargs = {})
    %rsqrt_36 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_114,), kwargs = {})
    %sub_36 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_204, %getitem_73), kwargs = {})
    %mul_81 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_36, %rsqrt_36), kwargs = {})
    %view_205 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_81, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant232 : [#users=1] = get_attr[target=_param_constant232]
    %unsqueeze_122 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant232, 0), kwargs = {})
    %unsqueeze_123 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_122, 2), kwargs = {})
    %unsqueeze_124 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_123, 3), kwargs = {})
    %_param_constant233 : [#users=1] = get_attr[target=_param_constant233]
    %unsqueeze_125 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant233, 0), kwargs = {})
    %unsqueeze_126 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_125, 2), kwargs = {})
    %unsqueeze_127 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_126, 3), kwargs = {})
    %mul_82 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_205, %unsqueeze_127), kwargs = {})
    %add_115 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_82, %unsqueeze_124), kwargs = {})
    %squeeze_72 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_73, 3), kwargs = {})
    %squeeze_73 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_72, 2), kwargs = {})
    %squeeze_74 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_36, 3), kwargs = {})
    %squeeze_75 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_74, 2), kwargs = {})
    %detach_48 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_73,), kwargs = {})
    %detach_49 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_75,), kwargs = {})
    %silu_19 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_115,), kwargs = {})
    %_param_constant234 : [#users=1] = get_attr[target=_param_constant234]
    %_param_constant235 : [#users=1] = get_attr[target=_param_constant235]
    %convolution_18 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_19, %_param_constant234, %_param_constant235, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_20 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant236 : [#users=1] = get_attr[target=_param_constant236]
    %t_80 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant236,), kwargs = {})
    %_param_constant237 : [#users=1] = get_attr[target=_param_constant237]
    %addmm_38 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant237, %silu_20, %t_80), kwargs = {})
    %slice_31 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_38, 0, 0, 9223372036854775807), kwargs = {})
    %slice_32 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_31, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_128 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_32, 2), kwargs = {})
    %unsqueeze_129 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_128, 3), kwargs = {})
    %add_116 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_18, %unsqueeze_129), kwargs = {})
    %view_206 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_116, [2, 32, 40, 64]), kwargs = {})
    %var_mean_37 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_206, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_74 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_37, 0), kwargs = {})
    %getitem_75 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_37, 1), kwargs = {})
    %add_117 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_74, 1e-05), kwargs = {})
    %rsqrt_37 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_117,), kwargs = {})
    %sub_37 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_206, %getitem_75), kwargs = {})
    %mul_83 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_37, %rsqrt_37), kwargs = {})
    %view_207 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_83, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant238 : [#users=1] = get_attr[target=_param_constant238]
    %unsqueeze_130 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant238, 0), kwargs = {})
    %unsqueeze_131 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_130, 2), kwargs = {})
    %unsqueeze_132 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_131, 3), kwargs = {})
    %_param_constant239 : [#users=1] = get_attr[target=_param_constant239]
    %unsqueeze_133 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant239, 0), kwargs = {})
    %unsqueeze_134 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_133, 2), kwargs = {})
    %unsqueeze_135 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_134, 3), kwargs = {})
    %mul_84 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_207, %unsqueeze_135), kwargs = {})
    %add_118 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_84, %unsqueeze_132), kwargs = {})
    %squeeze_76 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_75, 3), kwargs = {})
    %squeeze_77 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_76, 2), kwargs = {})
    %squeeze_78 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_37, 3), kwargs = {})
    %squeeze_79 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_78, 2), kwargs = {})
    %detach_50 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_77,), kwargs = {})
    %detach_51 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_79,), kwargs = {})
    %silu_21 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_118,), kwargs = {})
    %_param_constant240 : [#users=1] = get_attr[target=_param_constant240]
    %_param_constant241 : [#users=1] = get_attr[target=_param_constant241]
    %convolution_19 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_21, %_param_constant240, %_param_constant241, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_119 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_17, %convolution_19), kwargs = {})
    %div_7 : [#users=3] = call_function[target=torch.ops.aten.div](args = (%add_119, 1.0), kwargs = {})
    %view_208 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_7, [2, 32, 40, 64]), kwargs = {})
    %var_mean_38 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_208, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_76 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_38, 0), kwargs = {})
    %getitem_77 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_38, 1), kwargs = {})
    %add_120 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_76, 1e-05), kwargs = {})
    %rsqrt_38 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_120,), kwargs = {})
    %sub_38 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_208, %getitem_77), kwargs = {})
    %mul_85 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_38, %rsqrt_38), kwargs = {})
    %view_209 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_85, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant242 : [#users=1] = get_attr[target=_param_constant242]
    %unsqueeze_136 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant242, 0), kwargs = {})
    %unsqueeze_137 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_136, 2), kwargs = {})
    %unsqueeze_138 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_137, 3), kwargs = {})
    %_param_constant243 : [#users=1] = get_attr[target=_param_constant243]
    %unsqueeze_139 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant243, 0), kwargs = {})
    %unsqueeze_140 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_139, 2), kwargs = {})
    %unsqueeze_141 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_140, 3), kwargs = {})
    %mul_86 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_209, %unsqueeze_141), kwargs = {})
    %add_121 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_86, %unsqueeze_138), kwargs = {})
    %squeeze_80 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_77, 3), kwargs = {})
    %squeeze_81 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_80, 2), kwargs = {})
    %squeeze_82 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_38, 3), kwargs = {})
    %squeeze_83 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_82, 2), kwargs = {})
    %detach_52 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_81,), kwargs = {})
    %detach_53 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_83,), kwargs = {})
    %silu_22 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_121,), kwargs = {})
    %_param_constant244 : [#users=1] = get_attr[target=_param_constant244]
    %_param_constant245 : [#users=1] = get_attr[target=_param_constant245]
    %convolution_20 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_22, %_param_constant244, %_param_constant245, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_23 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant246 : [#users=1] = get_attr[target=_param_constant246]
    %t_81 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant246,), kwargs = {})
    %_param_constant247 : [#users=1] = get_attr[target=_param_constant247]
    %addmm_39 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant247, %silu_23, %t_81), kwargs = {})
    %slice_33 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_39, 0, 0, 9223372036854775807), kwargs = {})
    %slice_34 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_33, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_142 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_34, 2), kwargs = {})
    %unsqueeze_143 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_142, 3), kwargs = {})
    %add_122 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_20, %unsqueeze_143), kwargs = {})
    %view_210 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_122, [2, 32, 40, 64]), kwargs = {})
    %var_mean_39 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_210, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_78 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_39, 0), kwargs = {})
    %getitem_79 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_39, 1), kwargs = {})
    %add_123 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_78, 1e-05), kwargs = {})
    %rsqrt_39 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_123,), kwargs = {})
    %sub_39 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_210, %getitem_79), kwargs = {})
    %mul_87 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_39, %rsqrt_39), kwargs = {})
    %view_211 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_87, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant248 : [#users=1] = get_attr[target=_param_constant248]
    %unsqueeze_144 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant248, 0), kwargs = {})
    %unsqueeze_145 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_144, 2), kwargs = {})
    %unsqueeze_146 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_145, 3), kwargs = {})
    %_param_constant249 : [#users=1] = get_attr[target=_param_constant249]
    %unsqueeze_147 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant249, 0), kwargs = {})
    %unsqueeze_148 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_147, 2), kwargs = {})
    %unsqueeze_149 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_148, 3), kwargs = {})
    %mul_88 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_211, %unsqueeze_149), kwargs = {})
    %add_124 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_88, %unsqueeze_146), kwargs = {})
    %squeeze_84 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_79, 3), kwargs = {})
    %squeeze_85 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_84, 2), kwargs = {})
    %squeeze_86 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_39, 3), kwargs = {})
    %squeeze_87 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_86, 2), kwargs = {})
    %detach_54 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_85,), kwargs = {})
    %detach_55 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_87,), kwargs = {})
    %silu_24 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_124,), kwargs = {})
    %_param_constant250 : [#users=1] = get_attr[target=_param_constant250]
    %_param_constant251 : [#users=1] = get_attr[target=_param_constant251]
    %convolution_21 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_24, %_param_constant250, %_param_constant251, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_125 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%div_7, %convolution_21), kwargs = {})
    %div_8 : [#users=3] = call_function[target=torch.ops.aten.div](args = (%add_125, 1.0), kwargs = {})
    %view_212 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_8, [2, 32, 40, 64]), kwargs = {})
    %var_mean_40 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_212, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_80 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_40, 0), kwargs = {})
    %getitem_81 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_40, 1), kwargs = {})
    %add_126 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_80, 1e-05), kwargs = {})
    %rsqrt_40 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_126,), kwargs = {})
    %sub_40 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_212, %getitem_81), kwargs = {})
    %mul_89 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_40, %rsqrt_40), kwargs = {})
    %view_213 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_89, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant252 : [#users=1] = get_attr[target=_param_constant252]
    %unsqueeze_150 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant252, 0), kwargs = {})
    %unsqueeze_151 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_150, 2), kwargs = {})
    %unsqueeze_152 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_151, 3), kwargs = {})
    %_param_constant253 : [#users=1] = get_attr[target=_param_constant253]
    %unsqueeze_153 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant253, 0), kwargs = {})
    %unsqueeze_154 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_153, 2), kwargs = {})
    %unsqueeze_155 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_154, 3), kwargs = {})
    %mul_90 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_213, %unsqueeze_155), kwargs = {})
    %add_127 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_90, %unsqueeze_152), kwargs = {})
    %squeeze_88 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_81, 3), kwargs = {})
    %squeeze_89 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_88, 2), kwargs = {})
    %squeeze_90 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_40, 3), kwargs = {})
    %squeeze_91 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_90, 2), kwargs = {})
    %detach_56 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_89,), kwargs = {})
    %detach_57 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_91,), kwargs = {})
    %silu_25 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_127,), kwargs = {})
    %_param_constant254 : [#users=1] = get_attr[target=_param_constant254]
    %_param_constant255 : [#users=1] = get_attr[target=_param_constant255]
    %convolution_22 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_25, %_param_constant254, %_param_constant255, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_26 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant256 : [#users=1] = get_attr[target=_param_constant256]
    %t_82 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant256,), kwargs = {})
    %_param_constant257 : [#users=1] = get_attr[target=_param_constant257]
    %addmm_40 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant257, %silu_26, %t_82), kwargs = {})
    %slice_35 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_40, 0, 0, 9223372036854775807), kwargs = {})
    %slice_36 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_35, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_156 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_36, 2), kwargs = {})
    %unsqueeze_157 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_156, 3), kwargs = {})
    %add_128 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_22, %unsqueeze_157), kwargs = {})
    %view_214 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_128, [2, 32, 40, 64]), kwargs = {})
    %var_mean_41 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_214, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_82 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_41, 0), kwargs = {})
    %getitem_83 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_41, 1), kwargs = {})
    %add_129 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_82, 1e-05), kwargs = {})
    %rsqrt_41 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_129,), kwargs = {})
    %sub_41 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_214, %getitem_83), kwargs = {})
    %mul_91 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_41, %rsqrt_41), kwargs = {})
    %view_215 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_91, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant258 : [#users=1] = get_attr[target=_param_constant258]
    %unsqueeze_158 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant258, 0), kwargs = {})
    %unsqueeze_159 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_158, 2), kwargs = {})
    %unsqueeze_160 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_159, 3), kwargs = {})
    %_param_constant259 : [#users=1] = get_attr[target=_param_constant259]
    %unsqueeze_161 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant259, 0), kwargs = {})
    %unsqueeze_162 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_161, 2), kwargs = {})
    %unsqueeze_163 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_162, 3), kwargs = {})
    %mul_92 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_215, %unsqueeze_163), kwargs = {})
    %add_130 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_92, %unsqueeze_160), kwargs = {})
    %squeeze_92 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_83, 3), kwargs = {})
    %squeeze_93 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_92, 2), kwargs = {})
    %squeeze_94 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_41, 3), kwargs = {})
    %squeeze_95 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_94, 2), kwargs = {})
    %detach_58 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_93,), kwargs = {})
    %detach_59 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_95,), kwargs = {})
    %silu_27 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_130,), kwargs = {})
    %_param_constant260 : [#users=1] = get_attr[target=_param_constant260]
    %_param_constant261 : [#users=1] = get_attr[target=_param_constant261]
    %convolution_23 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_27, %_param_constant260, %_param_constant261, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_131 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%div_8, %convolution_23), kwargs = {})
    %div_9 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_131, 1), kwargs = {})
    %view_216 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_9, [2, 32, 40, 64]), kwargs = {})
    %var_mean_42 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_216, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_84 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_42, 0), kwargs = {})
    %getitem_85 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_42, 1), kwargs = {})
    %add_132 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_84, 1e-06), kwargs = {})
    %rsqrt_42 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_132,), kwargs = {})
    %sub_42 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_216, %getitem_85), kwargs = {})
    %mul_93 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_42, %rsqrt_42), kwargs = {})
    %view_217 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_93, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant262 : [#users=1] = get_attr[target=_param_constant262]
    %unsqueeze_164 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant262, 0), kwargs = {})
    %unsqueeze_165 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_164, 2), kwargs = {})
    %unsqueeze_166 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_165, 3), kwargs = {})
    %_param_constant263 : [#users=1] = get_attr[target=_param_constant263]
    %unsqueeze_167 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant263, 0), kwargs = {})
    %unsqueeze_168 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_167, 2), kwargs = {})
    %unsqueeze_169 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_168, 3), kwargs = {})
    %mul_94 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_217, %unsqueeze_169), kwargs = {})
    %add_133 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_94, %unsqueeze_166), kwargs = {})
    %squeeze_96 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_85, 3), kwargs = {})
    %squeeze_97 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_96, 2), kwargs = {})
    %squeeze_98 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_42, 3), kwargs = {})
    %squeeze_99 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_98, 2), kwargs = {})
    %detach_60 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_97,), kwargs = {})
    %detach_61 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_99,), kwargs = {})
    %permute_60 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_133, [0, 2, 3, 1]), kwargs = {})
    %view_218 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_60, [2, 64, 1280]), kwargs = {})
    %_param_constant264 : [#users=1] = get_attr[target=_param_constant264]
    %t_83 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant264,), kwargs = {})
    %expand_13 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_218, [2, 64, 1280]), kwargs = {})
    %view_219 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_13, [2, 64, 1280]), kwargs = {})
    %expand_14 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_83, [2, 1280, 1280]), kwargs = {})
    %view_220 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_14, [2, 1280, 1280]), kwargs = {})
    %bmm_18 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_219, %view_220), kwargs = {})
    %_unsafe_view_90 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_18, [2, 64, 1280]), kwargs = {})
    %_param_constant265 : [#users=1] = get_attr[target=_param_constant265]
    %add_134 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_90, %_param_constant265), kwargs = {})
    %var_mean_43 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_134, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_86 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_43, 0), kwargs = {})
    %getitem_87 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_43, 1), kwargs = {})
    %add_135 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_86, 1e-05), kwargs = {})
    %rsqrt_43 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_135,), kwargs = {})
    %sub_43 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_134, %getitem_87), kwargs = {})
    %mul_95 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_43, %rsqrt_43), kwargs = {})
    %_param_constant266 : [#users=1] = get_attr[target=_param_constant266]
    %mul_96 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_95, %_param_constant266), kwargs = {})
    %_param_constant267 : [#users=1] = get_attr[target=_param_constant267]
    %add_136 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_96, %_param_constant267), kwargs = {})
    %_param_constant268 : [#users=1] = get_attr[target=_param_constant268]
    %t_84 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant268,), kwargs = {})
    %view_221 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_136, [128, 1280]), kwargs = {})
    %mm_36 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_221, %t_84), kwargs = {})
    %_unsafe_view_91 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_36, [2, 64, 1280]), kwargs = {})
    %view_222 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_91, [2, 64, 20, 64]), kwargs = {})
    %permute_61 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_222, [0, 2, 1, 3]), kwargs = {})
    %clone_54 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_61,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_92 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_54, [40, 64, 64]), kwargs = {})
    %_param_constant269 : [#users=1] = get_attr[target=_param_constant269]
    %t_85 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant269,), kwargs = {})
    %view_223 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_136, [128, 1280]), kwargs = {})
    %mm_37 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_223, %t_85), kwargs = {})
    %_unsafe_view_93 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_37, [2, 64, 1280]), kwargs = {})
    %_param_constant270 : [#users=1] = get_attr[target=_param_constant270]
    %t_86 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant270,), kwargs = {})
    %view_224 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_136, [128, 1280]), kwargs = {})
    %mm_38 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_224, %t_86), kwargs = {})
    %_unsafe_view_94 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_38, [2, 64, 1280]), kwargs = {})
    %view_225 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_93, [2, 64, 20, 64]), kwargs = {})
    %permute_62 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_225, [0, 2, 1, 3]), kwargs = {})
    %clone_55 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_62,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_95 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_55, [40, 64, 64]), kwargs = {})
    %view_226 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_94, [2, 64, 20, 64]), kwargs = {})
    %permute_63 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_226, [0, 2, 1, 3]), kwargs = {})
    %clone_56 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_63,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_96 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_56, [40, 64, 64]), kwargs = {})
    %empty_12 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 64, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_12 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_95, -1, -2), kwargs = {})
    %baddbmm_12 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_12, %_unsafe_view_92, %transpose_12), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_12 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_12, -1, False), kwargs = {})
    %detach_62 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_12,), kwargs = {})
    %bmm_19 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_12, %_unsafe_view_96), kwargs = {})
    %view_227 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_19, [2, 20, 64, 64]), kwargs = {})
    %permute_64 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_227, [0, 2, 1, 3]), kwargs = {})
    %clone_57 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_64,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_97 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_57, [2, 64, 1280]), kwargs = {})
    %_param_constant271 : [#users=1] = get_attr[target=_param_constant271]
    %t_87 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant271,), kwargs = {})
    %view_228 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_97, [128, 1280]), kwargs = {})
    %_param_constant272 : [#users=1] = get_attr[target=_param_constant272]
    %addmm_41 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant272, %view_228, %t_87), kwargs = {})
    %view_229 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_41, [2, 64, 1280]), kwargs = {})
    %add_137 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_229, %add_134), kwargs = {})
    %var_mean_44 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_137, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_88 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_44, 0), kwargs = {})
    %getitem_89 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_44, 1), kwargs = {})
    %add_138 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_88, 1e-05), kwargs = {})
    %rsqrt_44 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_138,), kwargs = {})
    %sub_44 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_137, %getitem_89), kwargs = {})
    %mul_97 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_44, %rsqrt_44), kwargs = {})
    %_param_constant273 : [#users=1] = get_attr[target=_param_constant273]
    %mul_98 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_97, %_param_constant273), kwargs = {})
    %_param_constant274 : [#users=1] = get_attr[target=_param_constant274]
    %add_139 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_98, %_param_constant274), kwargs = {})
    %_param_constant275 : [#users=1] = get_attr[target=_param_constant275]
    %t_88 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant275,), kwargs = {})
    %view_230 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_139, [128, 1280]), kwargs = {})
    %mm_39 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_230, %t_88), kwargs = {})
    %_unsafe_view_98 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_39, [2, 64, 1280]), kwargs = {})
    %view_231 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_98, [2, 64, 20, 64]), kwargs = {})
    %permute_65 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_231, [0, 2, 1, 3]), kwargs = {})
    %clone_58 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_65,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_99 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_58, [40, 64, 64]), kwargs = {})
    %_param_constant276 : [#users=1] = get_attr[target=_param_constant276]
    %t_89 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant276,), kwargs = {})
    %view_232 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_40 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_232, %t_89), kwargs = {})
    %_unsafe_view_100 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_40, [2, 64, 1280]), kwargs = {})
    %_param_constant277 : [#users=1] = get_attr[target=_param_constant277]
    %t_90 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant277,), kwargs = {})
    %view_233 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_41 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_233, %t_90), kwargs = {})
    %_unsafe_view_101 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_41, [2, 64, 1280]), kwargs = {})
    %view_234 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_100, [2, 64, 20, 64]), kwargs = {})
    %permute_66 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_234, [0, 2, 1, 3]), kwargs = {})
    %clone_59 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_66,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_102 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_59, [40, 64, 64]), kwargs = {})
    %view_235 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_101, [2, 64, 20, 64]), kwargs = {})
    %permute_67 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_235, [0, 2, 1, 3]), kwargs = {})
    %clone_60 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_67,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_103 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_60, [40, 64, 64]), kwargs = {})
    %empty_13 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 64, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_13 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_102, -1, -2), kwargs = {})
    %baddbmm_13 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_13, %_unsafe_view_99, %transpose_13), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_13 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_13, -1, False), kwargs = {})
    %detach_63 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_13,), kwargs = {})
    %bmm_20 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_13, %_unsafe_view_103), kwargs = {})
    %view_236 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_20, [2, 20, 64, 64]), kwargs = {})
    %permute_68 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_236, [0, 2, 1, 3]), kwargs = {})
    %clone_61 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_68,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_104 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_61, [2, 64, 1280]), kwargs = {})
    %_param_constant278 : [#users=1] = get_attr[target=_param_constant278]
    %t_91 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant278,), kwargs = {})
    %view_237 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_104, [128, 1280]), kwargs = {})
    %_param_constant279 : [#users=1] = get_attr[target=_param_constant279]
    %addmm_42 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant279, %view_237, %t_91), kwargs = {})
    %view_238 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_42, [2, 64, 1280]), kwargs = {})
    %add_140 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_238, %add_137), kwargs = {})
    %var_mean_45 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_140, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_90 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_45, 0), kwargs = {})
    %getitem_91 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_45, 1), kwargs = {})
    %add_141 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_90, 1e-05), kwargs = {})
    %rsqrt_45 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_141,), kwargs = {})
    %sub_45 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_140, %getitem_91), kwargs = {})
    %mul_99 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_45, %rsqrt_45), kwargs = {})
    %_param_constant280 : [#users=1] = get_attr[target=_param_constant280]
    %mul_100 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_99, %_param_constant280), kwargs = {})
    %_param_constant281 : [#users=1] = get_attr[target=_param_constant281]
    %add_142 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_100, %_param_constant281), kwargs = {})
    %_param_constant282 : [#users=1] = get_attr[target=_param_constant282]
    %t_92 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant282,), kwargs = {})
    %view_239 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_142, [128, 1280]), kwargs = {})
    %_param_constant283 : [#users=1] = get_attr[target=_param_constant283]
    %addmm_43 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant283, %view_239, %t_92), kwargs = {})
    %view_240 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_43, [2, 64, 10240]), kwargs = {})
    %slice_37 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_240, -1, 0, 5120), kwargs = {})
    %slice_38 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_240, -1, 5120, 10240), kwargs = {})
    %gelu_6 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_38,), kwargs = {})
    %mul_101 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_37, %gelu_6), kwargs = {})
    %_param_constant284 : [#users=1] = get_attr[target=_param_constant284]
    %t_93 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant284,), kwargs = {})
    %view_241 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_101, [128, 5120]), kwargs = {})
    %_param_constant285 : [#users=1] = get_attr[target=_param_constant285]
    %addmm_44 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant285, %view_241, %t_93), kwargs = {})
    %view_242 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_44, [2, 64, 1280]), kwargs = {})
    %add_143 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_242, %add_140), kwargs = {})
    %_param_constant286 : [#users=1] = get_attr[target=_param_constant286]
    %t_94 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant286,), kwargs = {})
    %view_243 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_143, [128, 1280]), kwargs = {})
    %_param_constant287 : [#users=1] = get_attr[target=_param_constant287]
    %addmm_45 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant287, %view_243, %t_94), kwargs = {})
    %view_244 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_45, [2, 64, 1280]), kwargs = {})
    %view_245 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_244, [2, 8, 8, 1280]), kwargs = {})
    %permute_69 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_245, [0, 3, 1, 2]), kwargs = {})
    %clone_62 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_69,), kwargs = {memory_format: torch.contiguous_format})
    %add_144 : [#users=2] = call_function[target=torch.ops.aten.add](args = (%clone_62, %div_9), kwargs = {})
    %view_246 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_144, [2, 32, 40, 64]), kwargs = {})
    %var_mean_46 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_246, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_92 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_46, 0), kwargs = {})
    %getitem_93 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_46, 1), kwargs = {})
    %add_145 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_92, 1e-05), kwargs = {})
    %rsqrt_46 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_145,), kwargs = {})
    %sub_46 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_246, %getitem_93), kwargs = {})
    %mul_102 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_46, %rsqrt_46), kwargs = {})
    %view_247 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_102, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant288 : [#users=1] = get_attr[target=_param_constant288]
    %unsqueeze_170 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant288, 0), kwargs = {})
    %unsqueeze_171 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_170, 2), kwargs = {})
    %unsqueeze_172 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_171, 3), kwargs = {})
    %_param_constant289 : [#users=1] = get_attr[target=_param_constant289]
    %unsqueeze_173 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant289, 0), kwargs = {})
    %unsqueeze_174 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_173, 2), kwargs = {})
    %unsqueeze_175 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_174, 3), kwargs = {})
    %mul_103 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_247, %unsqueeze_175), kwargs = {})
    %add_146 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_103, %unsqueeze_172), kwargs = {})
    %squeeze_100 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_93, 3), kwargs = {})
    %squeeze_101 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_100, 2), kwargs = {})
    %squeeze_102 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_46, 3), kwargs = {})
    %squeeze_103 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_102, 2), kwargs = {})
    %detach_64 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_101,), kwargs = {})
    %detach_65 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_103,), kwargs = {})
    %silu_28 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_146,), kwargs = {})
    %_param_constant290 : [#users=1] = get_attr[target=_param_constant290]
    %_param_constant291 : [#users=1] = get_attr[target=_param_constant291]
    %convolution_24 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_28, %_param_constant290, %_param_constant291, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_29 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant292 : [#users=1] = get_attr[target=_param_constant292]
    %t_95 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant292,), kwargs = {})
    %_param_constant293 : [#users=1] = get_attr[target=_param_constant293]
    %addmm_46 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant293, %silu_29, %t_95), kwargs = {})
    %slice_39 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_46, 0, 0, 9223372036854775807), kwargs = {})
    %slice_40 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_39, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_176 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_40, 2), kwargs = {})
    %unsqueeze_177 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_176, 3), kwargs = {})
    %add_147 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_24, %unsqueeze_177), kwargs = {})
    %view_248 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_147, [2, 32, 40, 64]), kwargs = {})
    %var_mean_47 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_248, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_94 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_47, 0), kwargs = {})
    %getitem_95 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_47, 1), kwargs = {})
    %add_148 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_94, 1e-05), kwargs = {})
    %rsqrt_47 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_148,), kwargs = {})
    %sub_47 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_248, %getitem_95), kwargs = {})
    %mul_104 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_47, %rsqrt_47), kwargs = {})
    %view_249 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_104, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant294 : [#users=1] = get_attr[target=_param_constant294]
    %unsqueeze_178 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant294, 0), kwargs = {})
    %unsqueeze_179 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_178, 2), kwargs = {})
    %unsqueeze_180 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_179, 3), kwargs = {})
    %_param_constant295 : [#users=1] = get_attr[target=_param_constant295]
    %unsqueeze_181 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant295, 0), kwargs = {})
    %unsqueeze_182 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_181, 2), kwargs = {})
    %unsqueeze_183 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_182, 3), kwargs = {})
    %mul_105 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_249, %unsqueeze_183), kwargs = {})
    %add_149 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_105, %unsqueeze_180), kwargs = {})
    %squeeze_104 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_95, 3), kwargs = {})
    %squeeze_105 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_104, 2), kwargs = {})
    %squeeze_106 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_47, 3), kwargs = {})
    %squeeze_107 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_106, 2), kwargs = {})
    %detach_66 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_105,), kwargs = {})
    %detach_67 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_107,), kwargs = {})
    %silu_30 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_149,), kwargs = {})
    %_param_constant296 : [#users=1] = get_attr[target=_param_constant296]
    %_param_constant297 : [#users=1] = get_attr[target=_param_constant297]
    %convolution_25 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_30, %_param_constant296, %_param_constant297, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_150 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%add_144, %convolution_25), kwargs = {})
    %div_10 : [#users=1] = call_function[target=torch.ops.aten.div](args = (%add_150, 1), kwargs = {})
    %cat_3 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%div_10, %div_8], 1), kwargs = {})
    %view_250 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_3, [2, 32, 80, 64]), kwargs = {})
    %var_mean_48 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_250, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_96 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_48, 0), kwargs = {})
    %getitem_97 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_48, 1), kwargs = {})
    %add_151 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_96, 1e-05), kwargs = {})
    %rsqrt_48 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_151,), kwargs = {})
    %sub_48 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_250, %getitem_97), kwargs = {})
    %mul_106 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_48, %rsqrt_48), kwargs = {})
    %view_251 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_106, [2, 2560, 8, 8]), kwargs = {})
    %_param_constant298 : [#users=1] = get_attr[target=_param_constant298]
    %unsqueeze_184 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant298, 0), kwargs = {})
    %unsqueeze_185 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_184, 2), kwargs = {})
    %unsqueeze_186 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_185, 3), kwargs = {})
    %_param_constant299 : [#users=1] = get_attr[target=_param_constant299]
    %unsqueeze_187 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant299, 0), kwargs = {})
    %unsqueeze_188 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_187, 2), kwargs = {})
    %unsqueeze_189 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_188, 3), kwargs = {})
    %mul_107 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_251, %unsqueeze_189), kwargs = {})
    %add_152 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_107, %unsqueeze_186), kwargs = {})
    %squeeze_108 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_97, 3), kwargs = {})
    %squeeze_109 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_108, 2), kwargs = {})
    %squeeze_110 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_48, 3), kwargs = {})
    %squeeze_111 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_110, 2), kwargs = {})
    %detach_68 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_109,), kwargs = {})
    %detach_69 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_111,), kwargs = {})
    %silu_31 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_152,), kwargs = {})
    %_param_constant300 : [#users=1] = get_attr[target=_param_constant300]
    %_param_constant301 : [#users=1] = get_attr[target=_param_constant301]
    %convolution_26 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_31, %_param_constant300, %_param_constant301, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_32 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant302 : [#users=1] = get_attr[target=_param_constant302]
    %t_96 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant302,), kwargs = {})
    %_param_constant303 : [#users=1] = get_attr[target=_param_constant303]
    %addmm_47 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant303, %silu_32, %t_96), kwargs = {})
    %slice_41 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_47, 0, 0, 9223372036854775807), kwargs = {})
    %slice_42 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_41, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_190 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_42, 2), kwargs = {})
    %unsqueeze_191 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_190, 3), kwargs = {})
    %add_153 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_26, %unsqueeze_191), kwargs = {})
    %view_252 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_153, [2, 32, 40, 64]), kwargs = {})
    %var_mean_49 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_252, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_98 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_49, 0), kwargs = {})
    %getitem_99 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_49, 1), kwargs = {})
    %add_154 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_98, 1e-05), kwargs = {})
    %rsqrt_49 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_154,), kwargs = {})
    %sub_49 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_252, %getitem_99), kwargs = {})
    %mul_108 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_49, %rsqrt_49), kwargs = {})
    %view_253 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_108, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant304 : [#users=1] = get_attr[target=_param_constant304]
    %unsqueeze_192 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant304, 0), kwargs = {})
    %unsqueeze_193 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_192, 2), kwargs = {})
    %unsqueeze_194 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_193, 3), kwargs = {})
    %_param_constant305 : [#users=1] = get_attr[target=_param_constant305]
    %unsqueeze_195 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant305, 0), kwargs = {})
    %unsqueeze_196 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_195, 2), kwargs = {})
    %unsqueeze_197 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_196, 3), kwargs = {})
    %mul_109 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_253, %unsqueeze_197), kwargs = {})
    %add_155 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_109, %unsqueeze_194), kwargs = {})
    %squeeze_112 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_99, 3), kwargs = {})
    %squeeze_113 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_112, 2), kwargs = {})
    %squeeze_114 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_49, 3), kwargs = {})
    %squeeze_115 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_114, 2), kwargs = {})
    %detach_70 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_113,), kwargs = {})
    %detach_71 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_115,), kwargs = {})
    %silu_33 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_155,), kwargs = {})
    %_param_constant306 : [#users=1] = get_attr[target=_param_constant306]
    %_param_constant307 : [#users=1] = get_attr[target=_param_constant307]
    %convolution_27 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_33, %_param_constant306, %_param_constant307, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant308 : [#users=1] = get_attr[target=_param_constant308]
    %_param_constant309 : [#users=1] = get_attr[target=_param_constant309]
    %convolution_28 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_3, %_param_constant308, %_param_constant309, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_156 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_28, %convolution_27), kwargs = {})
    %div_11 : [#users=1] = call_function[target=torch.ops.aten.div](args = (%add_156, 1.0), kwargs = {})
    %cat_4 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%div_11, %div_7], 1), kwargs = {})
    %view_254 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_4, [2, 32, 80, 64]), kwargs = {})
    %var_mean_50 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_254, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_100 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_50, 0), kwargs = {})
    %getitem_101 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_50, 1), kwargs = {})
    %add_157 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_100, 1e-05), kwargs = {})
    %rsqrt_50 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_157,), kwargs = {})
    %sub_50 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_254, %getitem_101), kwargs = {})
    %mul_110 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_50, %rsqrt_50), kwargs = {})
    %view_255 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_110, [2, 2560, 8, 8]), kwargs = {})
    %_param_constant310 : [#users=1] = get_attr[target=_param_constant310]
    %unsqueeze_198 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant310, 0), kwargs = {})
    %unsqueeze_199 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_198, 2), kwargs = {})
    %unsqueeze_200 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_199, 3), kwargs = {})
    %_param_constant311 : [#users=1] = get_attr[target=_param_constant311]
    %unsqueeze_201 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant311, 0), kwargs = {})
    %unsqueeze_202 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_201, 2), kwargs = {})
    %unsqueeze_203 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_202, 3), kwargs = {})
    %mul_111 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_255, %unsqueeze_203), kwargs = {})
    %add_158 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_111, %unsqueeze_200), kwargs = {})
    %squeeze_116 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_101, 3), kwargs = {})
    %squeeze_117 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_116, 2), kwargs = {})
    %squeeze_118 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_50, 3), kwargs = {})
    %squeeze_119 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_118, 2), kwargs = {})
    %detach_72 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_117,), kwargs = {})
    %detach_73 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_119,), kwargs = {})
    %silu_34 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_158,), kwargs = {})
    %_param_constant312 : [#users=1] = get_attr[target=_param_constant312]
    %_param_constant313 : [#users=1] = get_attr[target=_param_constant313]
    %convolution_29 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_34, %_param_constant312, %_param_constant313, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_35 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant314 : [#users=1] = get_attr[target=_param_constant314]
    %t_97 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant314,), kwargs = {})
    %_param_constant315 : [#users=1] = get_attr[target=_param_constant315]
    %addmm_48 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant315, %silu_35, %t_97), kwargs = {})
    %slice_43 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_48, 0, 0, 9223372036854775807), kwargs = {})
    %slice_44 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_43, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_204 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_44, 2), kwargs = {})
    %unsqueeze_205 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_204, 3), kwargs = {})
    %add_159 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_29, %unsqueeze_205), kwargs = {})
    %view_256 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_159, [2, 32, 40, 64]), kwargs = {})
    %var_mean_51 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_256, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_102 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_51, 0), kwargs = {})
    %getitem_103 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_51, 1), kwargs = {})
    %add_160 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_102, 1e-05), kwargs = {})
    %rsqrt_51 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_160,), kwargs = {})
    %sub_51 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_256, %getitem_103), kwargs = {})
    %mul_112 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_51, %rsqrt_51), kwargs = {})
    %view_257 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_112, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant316 : [#users=1] = get_attr[target=_param_constant316]
    %unsqueeze_206 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant316, 0), kwargs = {})
    %unsqueeze_207 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_206, 2), kwargs = {})
    %unsqueeze_208 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_207, 3), kwargs = {})
    %_param_constant317 : [#users=1] = get_attr[target=_param_constant317]
    %unsqueeze_209 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant317, 0), kwargs = {})
    %unsqueeze_210 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_209, 2), kwargs = {})
    %unsqueeze_211 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_210, 3), kwargs = {})
    %mul_113 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_257, %unsqueeze_211), kwargs = {})
    %add_161 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_113, %unsqueeze_208), kwargs = {})
    %squeeze_120 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_103, 3), kwargs = {})
    %squeeze_121 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_120, 2), kwargs = {})
    %squeeze_122 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_51, 3), kwargs = {})
    %squeeze_123 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_122, 2), kwargs = {})
    %detach_74 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_121,), kwargs = {})
    %detach_75 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_123,), kwargs = {})
    %silu_36 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_161,), kwargs = {})
    %_param_constant318 : [#users=1] = get_attr[target=_param_constant318]
    %_param_constant319 : [#users=1] = get_attr[target=_param_constant319]
    %convolution_30 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_36, %_param_constant318, %_param_constant319, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant320 : [#users=1] = get_attr[target=_param_constant320]
    %_param_constant321 : [#users=1] = get_attr[target=_param_constant321]
    %convolution_31 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_4, %_param_constant320, %_param_constant321, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_162 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_31, %convolution_30), kwargs = {})
    %div_12 : [#users=1] = call_function[target=torch.ops.aten.div](args = (%add_162, 1.0), kwargs = {})
    %cat_5 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%div_12, %convolution_17], 1), kwargs = {})
    %view_258 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_5, [2, 32, 80, 64]), kwargs = {})
    %var_mean_52 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_258, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_104 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_52, 0), kwargs = {})
    %getitem_105 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_52, 1), kwargs = {})
    %add_163 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_104, 1e-05), kwargs = {})
    %rsqrt_52 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_163,), kwargs = {})
    %sub_52 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_258, %getitem_105), kwargs = {})
    %mul_114 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_52, %rsqrt_52), kwargs = {})
    %view_259 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_114, [2, 2560, 8, 8]), kwargs = {})
    %_param_constant322 : [#users=1] = get_attr[target=_param_constant322]
    %unsqueeze_212 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant322, 0), kwargs = {})
    %unsqueeze_213 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_212, 2), kwargs = {})
    %unsqueeze_214 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_213, 3), kwargs = {})
    %_param_constant323 : [#users=1] = get_attr[target=_param_constant323]
    %unsqueeze_215 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant323, 0), kwargs = {})
    %unsqueeze_216 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_215, 2), kwargs = {})
    %unsqueeze_217 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_216, 3), kwargs = {})
    %mul_115 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_259, %unsqueeze_217), kwargs = {})
    %add_164 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_115, %unsqueeze_214), kwargs = {})
    %squeeze_124 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_105, 3), kwargs = {})
    %squeeze_125 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_124, 2), kwargs = {})
    %squeeze_126 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_52, 3), kwargs = {})
    %squeeze_127 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_126, 2), kwargs = {})
    %detach_76 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_125,), kwargs = {})
    %detach_77 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_127,), kwargs = {})
    %silu_37 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_164,), kwargs = {})
    %_param_constant324 : [#users=1] = get_attr[target=_param_constant324]
    %_param_constant325 : [#users=1] = get_attr[target=_param_constant325]
    %convolution_32 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_37, %_param_constant324, %_param_constant325, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_38 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant326 : [#users=1] = get_attr[target=_param_constant326]
    %t_98 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant326,), kwargs = {})
    %_param_constant327 : [#users=1] = get_attr[target=_param_constant327]
    %addmm_49 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant327, %silu_38, %t_98), kwargs = {})
    %slice_45 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_49, 0, 0, 9223372036854775807), kwargs = {})
    %slice_46 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_45, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_218 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_46, 2), kwargs = {})
    %unsqueeze_219 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_218, 3), kwargs = {})
    %add_165 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_32, %unsqueeze_219), kwargs = {})
    %view_260 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_165, [2, 32, 40, 64]), kwargs = {})
    %var_mean_53 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_260, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_106 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_53, 0), kwargs = {})
    %getitem_107 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_53, 1), kwargs = {})
    %add_166 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_106, 1e-05), kwargs = {})
    %rsqrt_53 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_166,), kwargs = {})
    %sub_53 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_260, %getitem_107), kwargs = {})
    %mul_116 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_53, %rsqrt_53), kwargs = {})
    %view_261 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_116, [2, 1280, 8, 8]), kwargs = {})
    %_param_constant328 : [#users=1] = get_attr[target=_param_constant328]
    %unsqueeze_220 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant328, 0), kwargs = {})
    %unsqueeze_221 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_220, 2), kwargs = {})
    %unsqueeze_222 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_221, 3), kwargs = {})
    %_param_constant329 : [#users=1] = get_attr[target=_param_constant329]
    %unsqueeze_223 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant329, 0), kwargs = {})
    %unsqueeze_224 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_223, 2), kwargs = {})
    %unsqueeze_225 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_224, 3), kwargs = {})
    %mul_117 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_261, %unsqueeze_225), kwargs = {})
    %add_167 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_117, %unsqueeze_222), kwargs = {})
    %squeeze_128 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_107, 3), kwargs = {})
    %squeeze_129 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_128, 2), kwargs = {})
    %squeeze_130 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_53, 3), kwargs = {})
    %squeeze_131 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_130, 2), kwargs = {})
    %detach_78 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_129,), kwargs = {})
    %detach_79 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_131,), kwargs = {})
    %silu_39 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_167,), kwargs = {})
    %_param_constant330 : [#users=1] = get_attr[target=_param_constant330]
    %_param_constant331 : [#users=1] = get_attr[target=_param_constant331]
    %convolution_33 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_39, %_param_constant330, %_param_constant331, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant332 : [#users=1] = get_attr[target=_param_constant332]
    %_param_constant333 : [#users=1] = get_attr[target=_param_constant333]
    %convolution_34 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_5, %_param_constant332, %_param_constant333, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_168 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_34, %convolution_33), kwargs = {})
    %div_13 : [#users=1] = call_function[target=torch.ops.aten.div](args = (%add_168, 1.0), kwargs = {})
    %upsample_nearest2d : [#users=1] = call_function[target=torch.ops.aten.upsample_nearest2d](args = (%div_13, [16, 16], 2.0, 2.0), kwargs = {})
    %_param_constant334 : [#users=1] = get_attr[target=_param_constant334]
    %_param_constant335 : [#users=1] = get_attr[target=_param_constant335]
    %convolution_35 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%upsample_nearest2d, %_param_constant334, %_param_constant335, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %cat_6 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%convolution_35, %add_113], 1), kwargs = {})
    %view_262 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_6, [2, 32, 80, 256]), kwargs = {})
    %var_mean_54 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_262, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_108 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_54, 0), kwargs = {})
    %getitem_109 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_54, 1), kwargs = {})
    %add_169 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_108, 1e-05), kwargs = {})
    %rsqrt_54 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_169,), kwargs = {})
    %sub_54 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_262, %getitem_109), kwargs = {})
    %mul_118 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_54, %rsqrt_54), kwargs = {})
    %view_263 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_118, [2, 2560, 16, 16]), kwargs = {})
    %_param_constant336 : [#users=1] = get_attr[target=_param_constant336]
    %unsqueeze_226 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant336, 0), kwargs = {})
    %unsqueeze_227 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_226, 2), kwargs = {})
    %unsqueeze_228 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_227, 3), kwargs = {})
    %_param_constant337 : [#users=1] = get_attr[target=_param_constant337]
    %unsqueeze_229 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant337, 0), kwargs = {})
    %unsqueeze_230 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_229, 2), kwargs = {})
    %unsqueeze_231 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_230, 3), kwargs = {})
    %mul_119 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_263, %unsqueeze_231), kwargs = {})
    %add_170 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_119, %unsqueeze_228), kwargs = {})
    %squeeze_132 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_109, 3), kwargs = {})
    %squeeze_133 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_132, 2), kwargs = {})
    %squeeze_134 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_54, 3), kwargs = {})
    %squeeze_135 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_134, 2), kwargs = {})
    %detach_80 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_133,), kwargs = {})
    %detach_81 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_135,), kwargs = {})
    %silu_40 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_170,), kwargs = {})
    %_param_constant338 : [#users=1] = get_attr[target=_param_constant338]
    %_param_constant339 : [#users=1] = get_attr[target=_param_constant339]
    %convolution_36 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_40, %_param_constant338, %_param_constant339, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_41 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant340 : [#users=1] = get_attr[target=_param_constant340]
    %t_99 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant340,), kwargs = {})
    %_param_constant341 : [#users=1] = get_attr[target=_param_constant341]
    %addmm_50 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant341, %silu_41, %t_99), kwargs = {})
    %slice_47 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_50, 0, 0, 9223372036854775807), kwargs = {})
    %slice_48 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_47, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_232 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_48, 2), kwargs = {})
    %unsqueeze_233 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_232, 3), kwargs = {})
    %add_171 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_36, %unsqueeze_233), kwargs = {})
    %view_264 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_171, [2, 32, 40, 256]), kwargs = {})
    %var_mean_55 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_264, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_110 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_55, 0), kwargs = {})
    %getitem_111 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_55, 1), kwargs = {})
    %add_172 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_110, 1e-05), kwargs = {})
    %rsqrt_55 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_172,), kwargs = {})
    %sub_55 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_264, %getitem_111), kwargs = {})
    %mul_120 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_55, %rsqrt_55), kwargs = {})
    %view_265 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_120, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant342 : [#users=1] = get_attr[target=_param_constant342]
    %unsqueeze_234 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant342, 0), kwargs = {})
    %unsqueeze_235 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_234, 2), kwargs = {})
    %unsqueeze_236 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_235, 3), kwargs = {})
    %_param_constant343 : [#users=1] = get_attr[target=_param_constant343]
    %unsqueeze_237 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant343, 0), kwargs = {})
    %unsqueeze_238 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_237, 2), kwargs = {})
    %unsqueeze_239 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_238, 3), kwargs = {})
    %mul_121 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_265, %unsqueeze_239), kwargs = {})
    %add_173 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_121, %unsqueeze_236), kwargs = {})
    %squeeze_136 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_111, 3), kwargs = {})
    %squeeze_137 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_136, 2), kwargs = {})
    %squeeze_138 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_55, 3), kwargs = {})
    %squeeze_139 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_138, 2), kwargs = {})
    %detach_82 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_137,), kwargs = {})
    %detach_83 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_139,), kwargs = {})
    %silu_42 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_173,), kwargs = {})
    %_param_constant344 : [#users=1] = get_attr[target=_param_constant344]
    %_param_constant345 : [#users=1] = get_attr[target=_param_constant345]
    %convolution_37 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_42, %_param_constant344, %_param_constant345, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant346 : [#users=1] = get_attr[target=_param_constant346]
    %_param_constant347 : [#users=1] = get_attr[target=_param_constant347]
    %convolution_38 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_6, %_param_constant346, %_param_constant347, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_174 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_38, %convolution_37), kwargs = {})
    %div_14 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_174, 1.0), kwargs = {})
    %view_266 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_14, [2, 32, 40, 256]), kwargs = {})
    %var_mean_56 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_266, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_112 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_56, 0), kwargs = {})
    %getitem_113 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_56, 1), kwargs = {})
    %add_175 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_112, 1e-06), kwargs = {})
    %rsqrt_56 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_175,), kwargs = {})
    %sub_56 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_266, %getitem_113), kwargs = {})
    %mul_122 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_56, %rsqrt_56), kwargs = {})
    %view_267 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_122, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant348 : [#users=1] = get_attr[target=_param_constant348]
    %unsqueeze_240 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant348, 0), kwargs = {})
    %unsqueeze_241 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_240, 2), kwargs = {})
    %unsqueeze_242 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_241, 3), kwargs = {})
    %_param_constant349 : [#users=1] = get_attr[target=_param_constant349]
    %unsqueeze_243 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant349, 0), kwargs = {})
    %unsqueeze_244 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_243, 2), kwargs = {})
    %unsqueeze_245 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_244, 3), kwargs = {})
    %mul_123 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_267, %unsqueeze_245), kwargs = {})
    %add_176 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_123, %unsqueeze_242), kwargs = {})
    %squeeze_140 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_113, 3), kwargs = {})
    %squeeze_141 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_140, 2), kwargs = {})
    %squeeze_142 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_56, 3), kwargs = {})
    %squeeze_143 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_142, 2), kwargs = {})
    %detach_84 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_141,), kwargs = {})
    %detach_85 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_143,), kwargs = {})
    %permute_70 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_176, [0, 2, 3, 1]), kwargs = {})
    %view_268 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_70, [2, 256, 1280]), kwargs = {})
    %_param_constant350 : [#users=1] = get_attr[target=_param_constant350]
    %t_100 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant350,), kwargs = {})
    %expand_15 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_268, [2, 256, 1280]), kwargs = {})
    %view_269 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_15, [2, 256, 1280]), kwargs = {})
    %expand_16 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_100, [2, 1280, 1280]), kwargs = {})
    %view_270 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_16, [2, 1280, 1280]), kwargs = {})
    %bmm_21 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_269, %view_270), kwargs = {})
    %_unsafe_view_105 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_21, [2, 256, 1280]), kwargs = {})
    %_param_constant351 : [#users=1] = get_attr[target=_param_constant351]
    %add_177 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_105, %_param_constant351), kwargs = {})
    %var_mean_57 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_177, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_114 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_57, 0), kwargs = {})
    %getitem_115 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_57, 1), kwargs = {})
    %add_178 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_114, 1e-05), kwargs = {})
    %rsqrt_57 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_178,), kwargs = {})
    %sub_57 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_177, %getitem_115), kwargs = {})
    %mul_124 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_57, %rsqrt_57), kwargs = {})
    %_param_constant352 : [#users=1] = get_attr[target=_param_constant352]
    %mul_125 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_124, %_param_constant352), kwargs = {})
    %_param_constant353 : [#users=1] = get_attr[target=_param_constant353]
    %add_179 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_125, %_param_constant353), kwargs = {})
    %_param_constant354 : [#users=1] = get_attr[target=_param_constant354]
    %t_101 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant354,), kwargs = {})
    %view_271 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_179, [512, 1280]), kwargs = {})
    %mm_42 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_271, %t_101), kwargs = {})
    %_unsafe_view_106 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_42, [2, 256, 1280]), kwargs = {})
    %view_272 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_106, [2, 256, 20, 64]), kwargs = {})
    %permute_71 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_272, [0, 2, 1, 3]), kwargs = {})
    %clone_63 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_71,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_107 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_63, [40, 256, 64]), kwargs = {})
    %_param_constant355 : [#users=1] = get_attr[target=_param_constant355]
    %t_102 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant355,), kwargs = {})
    %view_273 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_179, [512, 1280]), kwargs = {})
    %mm_43 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_273, %t_102), kwargs = {})
    %_unsafe_view_108 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_43, [2, 256, 1280]), kwargs = {})
    %_param_constant356 : [#users=1] = get_attr[target=_param_constant356]
    %t_103 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant356,), kwargs = {})
    %view_274 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_179, [512, 1280]), kwargs = {})
    %mm_44 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_274, %t_103), kwargs = {})
    %_unsafe_view_109 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_44, [2, 256, 1280]), kwargs = {})
    %view_275 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_108, [2, 256, 20, 64]), kwargs = {})
    %permute_72 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_275, [0, 2, 1, 3]), kwargs = {})
    %clone_64 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_72,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_110 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_64, [40, 256, 64]), kwargs = {})
    %view_276 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_109, [2, 256, 20, 64]), kwargs = {})
    %permute_73 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_276, [0, 2, 1, 3]), kwargs = {})
    %clone_65 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_73,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_111 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_65, [40, 256, 64]), kwargs = {})
    %empty_14 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 256, 256],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_14 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_110, -1, -2), kwargs = {})
    %baddbmm_14 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_14, %_unsafe_view_107, %transpose_14), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_14 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_14, -1, False), kwargs = {})
    %detach_86 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_14,), kwargs = {})
    %bmm_22 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_14, %_unsafe_view_111), kwargs = {})
    %view_277 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_22, [2, 20, 256, 64]), kwargs = {})
    %permute_74 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_277, [0, 2, 1, 3]), kwargs = {})
    %clone_66 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_74,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_112 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_66, [2, 256, 1280]), kwargs = {})
    %_param_constant357 : [#users=1] = get_attr[target=_param_constant357]
    %t_104 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant357,), kwargs = {})
    %view_278 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_112, [512, 1280]), kwargs = {})
    %_param_constant358 : [#users=1] = get_attr[target=_param_constant358]
    %addmm_51 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant358, %view_278, %t_104), kwargs = {})
    %view_279 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_51, [2, 256, 1280]), kwargs = {})
    %add_180 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_279, %add_177), kwargs = {})
    %var_mean_58 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_180, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_116 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_58, 0), kwargs = {})
    %getitem_117 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_58, 1), kwargs = {})
    %add_181 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_116, 1e-05), kwargs = {})
    %rsqrt_58 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_181,), kwargs = {})
    %sub_58 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_180, %getitem_117), kwargs = {})
    %mul_126 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_58, %rsqrt_58), kwargs = {})
    %_param_constant359 : [#users=1] = get_attr[target=_param_constant359]
    %mul_127 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_126, %_param_constant359), kwargs = {})
    %_param_constant360 : [#users=1] = get_attr[target=_param_constant360]
    %add_182 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_127, %_param_constant360), kwargs = {})
    %_param_constant361 : [#users=1] = get_attr[target=_param_constant361]
    %t_105 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant361,), kwargs = {})
    %view_280 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_182, [512, 1280]), kwargs = {})
    %mm_45 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_280, %t_105), kwargs = {})
    %_unsafe_view_113 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_45, [2, 256, 1280]), kwargs = {})
    %view_281 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_113, [2, 256, 20, 64]), kwargs = {})
    %permute_75 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_281, [0, 2, 1, 3]), kwargs = {})
    %clone_67 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_75,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_114 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_67, [40, 256, 64]), kwargs = {})
    %_param_constant362 : [#users=1] = get_attr[target=_param_constant362]
    %t_106 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant362,), kwargs = {})
    %view_282 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_46 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_282, %t_106), kwargs = {})
    %_unsafe_view_115 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_46, [2, 64, 1280]), kwargs = {})
    %_param_constant363 : [#users=1] = get_attr[target=_param_constant363]
    %t_107 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant363,), kwargs = {})
    %view_283 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_47 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_283, %t_107), kwargs = {})
    %_unsafe_view_116 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_47, [2, 64, 1280]), kwargs = {})
    %view_284 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_115, [2, 64, 20, 64]), kwargs = {})
    %permute_76 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_284, [0, 2, 1, 3]), kwargs = {})
    %clone_68 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_76,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_117 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_68, [40, 64, 64]), kwargs = {})
    %view_285 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_116, [2, 64, 20, 64]), kwargs = {})
    %permute_77 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_285, [0, 2, 1, 3]), kwargs = {})
    %clone_69 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_77,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_118 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_69, [40, 64, 64]), kwargs = {})
    %empty_15 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 256, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_15 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_117, -1, -2), kwargs = {})
    %baddbmm_15 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_15, %_unsafe_view_114, %transpose_15), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_15 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_15, -1, False), kwargs = {})
    %detach_87 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_15,), kwargs = {})
    %bmm_23 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_15, %_unsafe_view_118), kwargs = {})
    %view_286 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_23, [2, 20, 256, 64]), kwargs = {})
    %permute_78 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_286, [0, 2, 1, 3]), kwargs = {})
    %clone_70 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_78,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_119 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_70, [2, 256, 1280]), kwargs = {})
    %_param_constant364 : [#users=1] = get_attr[target=_param_constant364]
    %t_108 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant364,), kwargs = {})
    %view_287 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_119, [512, 1280]), kwargs = {})
    %_param_constant365 : [#users=1] = get_attr[target=_param_constant365]
    %addmm_52 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant365, %view_287, %t_108), kwargs = {})
    %view_288 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_52, [2, 256, 1280]), kwargs = {})
    %add_183 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_288, %add_180), kwargs = {})
    %var_mean_59 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_183, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_118 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_59, 0), kwargs = {})
    %getitem_119 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_59, 1), kwargs = {})
    %add_184 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_118, 1e-05), kwargs = {})
    %rsqrt_59 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_184,), kwargs = {})
    %sub_59 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_183, %getitem_119), kwargs = {})
    %mul_128 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_59, %rsqrt_59), kwargs = {})
    %_param_constant366 : [#users=1] = get_attr[target=_param_constant366]
    %mul_129 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_128, %_param_constant366), kwargs = {})
    %_param_constant367 : [#users=1] = get_attr[target=_param_constant367]
    %add_185 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_129, %_param_constant367), kwargs = {})
    %_param_constant368 : [#users=1] = get_attr[target=_param_constant368]
    %t_109 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant368,), kwargs = {})
    %view_289 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_185, [512, 1280]), kwargs = {})
    %_param_constant369 : [#users=1] = get_attr[target=_param_constant369]
    %addmm_53 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant369, %view_289, %t_109), kwargs = {})
    %view_290 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_53, [2, 256, 10240]), kwargs = {})
    %slice_49 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_290, -1, 0, 5120), kwargs = {})
    %slice_50 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_290, -1, 5120, 10240), kwargs = {})
    %gelu_7 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_50,), kwargs = {})
    %mul_130 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_49, %gelu_7), kwargs = {})
    %_param_constant370 : [#users=1] = get_attr[target=_param_constant370]
    %t_110 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant370,), kwargs = {})
    %view_291 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_130, [512, 5120]), kwargs = {})
    %_param_constant371 : [#users=1] = get_attr[target=_param_constant371]
    %addmm_54 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant371, %view_291, %t_110), kwargs = {})
    %view_292 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_54, [2, 256, 1280]), kwargs = {})
    %add_186 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_292, %add_183), kwargs = {})
    %_param_constant372 : [#users=1] = get_attr[target=_param_constant372]
    %t_111 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant372,), kwargs = {})
    %view_293 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_186, [512, 1280]), kwargs = {})
    %_param_constant373 : [#users=1] = get_attr[target=_param_constant373]
    %addmm_55 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant373, %view_293, %t_111), kwargs = {})
    %view_294 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_55, [2, 256, 1280]), kwargs = {})
    %view_295 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_294, [2, 16, 16, 1280]), kwargs = {})
    %permute_79 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_295, [0, 3, 1, 2]), kwargs = {})
    %clone_71 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_79,), kwargs = {memory_format: torch.contiguous_format})
    %add_187 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%clone_71, %div_14), kwargs = {})
    %cat_7 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%add_187, %add_94], 1), kwargs = {})
    %view_296 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_7, [2, 32, 80, 256]), kwargs = {})
    %var_mean_60 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_296, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_120 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_60, 0), kwargs = {})
    %getitem_121 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_60, 1), kwargs = {})
    %add_188 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_120, 1e-05), kwargs = {})
    %rsqrt_60 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_188,), kwargs = {})
    %sub_60 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_296, %getitem_121), kwargs = {})
    %mul_131 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_60, %rsqrt_60), kwargs = {})
    %view_297 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_131, [2, 2560, 16, 16]), kwargs = {})
    %_param_constant374 : [#users=1] = get_attr[target=_param_constant374]
    %unsqueeze_246 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant374, 0), kwargs = {})
    %unsqueeze_247 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_246, 2), kwargs = {})
    %unsqueeze_248 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_247, 3), kwargs = {})
    %_param_constant375 : [#users=1] = get_attr[target=_param_constant375]
    %unsqueeze_249 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant375, 0), kwargs = {})
    %unsqueeze_250 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_249, 2), kwargs = {})
    %unsqueeze_251 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_250, 3), kwargs = {})
    %mul_132 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_297, %unsqueeze_251), kwargs = {})
    %add_189 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_132, %unsqueeze_248), kwargs = {})
    %squeeze_144 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_121, 3), kwargs = {})
    %squeeze_145 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_144, 2), kwargs = {})
    %squeeze_146 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_60, 3), kwargs = {})
    %squeeze_147 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_146, 2), kwargs = {})
    %detach_88 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_145,), kwargs = {})
    %detach_89 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_147,), kwargs = {})
    %silu_43 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_189,), kwargs = {})
    %_param_constant376 : [#users=1] = get_attr[target=_param_constant376]
    %_param_constant377 : [#users=1] = get_attr[target=_param_constant377]
    %convolution_39 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_43, %_param_constant376, %_param_constant377, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_44 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant378 : [#users=1] = get_attr[target=_param_constant378]
    %t_112 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant378,), kwargs = {})
    %_param_constant379 : [#users=1] = get_attr[target=_param_constant379]
    %addmm_56 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant379, %silu_44, %t_112), kwargs = {})
    %slice_51 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_56, 0, 0, 9223372036854775807), kwargs = {})
    %slice_52 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_51, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_252 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_52, 2), kwargs = {})
    %unsqueeze_253 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_252, 3), kwargs = {})
    %add_190 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_39, %unsqueeze_253), kwargs = {})
    %view_298 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_190, [2, 32, 40, 256]), kwargs = {})
    %var_mean_61 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_298, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_122 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_61, 0), kwargs = {})
    %getitem_123 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_61, 1), kwargs = {})
    %add_191 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_122, 1e-05), kwargs = {})
    %rsqrt_61 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_191,), kwargs = {})
    %sub_61 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_298, %getitem_123), kwargs = {})
    %mul_133 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_61, %rsqrt_61), kwargs = {})
    %view_299 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_133, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant380 : [#users=1] = get_attr[target=_param_constant380]
    %unsqueeze_254 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant380, 0), kwargs = {})
    %unsqueeze_255 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_254, 2), kwargs = {})
    %unsqueeze_256 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_255, 3), kwargs = {})
    %_param_constant381 : [#users=1] = get_attr[target=_param_constant381]
    %unsqueeze_257 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant381, 0), kwargs = {})
    %unsqueeze_258 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_257, 2), kwargs = {})
    %unsqueeze_259 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_258, 3), kwargs = {})
    %mul_134 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_299, %unsqueeze_259), kwargs = {})
    %add_192 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_134, %unsqueeze_256), kwargs = {})
    %squeeze_148 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_123, 3), kwargs = {})
    %squeeze_149 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_148, 2), kwargs = {})
    %squeeze_150 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_61, 3), kwargs = {})
    %squeeze_151 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_150, 2), kwargs = {})
    %detach_90 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_149,), kwargs = {})
    %detach_91 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_151,), kwargs = {})
    %silu_45 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_192,), kwargs = {})
    %_param_constant382 : [#users=1] = get_attr[target=_param_constant382]
    %_param_constant383 : [#users=1] = get_attr[target=_param_constant383]
    %convolution_40 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_45, %_param_constant382, %_param_constant383, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant384 : [#users=1] = get_attr[target=_param_constant384]
    %_param_constant385 : [#users=1] = get_attr[target=_param_constant385]
    %convolution_41 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_7, %_param_constant384, %_param_constant385, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_193 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_41, %convolution_40), kwargs = {})
    %div_15 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_193, 1.0), kwargs = {})
    %view_300 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_15, [2, 32, 40, 256]), kwargs = {})
    %var_mean_62 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_300, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_124 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_62, 0), kwargs = {})
    %getitem_125 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_62, 1), kwargs = {})
    %add_194 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_124, 1e-06), kwargs = {})
    %rsqrt_62 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_194,), kwargs = {})
    %sub_62 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_300, %getitem_125), kwargs = {})
    %mul_135 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_62, %rsqrt_62), kwargs = {})
    %view_301 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_135, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant386 : [#users=1] = get_attr[target=_param_constant386]
    %unsqueeze_260 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant386, 0), kwargs = {})
    %unsqueeze_261 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_260, 2), kwargs = {})
    %unsqueeze_262 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_261, 3), kwargs = {})
    %_param_constant387 : [#users=1] = get_attr[target=_param_constant387]
    %unsqueeze_263 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant387, 0), kwargs = {})
    %unsqueeze_264 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_263, 2), kwargs = {})
    %unsqueeze_265 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_264, 3), kwargs = {})
    %mul_136 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_301, %unsqueeze_265), kwargs = {})
    %add_195 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_136, %unsqueeze_262), kwargs = {})
    %squeeze_152 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_125, 3), kwargs = {})
    %squeeze_153 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_152, 2), kwargs = {})
    %squeeze_154 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_62, 3), kwargs = {})
    %squeeze_155 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_154, 2), kwargs = {})
    %detach_92 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_153,), kwargs = {})
    %detach_93 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_155,), kwargs = {})
    %permute_80 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_195, [0, 2, 3, 1]), kwargs = {})
    %view_302 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_80, [2, 256, 1280]), kwargs = {})
    %_param_constant388 : [#users=1] = get_attr[target=_param_constant388]
    %t_113 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant388,), kwargs = {})
    %expand_17 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_302, [2, 256, 1280]), kwargs = {})
    %view_303 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_17, [2, 256, 1280]), kwargs = {})
    %expand_18 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_113, [2, 1280, 1280]), kwargs = {})
    %view_304 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_18, [2, 1280, 1280]), kwargs = {})
    %bmm_24 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_303, %view_304), kwargs = {})
    %_unsafe_view_120 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_24, [2, 256, 1280]), kwargs = {})
    %_param_constant389 : [#users=1] = get_attr[target=_param_constant389]
    %add_196 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_120, %_param_constant389), kwargs = {})
    %var_mean_63 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_196, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_126 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_63, 0), kwargs = {})
    %getitem_127 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_63, 1), kwargs = {})
    %add_197 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_126, 1e-05), kwargs = {})
    %rsqrt_63 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_197,), kwargs = {})
    %sub_63 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_196, %getitem_127), kwargs = {})
    %mul_137 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_63, %rsqrt_63), kwargs = {})
    %_param_constant390 : [#users=1] = get_attr[target=_param_constant390]
    %mul_138 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_137, %_param_constant390), kwargs = {})
    %_param_constant391 : [#users=1] = get_attr[target=_param_constant391]
    %add_198 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_138, %_param_constant391), kwargs = {})
    %_param_constant392 : [#users=1] = get_attr[target=_param_constant392]
    %t_114 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant392,), kwargs = {})
    %view_305 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_198, [512, 1280]), kwargs = {})
    %mm_48 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_305, %t_114), kwargs = {})
    %_unsafe_view_121 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_48, [2, 256, 1280]), kwargs = {})
    %view_306 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_121, [2, 256, 20, 64]), kwargs = {})
    %permute_81 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_306, [0, 2, 1, 3]), kwargs = {})
    %clone_72 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_81,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_122 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_72, [40, 256, 64]), kwargs = {})
    %_param_constant393 : [#users=1] = get_attr[target=_param_constant393]
    %t_115 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant393,), kwargs = {})
    %view_307 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_198, [512, 1280]), kwargs = {})
    %mm_49 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_307, %t_115), kwargs = {})
    %_unsafe_view_123 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_49, [2, 256, 1280]), kwargs = {})
    %_param_constant394 : [#users=1] = get_attr[target=_param_constant394]
    %t_116 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant394,), kwargs = {})
    %view_308 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_198, [512, 1280]), kwargs = {})
    %mm_50 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_308, %t_116), kwargs = {})
    %_unsafe_view_124 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_50, [2, 256, 1280]), kwargs = {})
    %view_309 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_123, [2, 256, 20, 64]), kwargs = {})
    %permute_82 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_309, [0, 2, 1, 3]), kwargs = {})
    %clone_73 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_82,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_125 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_73, [40, 256, 64]), kwargs = {})
    %view_310 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_124, [2, 256, 20, 64]), kwargs = {})
    %permute_83 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_310, [0, 2, 1, 3]), kwargs = {})
    %clone_74 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_83,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_126 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_74, [40, 256, 64]), kwargs = {})
    %empty_16 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 256, 256],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_16 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_125, -1, -2), kwargs = {})
    %baddbmm_16 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_16, %_unsafe_view_122, %transpose_16), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_16 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_16, -1, False), kwargs = {})
    %detach_94 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_16,), kwargs = {})
    %bmm_25 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_16, %_unsafe_view_126), kwargs = {})
    %view_311 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_25, [2, 20, 256, 64]), kwargs = {})
    %permute_84 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_311, [0, 2, 1, 3]), kwargs = {})
    %clone_75 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_84,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_127 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_75, [2, 256, 1280]), kwargs = {})
    %_param_constant395 : [#users=1] = get_attr[target=_param_constant395]
    %t_117 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant395,), kwargs = {})
    %view_312 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_127, [512, 1280]), kwargs = {})
    %_param_constant396 : [#users=1] = get_attr[target=_param_constant396]
    %addmm_57 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant396, %view_312, %t_117), kwargs = {})
    %view_313 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_57, [2, 256, 1280]), kwargs = {})
    %add_199 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_313, %add_196), kwargs = {})
    %var_mean_64 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_199, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_128 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_64, 0), kwargs = {})
    %getitem_129 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_64, 1), kwargs = {})
    %add_200 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_128, 1e-05), kwargs = {})
    %rsqrt_64 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_200,), kwargs = {})
    %sub_64 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_199, %getitem_129), kwargs = {})
    %mul_139 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_64, %rsqrt_64), kwargs = {})
    %_param_constant397 : [#users=1] = get_attr[target=_param_constant397]
    %mul_140 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_139, %_param_constant397), kwargs = {})
    %_param_constant398 : [#users=1] = get_attr[target=_param_constant398]
    %add_201 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_140, %_param_constant398), kwargs = {})
    %_param_constant399 : [#users=1] = get_attr[target=_param_constant399]
    %t_118 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant399,), kwargs = {})
    %view_314 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_201, [512, 1280]), kwargs = {})
    %mm_51 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_314, %t_118), kwargs = {})
    %_unsafe_view_128 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_51, [2, 256, 1280]), kwargs = {})
    %view_315 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_128, [2, 256, 20, 64]), kwargs = {})
    %permute_85 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_315, [0, 2, 1, 3]), kwargs = {})
    %clone_76 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_85,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_129 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_76, [40, 256, 64]), kwargs = {})
    %_param_constant400 : [#users=1] = get_attr[target=_param_constant400]
    %t_119 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant400,), kwargs = {})
    %view_316 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_52 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_316, %t_119), kwargs = {})
    %_unsafe_view_130 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_52, [2, 64, 1280]), kwargs = {})
    %_param_constant401 : [#users=1] = get_attr[target=_param_constant401]
    %t_120 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant401,), kwargs = {})
    %view_317 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_53 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_317, %t_120), kwargs = {})
    %_unsafe_view_131 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_53, [2, 64, 1280]), kwargs = {})
    %view_318 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_130, [2, 64, 20, 64]), kwargs = {})
    %permute_86 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_318, [0, 2, 1, 3]), kwargs = {})
    %clone_77 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_86,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_132 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_77, [40, 64, 64]), kwargs = {})
    %view_319 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_131, [2, 64, 20, 64]), kwargs = {})
    %permute_87 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_319, [0, 2, 1, 3]), kwargs = {})
    %clone_78 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_87,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_133 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_78, [40, 64, 64]), kwargs = {})
    %empty_17 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 256, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_17 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_132, -1, -2), kwargs = {})
    %baddbmm_17 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_17, %_unsafe_view_129, %transpose_17), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_17 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_17, -1, False), kwargs = {})
    %detach_95 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_17,), kwargs = {})
    %bmm_26 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_17, %_unsafe_view_133), kwargs = {})
    %view_320 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_26, [2, 20, 256, 64]), kwargs = {})
    %permute_88 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_320, [0, 2, 1, 3]), kwargs = {})
    %clone_79 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_88,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_134 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_79, [2, 256, 1280]), kwargs = {})
    %_param_constant402 : [#users=1] = get_attr[target=_param_constant402]
    %t_121 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant402,), kwargs = {})
    %view_321 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_134, [512, 1280]), kwargs = {})
    %_param_constant403 : [#users=1] = get_attr[target=_param_constant403]
    %addmm_58 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant403, %view_321, %t_121), kwargs = {})
    %view_322 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_58, [2, 256, 1280]), kwargs = {})
    %add_202 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_322, %add_199), kwargs = {})
    %var_mean_65 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_202, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_130 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_65, 0), kwargs = {})
    %getitem_131 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_65, 1), kwargs = {})
    %add_203 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_130, 1e-05), kwargs = {})
    %rsqrt_65 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_203,), kwargs = {})
    %sub_65 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_202, %getitem_131), kwargs = {})
    %mul_141 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_65, %rsqrt_65), kwargs = {})
    %_param_constant404 : [#users=1] = get_attr[target=_param_constant404]
    %mul_142 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_141, %_param_constant404), kwargs = {})
    %_param_constant405 : [#users=1] = get_attr[target=_param_constant405]
    %add_204 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_142, %_param_constant405), kwargs = {})
    %_param_constant406 : [#users=1] = get_attr[target=_param_constant406]
    %t_122 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant406,), kwargs = {})
    %view_323 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_204, [512, 1280]), kwargs = {})
    %_param_constant407 : [#users=1] = get_attr[target=_param_constant407]
    %addmm_59 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant407, %view_323, %t_122), kwargs = {})
    %view_324 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_59, [2, 256, 10240]), kwargs = {})
    %slice_53 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_324, -1, 0, 5120), kwargs = {})
    %slice_54 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_324, -1, 5120, 10240), kwargs = {})
    %gelu_8 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_54,), kwargs = {})
    %mul_143 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_53, %gelu_8), kwargs = {})
    %_param_constant408 : [#users=1] = get_attr[target=_param_constant408]
    %t_123 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant408,), kwargs = {})
    %view_325 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_143, [512, 5120]), kwargs = {})
    %_param_constant409 : [#users=1] = get_attr[target=_param_constant409]
    %addmm_60 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant409, %view_325, %t_123), kwargs = {})
    %view_326 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_60, [2, 256, 1280]), kwargs = {})
    %add_205 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_326, %add_202), kwargs = {})
    %_param_constant410 : [#users=1] = get_attr[target=_param_constant410]
    %t_124 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant410,), kwargs = {})
    %view_327 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_205, [512, 1280]), kwargs = {})
    %_param_constant411 : [#users=1] = get_attr[target=_param_constant411]
    %addmm_61 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant411, %view_327, %t_124), kwargs = {})
    %view_328 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_61, [2, 256, 1280]), kwargs = {})
    %view_329 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_328, [2, 16, 16, 1280]), kwargs = {})
    %permute_89 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_329, [0, 3, 1, 2]), kwargs = {})
    %clone_80 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_89,), kwargs = {memory_format: torch.contiguous_format})
    %add_206 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%clone_80, %div_15), kwargs = {})
    %cat_8 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%add_206, %convolution_11], 1), kwargs = {})
    %view_330 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_8, [2, 32, 60, 256]), kwargs = {})
    %var_mean_66 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_330, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_132 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_66, 0), kwargs = {})
    %getitem_133 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_66, 1), kwargs = {})
    %add_207 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_132, 1e-05), kwargs = {})
    %rsqrt_66 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_207,), kwargs = {})
    %sub_66 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_330, %getitem_133), kwargs = {})
    %mul_144 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_66, %rsqrt_66), kwargs = {})
    %view_331 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_144, [2, 1920, 16, 16]), kwargs = {})
    %_param_constant412 : [#users=1] = get_attr[target=_param_constant412]
    %unsqueeze_266 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant412, 0), kwargs = {})
    %unsqueeze_267 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_266, 2), kwargs = {})
    %unsqueeze_268 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_267, 3), kwargs = {})
    %_param_constant413 : [#users=1] = get_attr[target=_param_constant413]
    %unsqueeze_269 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant413, 0), kwargs = {})
    %unsqueeze_270 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_269, 2), kwargs = {})
    %unsqueeze_271 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_270, 3), kwargs = {})
    %mul_145 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_331, %unsqueeze_271), kwargs = {})
    %add_208 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_145, %unsqueeze_268), kwargs = {})
    %squeeze_156 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_133, 3), kwargs = {})
    %squeeze_157 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_156, 2), kwargs = {})
    %squeeze_158 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_66, 3), kwargs = {})
    %squeeze_159 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_158, 2), kwargs = {})
    %detach_96 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_157,), kwargs = {})
    %detach_97 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_159,), kwargs = {})
    %silu_46 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_208,), kwargs = {})
    %_param_constant414 : [#users=1] = get_attr[target=_param_constant414]
    %_param_constant415 : [#users=1] = get_attr[target=_param_constant415]
    %convolution_42 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_46, %_param_constant414, %_param_constant415, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_47 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant416 : [#users=1] = get_attr[target=_param_constant416]
    %t_125 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant416,), kwargs = {})
    %_param_constant417 : [#users=1] = get_attr[target=_param_constant417]
    %addmm_62 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant417, %silu_47, %t_125), kwargs = {})
    %slice_55 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_62, 0, 0, 9223372036854775807), kwargs = {})
    %slice_56 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_55, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_272 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_56, 2), kwargs = {})
    %unsqueeze_273 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_272, 3), kwargs = {})
    %add_209 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_42, %unsqueeze_273), kwargs = {})
    %view_332 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_209, [2, 32, 40, 256]), kwargs = {})
    %var_mean_67 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_332, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_134 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_67, 0), kwargs = {})
    %getitem_135 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_67, 1), kwargs = {})
    %add_210 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_134, 1e-05), kwargs = {})
    %rsqrt_67 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_210,), kwargs = {})
    %sub_67 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_332, %getitem_135), kwargs = {})
    %mul_146 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_67, %rsqrt_67), kwargs = {})
    %view_333 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_146, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant418 : [#users=1] = get_attr[target=_param_constant418]
    %unsqueeze_274 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant418, 0), kwargs = {})
    %unsqueeze_275 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_274, 2), kwargs = {})
    %unsqueeze_276 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_275, 3), kwargs = {})
    %_param_constant419 : [#users=1] = get_attr[target=_param_constant419]
    %unsqueeze_277 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant419, 0), kwargs = {})
    %unsqueeze_278 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_277, 2), kwargs = {})
    %unsqueeze_279 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_278, 3), kwargs = {})
    %mul_147 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_333, %unsqueeze_279), kwargs = {})
    %add_211 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_147, %unsqueeze_276), kwargs = {})
    %squeeze_160 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_135, 3), kwargs = {})
    %squeeze_161 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_160, 2), kwargs = {})
    %squeeze_162 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_67, 3), kwargs = {})
    %squeeze_163 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_162, 2), kwargs = {})
    %detach_98 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_161,), kwargs = {})
    %detach_99 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_163,), kwargs = {})
    %silu_48 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_211,), kwargs = {})
    %_param_constant420 : [#users=1] = get_attr[target=_param_constant420]
    %_param_constant421 : [#users=1] = get_attr[target=_param_constant421]
    %convolution_43 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_48, %_param_constant420, %_param_constant421, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant422 : [#users=1] = get_attr[target=_param_constant422]
    %_param_constant423 : [#users=1] = get_attr[target=_param_constant423]
    %convolution_44 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_8, %_param_constant422, %_param_constant423, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_212 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_44, %convolution_43), kwargs = {})
    %div_16 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_212, 1.0), kwargs = {})
    %view_334 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_16, [2, 32, 40, 256]), kwargs = {})
    %var_mean_68 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_334, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_136 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_68, 0), kwargs = {})
    %getitem_137 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_68, 1), kwargs = {})
    %add_213 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_136, 1e-06), kwargs = {})
    %rsqrt_68 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_213,), kwargs = {})
    %sub_68 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_334, %getitem_137), kwargs = {})
    %mul_148 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_68, %rsqrt_68), kwargs = {})
    %view_335 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_148, [2, 1280, 16, 16]), kwargs = {})
    %_param_constant424 : [#users=1] = get_attr[target=_param_constant424]
    %unsqueeze_280 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant424, 0), kwargs = {})
    %unsqueeze_281 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_280, 2), kwargs = {})
    %unsqueeze_282 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_281, 3), kwargs = {})
    %_param_constant425 : [#users=1] = get_attr[target=_param_constant425]
    %unsqueeze_283 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant425, 0), kwargs = {})
    %unsqueeze_284 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_283, 2), kwargs = {})
    %unsqueeze_285 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_284, 3), kwargs = {})
    %mul_149 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_335, %unsqueeze_285), kwargs = {})
    %add_214 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_149, %unsqueeze_282), kwargs = {})
    %squeeze_164 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_137, 3), kwargs = {})
    %squeeze_165 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_164, 2), kwargs = {})
    %squeeze_166 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_68, 3), kwargs = {})
    %squeeze_167 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_166, 2), kwargs = {})
    %detach_100 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_165,), kwargs = {})
    %detach_101 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_167,), kwargs = {})
    %permute_90 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_214, [0, 2, 3, 1]), kwargs = {})
    %view_336 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_90, [2, 256, 1280]), kwargs = {})
    %_param_constant426 : [#users=1] = get_attr[target=_param_constant426]
    %t_126 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant426,), kwargs = {})
    %expand_19 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_336, [2, 256, 1280]), kwargs = {})
    %view_337 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_19, [2, 256, 1280]), kwargs = {})
    %expand_20 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_126, [2, 1280, 1280]), kwargs = {})
    %view_338 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_20, [2, 1280, 1280]), kwargs = {})
    %bmm_27 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_337, %view_338), kwargs = {})
    %_unsafe_view_135 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_27, [2, 256, 1280]), kwargs = {})
    %_param_constant427 : [#users=1] = get_attr[target=_param_constant427]
    %add_215 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_135, %_param_constant427), kwargs = {})
    %var_mean_69 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_215, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_138 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_69, 0), kwargs = {})
    %getitem_139 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_69, 1), kwargs = {})
    %add_216 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_138, 1e-05), kwargs = {})
    %rsqrt_69 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_216,), kwargs = {})
    %sub_69 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_215, %getitem_139), kwargs = {})
    %mul_150 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_69, %rsqrt_69), kwargs = {})
    %_param_constant428 : [#users=1] = get_attr[target=_param_constant428]
    %mul_151 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_150, %_param_constant428), kwargs = {})
    %_param_constant429 : [#users=1] = get_attr[target=_param_constant429]
    %add_217 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_151, %_param_constant429), kwargs = {})
    %_param_constant430 : [#users=1] = get_attr[target=_param_constant430]
    %t_127 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant430,), kwargs = {})
    %view_339 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_217, [512, 1280]), kwargs = {})
    %mm_54 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_339, %t_127), kwargs = {})
    %_unsafe_view_136 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_54, [2, 256, 1280]), kwargs = {})
    %view_340 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_136, [2, 256, 20, 64]), kwargs = {})
    %permute_91 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_340, [0, 2, 1, 3]), kwargs = {})
    %clone_81 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_91,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_137 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_81, [40, 256, 64]), kwargs = {})
    %_param_constant431 : [#users=1] = get_attr[target=_param_constant431]
    %t_128 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant431,), kwargs = {})
    %view_341 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_217, [512, 1280]), kwargs = {})
    %mm_55 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_341, %t_128), kwargs = {})
    %_unsafe_view_138 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_55, [2, 256, 1280]), kwargs = {})
    %_param_constant432 : [#users=1] = get_attr[target=_param_constant432]
    %t_129 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant432,), kwargs = {})
    %view_342 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_217, [512, 1280]), kwargs = {})
    %mm_56 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_342, %t_129), kwargs = {})
    %_unsafe_view_139 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_56, [2, 256, 1280]), kwargs = {})
    %view_343 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_138, [2, 256, 20, 64]), kwargs = {})
    %permute_92 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_343, [0, 2, 1, 3]), kwargs = {})
    %clone_82 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_92,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_140 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_82, [40, 256, 64]), kwargs = {})
    %view_344 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_139, [2, 256, 20, 64]), kwargs = {})
    %permute_93 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_344, [0, 2, 1, 3]), kwargs = {})
    %clone_83 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_93,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_141 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_83, [40, 256, 64]), kwargs = {})
    %empty_18 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 256, 256],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_18 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_140, -1, -2), kwargs = {})
    %baddbmm_18 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_18, %_unsafe_view_137, %transpose_18), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_18 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_18, -1, False), kwargs = {})
    %detach_102 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_18,), kwargs = {})
    %bmm_28 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_18, %_unsafe_view_141), kwargs = {})
    %view_345 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_28, [2, 20, 256, 64]), kwargs = {})
    %permute_94 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_345, [0, 2, 1, 3]), kwargs = {})
    %clone_84 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_94,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_142 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_84, [2, 256, 1280]), kwargs = {})
    %_param_constant433 : [#users=1] = get_attr[target=_param_constant433]
    %t_130 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant433,), kwargs = {})
    %view_346 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_142, [512, 1280]), kwargs = {})
    %_param_constant434 : [#users=1] = get_attr[target=_param_constant434]
    %addmm_63 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant434, %view_346, %t_130), kwargs = {})
    %view_347 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_63, [2, 256, 1280]), kwargs = {})
    %add_218 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_347, %add_215), kwargs = {})
    %var_mean_70 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_218, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_140 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_70, 0), kwargs = {})
    %getitem_141 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_70, 1), kwargs = {})
    %add_219 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_140, 1e-05), kwargs = {})
    %rsqrt_70 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_219,), kwargs = {})
    %sub_70 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_218, %getitem_141), kwargs = {})
    %mul_152 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_70, %rsqrt_70), kwargs = {})
    %_param_constant435 : [#users=1] = get_attr[target=_param_constant435]
    %mul_153 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_152, %_param_constant435), kwargs = {})
    %_param_constant436 : [#users=1] = get_attr[target=_param_constant436]
    %add_220 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_153, %_param_constant436), kwargs = {})
    %_param_constant437 : [#users=1] = get_attr[target=_param_constant437]
    %t_131 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant437,), kwargs = {})
    %view_348 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_220, [512, 1280]), kwargs = {})
    %mm_57 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_348, %t_131), kwargs = {})
    %_unsafe_view_143 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_57, [2, 256, 1280]), kwargs = {})
    %view_349 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_143, [2, 256, 20, 64]), kwargs = {})
    %permute_95 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_349, [0, 2, 1, 3]), kwargs = {})
    %clone_85 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_95,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_144 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_85, [40, 256, 64]), kwargs = {})
    %_param_constant438 : [#users=1] = get_attr[target=_param_constant438]
    %t_132 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant438,), kwargs = {})
    %view_350 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_58 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_350, %t_132), kwargs = {})
    %_unsafe_view_145 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_58, [2, 64, 1280]), kwargs = {})
    %_param_constant439 : [#users=1] = get_attr[target=_param_constant439]
    %t_133 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant439,), kwargs = {})
    %view_351 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_59 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_351, %t_133), kwargs = {})
    %_unsafe_view_146 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_59, [2, 64, 1280]), kwargs = {})
    %view_352 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_145, [2, 64, 20, 64]), kwargs = {})
    %permute_96 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_352, [0, 2, 1, 3]), kwargs = {})
    %clone_86 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_96,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_147 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_86, [40, 64, 64]), kwargs = {})
    %view_353 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_146, [2, 64, 20, 64]), kwargs = {})
    %permute_97 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_353, [0, 2, 1, 3]), kwargs = {})
    %clone_87 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_97,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_148 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_87, [40, 64, 64]), kwargs = {})
    %empty_19 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([40, 256, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_19 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_147, -1, -2), kwargs = {})
    %baddbmm_19 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_19, %_unsafe_view_144, %transpose_19), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_19 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_19, -1, False), kwargs = {})
    %detach_103 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_19,), kwargs = {})
    %bmm_29 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_19, %_unsafe_view_148), kwargs = {})
    %view_354 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_29, [2, 20, 256, 64]), kwargs = {})
    %permute_98 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_354, [0, 2, 1, 3]), kwargs = {})
    %clone_88 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_98,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_149 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_88, [2, 256, 1280]), kwargs = {})
    %_param_constant440 : [#users=1] = get_attr[target=_param_constant440]
    %t_134 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant440,), kwargs = {})
    %view_355 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_149, [512, 1280]), kwargs = {})
    %_param_constant441 : [#users=1] = get_attr[target=_param_constant441]
    %addmm_64 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant441, %view_355, %t_134), kwargs = {})
    %view_356 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_64, [2, 256, 1280]), kwargs = {})
    %add_221 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_356, %add_218), kwargs = {})
    %var_mean_71 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_221, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_142 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_71, 0), kwargs = {})
    %getitem_143 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_71, 1), kwargs = {})
    %add_222 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_142, 1e-05), kwargs = {})
    %rsqrt_71 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_222,), kwargs = {})
    %sub_71 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_221, %getitem_143), kwargs = {})
    %mul_154 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_71, %rsqrt_71), kwargs = {})
    %_param_constant442 : [#users=1] = get_attr[target=_param_constant442]
    %mul_155 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_154, %_param_constant442), kwargs = {})
    %_param_constant443 : [#users=1] = get_attr[target=_param_constant443]
    %add_223 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_155, %_param_constant443), kwargs = {})
    %_param_constant444 : [#users=1] = get_attr[target=_param_constant444]
    %t_135 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant444,), kwargs = {})
    %view_357 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_223, [512, 1280]), kwargs = {})
    %_param_constant445 : [#users=1] = get_attr[target=_param_constant445]
    %addmm_65 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant445, %view_357, %t_135), kwargs = {})
    %view_358 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_65, [2, 256, 10240]), kwargs = {})
    %slice_57 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_358, -1, 0, 5120), kwargs = {})
    %slice_58 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_358, -1, 5120, 10240), kwargs = {})
    %gelu_9 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_58,), kwargs = {})
    %mul_156 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_57, %gelu_9), kwargs = {})
    %_param_constant446 : [#users=1] = get_attr[target=_param_constant446]
    %t_136 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant446,), kwargs = {})
    %view_359 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_156, [512, 5120]), kwargs = {})
    %_param_constant447 : [#users=1] = get_attr[target=_param_constant447]
    %addmm_66 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant447, %view_359, %t_136), kwargs = {})
    %view_360 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_66, [2, 256, 1280]), kwargs = {})
    %add_224 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_360, %add_221), kwargs = {})
    %_param_constant448 : [#users=1] = get_attr[target=_param_constant448]
    %t_137 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant448,), kwargs = {})
    %view_361 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_224, [512, 1280]), kwargs = {})
    %_param_constant449 : [#users=1] = get_attr[target=_param_constant449]
    %addmm_67 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant449, %view_361, %t_137), kwargs = {})
    %view_362 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_67, [2, 256, 1280]), kwargs = {})
    %view_363 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_362, [2, 16, 16, 1280]), kwargs = {})
    %permute_99 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_363, [0, 3, 1, 2]), kwargs = {})
    %clone_89 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_99,), kwargs = {memory_format: torch.contiguous_format})
    %add_225 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%clone_89, %div_16), kwargs = {})
    %upsample_nearest2d_1 : [#users=1] = call_function[target=torch.ops.aten.upsample_nearest2d](args = (%add_225, [32, 32], 2.0, 2.0), kwargs = {})
    %_param_constant450 : [#users=1] = get_attr[target=_param_constant450]
    %_param_constant451 : [#users=1] = get_attr[target=_param_constant451]
    %convolution_45 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%upsample_nearest2d_1, %_param_constant450, %_param_constant451, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %cat_9 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%convolution_45, %add_75], 1), kwargs = {})
    %view_364 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_9, [2, 32, 60, 1024]), kwargs = {})
    %var_mean_72 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_364, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_144 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_72, 0), kwargs = {})
    %getitem_145 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_72, 1), kwargs = {})
    %add_226 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_144, 1e-05), kwargs = {})
    %rsqrt_72 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_226,), kwargs = {})
    %sub_72 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_364, %getitem_145), kwargs = {})
    %mul_157 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_72, %rsqrt_72), kwargs = {})
    %view_365 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_157, [2, 1920, 32, 32]), kwargs = {})
    %_param_constant452 : [#users=1] = get_attr[target=_param_constant452]
    %unsqueeze_286 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant452, 0), kwargs = {})
    %unsqueeze_287 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_286, 2), kwargs = {})
    %unsqueeze_288 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_287, 3), kwargs = {})
    %_param_constant453 : [#users=1] = get_attr[target=_param_constant453]
    %unsqueeze_289 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant453, 0), kwargs = {})
    %unsqueeze_290 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_289, 2), kwargs = {})
    %unsqueeze_291 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_290, 3), kwargs = {})
    %mul_158 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_365, %unsqueeze_291), kwargs = {})
    %add_227 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_158, %unsqueeze_288), kwargs = {})
    %squeeze_168 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_145, 3), kwargs = {})
    %squeeze_169 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_168, 2), kwargs = {})
    %squeeze_170 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_72, 3), kwargs = {})
    %squeeze_171 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_170, 2), kwargs = {})
    %detach_104 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_169,), kwargs = {})
    %detach_105 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_171,), kwargs = {})
    %silu_49 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_227,), kwargs = {})
    %_param_constant454 : [#users=1] = get_attr[target=_param_constant454]
    %_param_constant455 : [#users=1] = get_attr[target=_param_constant455]
    %convolution_46 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_49, %_param_constant454, %_param_constant455, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_50 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant456 : [#users=1] = get_attr[target=_param_constant456]
    %t_138 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant456,), kwargs = {})
    %_param_constant457 : [#users=1] = get_attr[target=_param_constant457]
    %addmm_68 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant457, %silu_50, %t_138), kwargs = {})
    %slice_59 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_68, 0, 0, 9223372036854775807), kwargs = {})
    %slice_60 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_59, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_292 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_60, 2), kwargs = {})
    %unsqueeze_293 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_292, 3), kwargs = {})
    %add_228 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_46, %unsqueeze_293), kwargs = {})
    %view_366 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_228, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_73 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_366, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_146 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_73, 0), kwargs = {})
    %getitem_147 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_73, 1), kwargs = {})
    %add_229 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_146, 1e-05), kwargs = {})
    %rsqrt_73 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_229,), kwargs = {})
    %sub_73 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_366, %getitem_147), kwargs = {})
    %mul_159 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_73, %rsqrt_73), kwargs = {})
    %view_367 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_159, [2, 640, 32, 32]), kwargs = {})
    %_param_constant458 : [#users=1] = get_attr[target=_param_constant458]
    %unsqueeze_294 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant458, 0), kwargs = {})
    %unsqueeze_295 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_294, 2), kwargs = {})
    %unsqueeze_296 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_295, 3), kwargs = {})
    %_param_constant459 : [#users=1] = get_attr[target=_param_constant459]
    %unsqueeze_297 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant459, 0), kwargs = {})
    %unsqueeze_298 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_297, 2), kwargs = {})
    %unsqueeze_299 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_298, 3), kwargs = {})
    %mul_160 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_367, %unsqueeze_299), kwargs = {})
    %add_230 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_160, %unsqueeze_296), kwargs = {})
    %squeeze_172 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_147, 3), kwargs = {})
    %squeeze_173 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_172, 2), kwargs = {})
    %squeeze_174 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_73, 3), kwargs = {})
    %squeeze_175 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_174, 2), kwargs = {})
    %detach_106 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_173,), kwargs = {})
    %detach_107 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_175,), kwargs = {})
    %silu_51 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_230,), kwargs = {})
    %_param_constant460 : [#users=1] = get_attr[target=_param_constant460]
    %_param_constant461 : [#users=1] = get_attr[target=_param_constant461]
    %convolution_47 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_51, %_param_constant460, %_param_constant461, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant462 : [#users=1] = get_attr[target=_param_constant462]
    %_param_constant463 : [#users=1] = get_attr[target=_param_constant463]
    %convolution_48 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_9, %_param_constant462, %_param_constant463, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_231 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_48, %convolution_47), kwargs = {})
    %div_17 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_231, 1.0), kwargs = {})
    %view_368 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_17, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_74 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_368, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_148 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_74, 0), kwargs = {})
    %getitem_149 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_74, 1), kwargs = {})
    %add_232 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_148, 1e-06), kwargs = {})
    %rsqrt_74 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_232,), kwargs = {})
    %sub_74 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_368, %getitem_149), kwargs = {})
    %mul_161 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_74, %rsqrt_74), kwargs = {})
    %view_369 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_161, [2, 640, 32, 32]), kwargs = {})
    %_param_constant464 : [#users=1] = get_attr[target=_param_constant464]
    %unsqueeze_300 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant464, 0), kwargs = {})
    %unsqueeze_301 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_300, 2), kwargs = {})
    %unsqueeze_302 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_301, 3), kwargs = {})
    %_param_constant465 : [#users=1] = get_attr[target=_param_constant465]
    %unsqueeze_303 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant465, 0), kwargs = {})
    %unsqueeze_304 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_303, 2), kwargs = {})
    %unsqueeze_305 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_304, 3), kwargs = {})
    %mul_162 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_369, %unsqueeze_305), kwargs = {})
    %add_233 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_162, %unsqueeze_302), kwargs = {})
    %squeeze_176 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_149, 3), kwargs = {})
    %squeeze_177 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_176, 2), kwargs = {})
    %squeeze_178 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_74, 3), kwargs = {})
    %squeeze_179 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_178, 2), kwargs = {})
    %detach_108 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_177,), kwargs = {})
    %detach_109 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_179,), kwargs = {})
    %permute_100 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_233, [0, 2, 3, 1]), kwargs = {})
    %view_370 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_100, [2, 1024, 640]), kwargs = {})
    %_param_constant466 : [#users=1] = get_attr[target=_param_constant466]
    %t_139 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant466,), kwargs = {})
    %expand_21 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_370, [2, 1024, 640]), kwargs = {})
    %view_371 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_21, [2, 1024, 640]), kwargs = {})
    %expand_22 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_139, [2, 640, 640]), kwargs = {})
    %view_372 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_22, [2, 640, 640]), kwargs = {})
    %bmm_30 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_371, %view_372), kwargs = {})
    %_unsafe_view_150 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_30, [2, 1024, 640]), kwargs = {})
    %_param_constant467 : [#users=1] = get_attr[target=_param_constant467]
    %add_234 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_150, %_param_constant467), kwargs = {})
    %var_mean_75 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_234, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_150 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_75, 0), kwargs = {})
    %getitem_151 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_75, 1), kwargs = {})
    %add_235 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_150, 1e-05), kwargs = {})
    %rsqrt_75 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_235,), kwargs = {})
    %sub_75 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_234, %getitem_151), kwargs = {})
    %mul_163 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_75, %rsqrt_75), kwargs = {})
    %_param_constant468 : [#users=1] = get_attr[target=_param_constant468]
    %mul_164 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_163, %_param_constant468), kwargs = {})
    %_param_constant469 : [#users=1] = get_attr[target=_param_constant469]
    %add_236 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_164, %_param_constant469), kwargs = {})
    %_param_constant470 : [#users=1] = get_attr[target=_param_constant470]
    %t_140 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant470,), kwargs = {})
    %view_373 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_236, [2048, 640]), kwargs = {})
    %mm_60 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_373, %t_140), kwargs = {})
    %_unsafe_view_151 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_60, [2, 1024, 640]), kwargs = {})
    %view_374 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_151, [2, 1024, 10, 64]), kwargs = {})
    %permute_101 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_374, [0, 2, 1, 3]), kwargs = {})
    %clone_90 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_101,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_152 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_90, [20, 1024, 64]), kwargs = {})
    %_param_constant471 : [#users=1] = get_attr[target=_param_constant471]
    %t_141 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant471,), kwargs = {})
    %view_375 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_236, [2048, 640]), kwargs = {})
    %mm_61 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_375, %t_141), kwargs = {})
    %_unsafe_view_153 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_61, [2, 1024, 640]), kwargs = {})
    %_param_constant472 : [#users=1] = get_attr[target=_param_constant472]
    %t_142 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant472,), kwargs = {})
    %view_376 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_236, [2048, 640]), kwargs = {})
    %mm_62 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_376, %t_142), kwargs = {})
    %_unsafe_view_154 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_62, [2, 1024, 640]), kwargs = {})
    %view_377 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_153, [2, 1024, 10, 64]), kwargs = {})
    %permute_102 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_377, [0, 2, 1, 3]), kwargs = {})
    %clone_91 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_102,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_155 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_91, [20, 1024, 64]), kwargs = {})
    %view_378 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_154, [2, 1024, 10, 64]), kwargs = {})
    %permute_103 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_378, [0, 2, 1, 3]), kwargs = {})
    %clone_92 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_103,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_156 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_92, [20, 1024, 64]), kwargs = {})
    %empty_20 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([20, 1024, 1024],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_20 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_155, -1, -2), kwargs = {})
    %baddbmm_20 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_20, %_unsafe_view_152, %transpose_20), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_20 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_20, -1, False), kwargs = {})
    %detach_110 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_20,), kwargs = {})
    %bmm_31 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_20, %_unsafe_view_156), kwargs = {})
    %view_379 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_31, [2, 10, 1024, 64]), kwargs = {})
    %permute_104 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_379, [0, 2, 1, 3]), kwargs = {})
    %clone_93 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_104,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_157 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_93, [2, 1024, 640]), kwargs = {})
    %_param_constant473 : [#users=1] = get_attr[target=_param_constant473]
    %t_143 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant473,), kwargs = {})
    %view_380 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_157, [2048, 640]), kwargs = {})
    %_param_constant474 : [#users=1] = get_attr[target=_param_constant474]
    %addmm_69 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant474, %view_380, %t_143), kwargs = {})
    %view_381 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_69, [2, 1024, 640]), kwargs = {})
    %add_237 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_381, %add_234), kwargs = {})
    %var_mean_76 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_237, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_152 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_76, 0), kwargs = {})
    %getitem_153 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_76, 1), kwargs = {})
    %add_238 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_152, 1e-05), kwargs = {})
    %rsqrt_76 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_238,), kwargs = {})
    %sub_76 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_237, %getitem_153), kwargs = {})
    %mul_165 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_76, %rsqrt_76), kwargs = {})
    %_param_constant475 : [#users=1] = get_attr[target=_param_constant475]
    %mul_166 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_165, %_param_constant475), kwargs = {})
    %_param_constant476 : [#users=1] = get_attr[target=_param_constant476]
    %add_239 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_166, %_param_constant476), kwargs = {})
    %_param_constant477 : [#users=1] = get_attr[target=_param_constant477]
    %t_144 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant477,), kwargs = {})
    %view_382 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_239, [2048, 640]), kwargs = {})
    %mm_63 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_382, %t_144), kwargs = {})
    %_unsafe_view_158 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_63, [2, 1024, 640]), kwargs = {})
    %view_383 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_158, [2, 1024, 10, 64]), kwargs = {})
    %permute_105 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_383, [0, 2, 1, 3]), kwargs = {})
    %clone_94 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_105,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_159 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_94, [20, 1024, 64]), kwargs = {})
    %_param_constant478 : [#users=1] = get_attr[target=_param_constant478]
    %t_145 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant478,), kwargs = {})
    %view_384 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_64 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_384, %t_145), kwargs = {})
    %_unsafe_view_160 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_64, [2, 64, 640]), kwargs = {})
    %_param_constant479 : [#users=1] = get_attr[target=_param_constant479]
    %t_146 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant479,), kwargs = {})
    %view_385 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_65 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_385, %t_146), kwargs = {})
    %_unsafe_view_161 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_65, [2, 64, 640]), kwargs = {})
    %view_386 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_160, [2, 64, 10, 64]), kwargs = {})
    %permute_106 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_386, [0, 2, 1, 3]), kwargs = {})
    %clone_95 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_106,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_162 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_95, [20, 64, 64]), kwargs = {})
    %view_387 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_161, [2, 64, 10, 64]), kwargs = {})
    %permute_107 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_387, [0, 2, 1, 3]), kwargs = {})
    %clone_96 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_107,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_163 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_96, [20, 64, 64]), kwargs = {})
    %empty_21 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([20, 1024, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_21 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_162, -1, -2), kwargs = {})
    %baddbmm_21 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_21, %_unsafe_view_159, %transpose_21), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_21 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_21, -1, False), kwargs = {})
    %detach_111 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_21,), kwargs = {})
    %bmm_32 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_21, %_unsafe_view_163), kwargs = {})
    %view_388 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_32, [2, 10, 1024, 64]), kwargs = {})
    %permute_108 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_388, [0, 2, 1, 3]), kwargs = {})
    %clone_97 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_108,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_164 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_97, [2, 1024, 640]), kwargs = {})
    %_param_constant480 : [#users=1] = get_attr[target=_param_constant480]
    %t_147 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant480,), kwargs = {})
    %view_389 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_164, [2048, 640]), kwargs = {})
    %_param_constant481 : [#users=1] = get_attr[target=_param_constant481]
    %addmm_70 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant481, %view_389, %t_147), kwargs = {})
    %view_390 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_70, [2, 1024, 640]), kwargs = {})
    %add_240 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_390, %add_237), kwargs = {})
    %var_mean_77 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_240, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_154 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_77, 0), kwargs = {})
    %getitem_155 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_77, 1), kwargs = {})
    %add_241 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_154, 1e-05), kwargs = {})
    %rsqrt_77 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_241,), kwargs = {})
    %sub_77 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_240, %getitem_155), kwargs = {})
    %mul_167 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_77, %rsqrt_77), kwargs = {})
    %_param_constant482 : [#users=1] = get_attr[target=_param_constant482]
    %mul_168 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_167, %_param_constant482), kwargs = {})
    %_param_constant483 : [#users=1] = get_attr[target=_param_constant483]
    %add_242 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_168, %_param_constant483), kwargs = {})
    %_param_constant484 : [#users=1] = get_attr[target=_param_constant484]
    %t_148 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant484,), kwargs = {})
    %view_391 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_242, [2048, 640]), kwargs = {})
    %_param_constant485 : [#users=1] = get_attr[target=_param_constant485]
    %addmm_71 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant485, %view_391, %t_148), kwargs = {})
    %view_392 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_71, [2, 1024, 5120]), kwargs = {})
    %slice_61 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_392, -1, 0, 2560), kwargs = {})
    %slice_62 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_392, -1, 2560, 5120), kwargs = {})
    %gelu_10 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_62,), kwargs = {})
    %mul_169 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_61, %gelu_10), kwargs = {})
    %_param_constant486 : [#users=1] = get_attr[target=_param_constant486]
    %t_149 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant486,), kwargs = {})
    %view_393 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_169, [2048, 2560]), kwargs = {})
    %_param_constant487 : [#users=1] = get_attr[target=_param_constant487]
    %addmm_72 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant487, %view_393, %t_149), kwargs = {})
    %view_394 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_72, [2, 1024, 640]), kwargs = {})
    %add_243 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_394, %add_240), kwargs = {})
    %_param_constant488 : [#users=1] = get_attr[target=_param_constant488]
    %t_150 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant488,), kwargs = {})
    %view_395 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_243, [2048, 640]), kwargs = {})
    %_param_constant489 : [#users=1] = get_attr[target=_param_constant489]
    %addmm_73 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant489, %view_395, %t_150), kwargs = {})
    %view_396 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_73, [2, 1024, 640]), kwargs = {})
    %view_397 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_396, [2, 32, 32, 640]), kwargs = {})
    %permute_109 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_397, [0, 3, 1, 2]), kwargs = {})
    %clone_98 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_109,), kwargs = {memory_format: torch.contiguous_format})
    %add_244 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%clone_98, %div_17), kwargs = {})
    %cat_10 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%add_244, %add_56], 1), kwargs = {})
    %view_398 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_10, [2, 32, 40, 1024]), kwargs = {})
    %var_mean_78 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_398, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_156 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_78, 0), kwargs = {})
    %getitem_157 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_78, 1), kwargs = {})
    %add_245 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_156, 1e-05), kwargs = {})
    %rsqrt_78 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_245,), kwargs = {})
    %sub_78 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_398, %getitem_157), kwargs = {})
    %mul_170 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_78, %rsqrt_78), kwargs = {})
    %view_399 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_170, [2, 1280, 32, 32]), kwargs = {})
    %_param_constant490 : [#users=1] = get_attr[target=_param_constant490]
    %unsqueeze_306 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant490, 0), kwargs = {})
    %unsqueeze_307 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_306, 2), kwargs = {})
    %unsqueeze_308 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_307, 3), kwargs = {})
    %_param_constant491 : [#users=1] = get_attr[target=_param_constant491]
    %unsqueeze_309 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant491, 0), kwargs = {})
    %unsqueeze_310 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_309, 2), kwargs = {})
    %unsqueeze_311 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_310, 3), kwargs = {})
    %mul_171 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_399, %unsqueeze_311), kwargs = {})
    %add_246 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_171, %unsqueeze_308), kwargs = {})
    %squeeze_180 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_157, 3), kwargs = {})
    %squeeze_181 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_180, 2), kwargs = {})
    %squeeze_182 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_78, 3), kwargs = {})
    %squeeze_183 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_182, 2), kwargs = {})
    %detach_112 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_181,), kwargs = {})
    %detach_113 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_183,), kwargs = {})
    %silu_52 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_246,), kwargs = {})
    %_param_constant492 : [#users=1] = get_attr[target=_param_constant492]
    %_param_constant493 : [#users=1] = get_attr[target=_param_constant493]
    %convolution_49 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_52, %_param_constant492, %_param_constant493, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_53 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant494 : [#users=1] = get_attr[target=_param_constant494]
    %t_151 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant494,), kwargs = {})
    %_param_constant495 : [#users=1] = get_attr[target=_param_constant495]
    %addmm_74 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant495, %silu_53, %t_151), kwargs = {})
    %slice_63 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_74, 0, 0, 9223372036854775807), kwargs = {})
    %slice_64 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_63, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_312 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_64, 2), kwargs = {})
    %unsqueeze_313 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_312, 3), kwargs = {})
    %add_247 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_49, %unsqueeze_313), kwargs = {})
    %view_400 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_247, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_79 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_400, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_158 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_79, 0), kwargs = {})
    %getitem_159 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_79, 1), kwargs = {})
    %add_248 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_158, 1e-05), kwargs = {})
    %rsqrt_79 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_248,), kwargs = {})
    %sub_79 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_400, %getitem_159), kwargs = {})
    %mul_172 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_79, %rsqrt_79), kwargs = {})
    %view_401 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_172, [2, 640, 32, 32]), kwargs = {})
    %_param_constant496 : [#users=1] = get_attr[target=_param_constant496]
    %unsqueeze_314 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant496, 0), kwargs = {})
    %unsqueeze_315 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_314, 2), kwargs = {})
    %unsqueeze_316 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_315, 3), kwargs = {})
    %_param_constant497 : [#users=1] = get_attr[target=_param_constant497]
    %unsqueeze_317 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant497, 0), kwargs = {})
    %unsqueeze_318 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_317, 2), kwargs = {})
    %unsqueeze_319 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_318, 3), kwargs = {})
    %mul_173 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_401, %unsqueeze_319), kwargs = {})
    %add_249 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_173, %unsqueeze_316), kwargs = {})
    %squeeze_184 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_159, 3), kwargs = {})
    %squeeze_185 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_184, 2), kwargs = {})
    %squeeze_186 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_79, 3), kwargs = {})
    %squeeze_187 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_186, 2), kwargs = {})
    %detach_114 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_185,), kwargs = {})
    %detach_115 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_187,), kwargs = {})
    %silu_54 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_249,), kwargs = {})
    %_param_constant498 : [#users=1] = get_attr[target=_param_constant498]
    %_param_constant499 : [#users=1] = get_attr[target=_param_constant499]
    %convolution_50 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_54, %_param_constant498, %_param_constant499, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant500 : [#users=1] = get_attr[target=_param_constant500]
    %_param_constant501 : [#users=1] = get_attr[target=_param_constant501]
    %convolution_51 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_10, %_param_constant500, %_param_constant501, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_250 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_51, %convolution_50), kwargs = {})
    %div_18 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_250, 1.0), kwargs = {})
    %view_402 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_18, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_80 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_402, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_160 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_80, 0), kwargs = {})
    %getitem_161 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_80, 1), kwargs = {})
    %add_251 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_160, 1e-06), kwargs = {})
    %rsqrt_80 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_251,), kwargs = {})
    %sub_80 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_402, %getitem_161), kwargs = {})
    %mul_174 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_80, %rsqrt_80), kwargs = {})
    %view_403 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_174, [2, 640, 32, 32]), kwargs = {})
    %_param_constant502 : [#users=1] = get_attr[target=_param_constant502]
    %unsqueeze_320 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant502, 0), kwargs = {})
    %unsqueeze_321 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_320, 2), kwargs = {})
    %unsqueeze_322 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_321, 3), kwargs = {})
    %_param_constant503 : [#users=1] = get_attr[target=_param_constant503]
    %unsqueeze_323 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant503, 0), kwargs = {})
    %unsqueeze_324 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_323, 2), kwargs = {})
    %unsqueeze_325 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_324, 3), kwargs = {})
    %mul_175 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_403, %unsqueeze_325), kwargs = {})
    %add_252 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_175, %unsqueeze_322), kwargs = {})
    %squeeze_188 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_161, 3), kwargs = {})
    %squeeze_189 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_188, 2), kwargs = {})
    %squeeze_190 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_80, 3), kwargs = {})
    %squeeze_191 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_190, 2), kwargs = {})
    %detach_116 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_189,), kwargs = {})
    %detach_117 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_191,), kwargs = {})
    %permute_110 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_252, [0, 2, 3, 1]), kwargs = {})
    %view_404 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_110, [2, 1024, 640]), kwargs = {})
    %_param_constant504 : [#users=1] = get_attr[target=_param_constant504]
    %t_152 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant504,), kwargs = {})
    %expand_23 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_404, [2, 1024, 640]), kwargs = {})
    %view_405 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_23, [2, 1024, 640]), kwargs = {})
    %expand_24 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_152, [2, 640, 640]), kwargs = {})
    %view_406 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_24, [2, 640, 640]), kwargs = {})
    %bmm_33 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_405, %view_406), kwargs = {})
    %_unsafe_view_165 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_33, [2, 1024, 640]), kwargs = {})
    %_param_constant505 : [#users=1] = get_attr[target=_param_constant505]
    %add_253 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_165, %_param_constant505), kwargs = {})
    %var_mean_81 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_253, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_162 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_81, 0), kwargs = {})
    %getitem_163 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_81, 1), kwargs = {})
    %add_254 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_162, 1e-05), kwargs = {})
    %rsqrt_81 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_254,), kwargs = {})
    %sub_81 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_253, %getitem_163), kwargs = {})
    %mul_176 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_81, %rsqrt_81), kwargs = {})
    %_param_constant506 : [#users=1] = get_attr[target=_param_constant506]
    %mul_177 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_176, %_param_constant506), kwargs = {})
    %_param_constant507 : [#users=1] = get_attr[target=_param_constant507]
    %add_255 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_177, %_param_constant507), kwargs = {})
    %_param_constant508 : [#users=1] = get_attr[target=_param_constant508]
    %t_153 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant508,), kwargs = {})
    %view_407 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_255, [2048, 640]), kwargs = {})
    %mm_66 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_407, %t_153), kwargs = {})
    %_unsafe_view_166 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_66, [2, 1024, 640]), kwargs = {})
    %view_408 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_166, [2, 1024, 10, 64]), kwargs = {})
    %permute_111 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_408, [0, 2, 1, 3]), kwargs = {})
    %clone_99 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_111,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_167 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_99, [20, 1024, 64]), kwargs = {})
    %_param_constant509 : [#users=1] = get_attr[target=_param_constant509]
    %t_154 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant509,), kwargs = {})
    %view_409 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_255, [2048, 640]), kwargs = {})
    %mm_67 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_409, %t_154), kwargs = {})
    %_unsafe_view_168 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_67, [2, 1024, 640]), kwargs = {})
    %_param_constant510 : [#users=1] = get_attr[target=_param_constant510]
    %t_155 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant510,), kwargs = {})
    %view_410 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_255, [2048, 640]), kwargs = {})
    %mm_68 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_410, %t_155), kwargs = {})
    %_unsafe_view_169 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_68, [2, 1024, 640]), kwargs = {})
    %view_411 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_168, [2, 1024, 10, 64]), kwargs = {})
    %permute_112 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_411, [0, 2, 1, 3]), kwargs = {})
    %clone_100 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_112,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_170 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_100, [20, 1024, 64]), kwargs = {})
    %view_412 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_169, [2, 1024, 10, 64]), kwargs = {})
    %permute_113 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_412, [0, 2, 1, 3]), kwargs = {})
    %clone_101 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_113,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_171 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_101, [20, 1024, 64]), kwargs = {})
    %empty_22 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([20, 1024, 1024],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_22 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_170, -1, -2), kwargs = {})
    %baddbmm_22 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_22, %_unsafe_view_167, %transpose_22), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_22 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_22, -1, False), kwargs = {})
    %detach_118 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_22,), kwargs = {})
    %bmm_34 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_22, %_unsafe_view_171), kwargs = {})
    %view_413 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_34, [2, 10, 1024, 64]), kwargs = {})
    %permute_114 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_413, [0, 2, 1, 3]), kwargs = {})
    %clone_102 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_114,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_172 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_102, [2, 1024, 640]), kwargs = {})
    %_param_constant511 : [#users=1] = get_attr[target=_param_constant511]
    %t_156 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant511,), kwargs = {})
    %view_414 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_172, [2048, 640]), kwargs = {})
    %_param_constant512 : [#users=1] = get_attr[target=_param_constant512]
    %addmm_75 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant512, %view_414, %t_156), kwargs = {})
    %view_415 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_75, [2, 1024, 640]), kwargs = {})
    %add_256 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_415, %add_253), kwargs = {})
    %var_mean_82 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_256, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_164 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_82, 0), kwargs = {})
    %getitem_165 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_82, 1), kwargs = {})
    %add_257 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_164, 1e-05), kwargs = {})
    %rsqrt_82 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_257,), kwargs = {})
    %sub_82 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_256, %getitem_165), kwargs = {})
    %mul_178 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_82, %rsqrt_82), kwargs = {})
    %_param_constant513 : [#users=1] = get_attr[target=_param_constant513]
    %mul_179 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_178, %_param_constant513), kwargs = {})
    %_param_constant514 : [#users=1] = get_attr[target=_param_constant514]
    %add_258 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_179, %_param_constant514), kwargs = {})
    %_param_constant515 : [#users=1] = get_attr[target=_param_constant515]
    %t_157 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant515,), kwargs = {})
    %view_416 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_258, [2048, 640]), kwargs = {})
    %mm_69 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_416, %t_157), kwargs = {})
    %_unsafe_view_173 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_69, [2, 1024, 640]), kwargs = {})
    %view_417 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_173, [2, 1024, 10, 64]), kwargs = {})
    %permute_115 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_417, [0, 2, 1, 3]), kwargs = {})
    %clone_103 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_115,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_174 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_103, [20, 1024, 64]), kwargs = {})
    %_param_constant516 : [#users=1] = get_attr[target=_param_constant516]
    %t_158 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant516,), kwargs = {})
    %view_418 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_70 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_418, %t_158), kwargs = {})
    %_unsafe_view_175 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_70, [2, 64, 640]), kwargs = {})
    %_param_constant517 : [#users=1] = get_attr[target=_param_constant517]
    %t_159 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant517,), kwargs = {})
    %view_419 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_71 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_419, %t_159), kwargs = {})
    %_unsafe_view_176 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_71, [2, 64, 640]), kwargs = {})
    %view_420 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_175, [2, 64, 10, 64]), kwargs = {})
    %permute_116 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_420, [0, 2, 1, 3]), kwargs = {})
    %clone_104 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_116,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_177 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_104, [20, 64, 64]), kwargs = {})
    %view_421 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_176, [2, 64, 10, 64]), kwargs = {})
    %permute_117 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_421, [0, 2, 1, 3]), kwargs = {})
    %clone_105 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_117,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_178 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_105, [20, 64, 64]), kwargs = {})
    %empty_23 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([20, 1024, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_23 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_177, -1, -2), kwargs = {})
    %baddbmm_23 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_23, %_unsafe_view_174, %transpose_23), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_23 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_23, -1, False), kwargs = {})
    %detach_119 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_23,), kwargs = {})
    %bmm_35 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_23, %_unsafe_view_178), kwargs = {})
    %view_422 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_35, [2, 10, 1024, 64]), kwargs = {})
    %permute_118 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_422, [0, 2, 1, 3]), kwargs = {})
    %clone_106 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_118,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_179 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_106, [2, 1024, 640]), kwargs = {})
    %_param_constant518 : [#users=1] = get_attr[target=_param_constant518]
    %t_160 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant518,), kwargs = {})
    %view_423 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_179, [2048, 640]), kwargs = {})
    %_param_constant519 : [#users=1] = get_attr[target=_param_constant519]
    %addmm_76 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant519, %view_423, %t_160), kwargs = {})
    %view_424 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_76, [2, 1024, 640]), kwargs = {})
    %add_259 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_424, %add_256), kwargs = {})
    %var_mean_83 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_259, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_166 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_83, 0), kwargs = {})
    %getitem_167 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_83, 1), kwargs = {})
    %add_260 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_166, 1e-05), kwargs = {})
    %rsqrt_83 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_260,), kwargs = {})
    %sub_83 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_259, %getitem_167), kwargs = {})
    %mul_180 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_83, %rsqrt_83), kwargs = {})
    %_param_constant520 : [#users=1] = get_attr[target=_param_constant520]
    %mul_181 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_180, %_param_constant520), kwargs = {})
    %_param_constant521 : [#users=1] = get_attr[target=_param_constant521]
    %add_261 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_181, %_param_constant521), kwargs = {})
    %_param_constant522 : [#users=1] = get_attr[target=_param_constant522]
    %t_161 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant522,), kwargs = {})
    %view_425 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_261, [2048, 640]), kwargs = {})
    %_param_constant523 : [#users=1] = get_attr[target=_param_constant523]
    %addmm_77 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant523, %view_425, %t_161), kwargs = {})
    %view_426 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_77, [2, 1024, 5120]), kwargs = {})
    %slice_65 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_426, -1, 0, 2560), kwargs = {})
    %slice_66 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_426, -1, 2560, 5120), kwargs = {})
    %gelu_11 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_66,), kwargs = {})
    %mul_182 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_65, %gelu_11), kwargs = {})
    %_param_constant524 : [#users=1] = get_attr[target=_param_constant524]
    %t_162 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant524,), kwargs = {})
    %view_427 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_182, [2048, 2560]), kwargs = {})
    %_param_constant525 : [#users=1] = get_attr[target=_param_constant525]
    %addmm_78 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant525, %view_427, %t_162), kwargs = {})
    %view_428 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_78, [2, 1024, 640]), kwargs = {})
    %add_262 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_428, %add_259), kwargs = {})
    %_param_constant526 : [#users=1] = get_attr[target=_param_constant526]
    %t_163 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant526,), kwargs = {})
    %view_429 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_262, [2048, 640]), kwargs = {})
    %_param_constant527 : [#users=1] = get_attr[target=_param_constant527]
    %addmm_79 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant527, %view_429, %t_163), kwargs = {})
    %view_430 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_79, [2, 1024, 640]), kwargs = {})
    %view_431 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_430, [2, 32, 32, 640]), kwargs = {})
    %permute_119 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_431, [0, 3, 1, 2]), kwargs = {})
    %clone_107 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_119,), kwargs = {memory_format: torch.contiguous_format})
    %add_263 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%clone_107, %div_18), kwargs = {})
    %cat_11 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%add_263, %convolution_5], 1), kwargs = {})
    %view_432 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_11, [2, 32, 30, 1024]), kwargs = {})
    %var_mean_84 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_432, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_168 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_84, 0), kwargs = {})
    %getitem_169 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_84, 1), kwargs = {})
    %add_264 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_168, 1e-05), kwargs = {})
    %rsqrt_84 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_264,), kwargs = {})
    %sub_84 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_432, %getitem_169), kwargs = {})
    %mul_183 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_84, %rsqrt_84), kwargs = {})
    %view_433 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_183, [2, 960, 32, 32]), kwargs = {})
    %_param_constant528 : [#users=1] = get_attr[target=_param_constant528]
    %unsqueeze_326 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant528, 0), kwargs = {})
    %unsqueeze_327 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_326, 2), kwargs = {})
    %unsqueeze_328 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_327, 3), kwargs = {})
    %_param_constant529 : [#users=1] = get_attr[target=_param_constant529]
    %unsqueeze_329 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant529, 0), kwargs = {})
    %unsqueeze_330 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_329, 2), kwargs = {})
    %unsqueeze_331 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_330, 3), kwargs = {})
    %mul_184 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_433, %unsqueeze_331), kwargs = {})
    %add_265 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_184, %unsqueeze_328), kwargs = {})
    %squeeze_192 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_169, 3), kwargs = {})
    %squeeze_193 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_192, 2), kwargs = {})
    %squeeze_194 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_84, 3), kwargs = {})
    %squeeze_195 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_194, 2), kwargs = {})
    %detach_120 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_193,), kwargs = {})
    %detach_121 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_195,), kwargs = {})
    %silu_55 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_265,), kwargs = {})
    %_param_constant530 : [#users=1] = get_attr[target=_param_constant530]
    %_param_constant531 : [#users=1] = get_attr[target=_param_constant531]
    %convolution_52 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_55, %_param_constant530, %_param_constant531, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_56 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant532 : [#users=1] = get_attr[target=_param_constant532]
    %t_164 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant532,), kwargs = {})
    %_param_constant533 : [#users=1] = get_attr[target=_param_constant533]
    %addmm_80 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant533, %silu_56, %t_164), kwargs = {})
    %slice_67 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_80, 0, 0, 9223372036854775807), kwargs = {})
    %slice_68 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_67, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_332 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_68, 2), kwargs = {})
    %unsqueeze_333 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_332, 3), kwargs = {})
    %add_266 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_52, %unsqueeze_333), kwargs = {})
    %view_434 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_266, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_85 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_434, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_170 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_85, 0), kwargs = {})
    %getitem_171 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_85, 1), kwargs = {})
    %add_267 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_170, 1e-05), kwargs = {})
    %rsqrt_85 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_267,), kwargs = {})
    %sub_85 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_434, %getitem_171), kwargs = {})
    %mul_185 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_85, %rsqrt_85), kwargs = {})
    %view_435 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_185, [2, 640, 32, 32]), kwargs = {})
    %_param_constant534 : [#users=1] = get_attr[target=_param_constant534]
    %unsqueeze_334 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant534, 0), kwargs = {})
    %unsqueeze_335 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_334, 2), kwargs = {})
    %unsqueeze_336 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_335, 3), kwargs = {})
    %_param_constant535 : [#users=1] = get_attr[target=_param_constant535]
    %unsqueeze_337 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant535, 0), kwargs = {})
    %unsqueeze_338 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_337, 2), kwargs = {})
    %unsqueeze_339 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_338, 3), kwargs = {})
    %mul_186 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_435, %unsqueeze_339), kwargs = {})
    %add_268 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_186, %unsqueeze_336), kwargs = {})
    %squeeze_196 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_171, 3), kwargs = {})
    %squeeze_197 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_196, 2), kwargs = {})
    %squeeze_198 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_85, 3), kwargs = {})
    %squeeze_199 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_198, 2), kwargs = {})
    %detach_122 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_197,), kwargs = {})
    %detach_123 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_199,), kwargs = {})
    %silu_57 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_268,), kwargs = {})
    %_param_constant536 : [#users=1] = get_attr[target=_param_constant536]
    %_param_constant537 : [#users=1] = get_attr[target=_param_constant537]
    %convolution_53 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_57, %_param_constant536, %_param_constant537, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant538 : [#users=1] = get_attr[target=_param_constant538]
    %_param_constant539 : [#users=1] = get_attr[target=_param_constant539]
    %convolution_54 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_11, %_param_constant538, %_param_constant539, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_269 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_54, %convolution_53), kwargs = {})
    %div_19 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_269, 1.0), kwargs = {})
    %view_436 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_19, [2, 32, 20, 1024]), kwargs = {})
    %var_mean_86 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_436, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_172 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_86, 0), kwargs = {})
    %getitem_173 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_86, 1), kwargs = {})
    %add_270 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_172, 1e-06), kwargs = {})
    %rsqrt_86 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_270,), kwargs = {})
    %sub_86 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_436, %getitem_173), kwargs = {})
    %mul_187 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_86, %rsqrt_86), kwargs = {})
    %view_437 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_187, [2, 640, 32, 32]), kwargs = {})
    %_param_constant540 : [#users=1] = get_attr[target=_param_constant540]
    %unsqueeze_340 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant540, 0), kwargs = {})
    %unsqueeze_341 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_340, 2), kwargs = {})
    %unsqueeze_342 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_341, 3), kwargs = {})
    %_param_constant541 : [#users=1] = get_attr[target=_param_constant541]
    %unsqueeze_343 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant541, 0), kwargs = {})
    %unsqueeze_344 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_343, 2), kwargs = {})
    %unsqueeze_345 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_344, 3), kwargs = {})
    %mul_188 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_437, %unsqueeze_345), kwargs = {})
    %add_271 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_188, %unsqueeze_342), kwargs = {})
    %squeeze_200 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_173, 3), kwargs = {})
    %squeeze_201 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_200, 2), kwargs = {})
    %squeeze_202 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_86, 3), kwargs = {})
    %squeeze_203 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_202, 2), kwargs = {})
    %detach_124 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_201,), kwargs = {})
    %detach_125 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_203,), kwargs = {})
    %permute_120 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_271, [0, 2, 3, 1]), kwargs = {})
    %view_438 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_120, [2, 1024, 640]), kwargs = {})
    %_param_constant542 : [#users=1] = get_attr[target=_param_constant542]
    %t_165 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant542,), kwargs = {})
    %expand_25 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_438, [2, 1024, 640]), kwargs = {})
    %view_439 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_25, [2, 1024, 640]), kwargs = {})
    %expand_26 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_165, [2, 640, 640]), kwargs = {})
    %view_440 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_26, [2, 640, 640]), kwargs = {})
    %bmm_36 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_439, %view_440), kwargs = {})
    %_unsafe_view_180 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_36, [2, 1024, 640]), kwargs = {})
    %_param_constant543 : [#users=1] = get_attr[target=_param_constant543]
    %add_272 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_180, %_param_constant543), kwargs = {})
    %var_mean_87 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_272, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_174 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_87, 0), kwargs = {})
    %getitem_175 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_87, 1), kwargs = {})
    %add_273 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_174, 1e-05), kwargs = {})
    %rsqrt_87 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_273,), kwargs = {})
    %sub_87 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_272, %getitem_175), kwargs = {})
    %mul_189 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_87, %rsqrt_87), kwargs = {})
    %_param_constant544 : [#users=1] = get_attr[target=_param_constant544]
    %mul_190 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_189, %_param_constant544), kwargs = {})
    %_param_constant545 : [#users=1] = get_attr[target=_param_constant545]
    %add_274 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_190, %_param_constant545), kwargs = {})
    %_param_constant546 : [#users=1] = get_attr[target=_param_constant546]
    %t_166 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant546,), kwargs = {})
    %view_441 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_274, [2048, 640]), kwargs = {})
    %mm_72 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_441, %t_166), kwargs = {})
    %_unsafe_view_181 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_72, [2, 1024, 640]), kwargs = {})
    %view_442 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_181, [2, 1024, 10, 64]), kwargs = {})
    %permute_121 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_442, [0, 2, 1, 3]), kwargs = {})
    %clone_108 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_121,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_182 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_108, [20, 1024, 64]), kwargs = {})
    %_param_constant547 : [#users=1] = get_attr[target=_param_constant547]
    %t_167 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant547,), kwargs = {})
    %view_443 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_274, [2048, 640]), kwargs = {})
    %mm_73 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_443, %t_167), kwargs = {})
    %_unsafe_view_183 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_73, [2, 1024, 640]), kwargs = {})
    %_param_constant548 : [#users=1] = get_attr[target=_param_constant548]
    %t_168 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant548,), kwargs = {})
    %view_444 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_274, [2048, 640]), kwargs = {})
    %mm_74 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_444, %t_168), kwargs = {})
    %_unsafe_view_184 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_74, [2, 1024, 640]), kwargs = {})
    %view_445 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_183, [2, 1024, 10, 64]), kwargs = {})
    %permute_122 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_445, [0, 2, 1, 3]), kwargs = {})
    %clone_109 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_122,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_185 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_109, [20, 1024, 64]), kwargs = {})
    %view_446 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_184, [2, 1024, 10, 64]), kwargs = {})
    %permute_123 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_446, [0, 2, 1, 3]), kwargs = {})
    %clone_110 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_123,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_186 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_110, [20, 1024, 64]), kwargs = {})
    %empty_24 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([20, 1024, 1024],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_24 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_185, -1, -2), kwargs = {})
    %baddbmm_24 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_24, %_unsafe_view_182, %transpose_24), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_24 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_24, -1, False), kwargs = {})
    %detach_126 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_24,), kwargs = {})
    %bmm_37 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_24, %_unsafe_view_186), kwargs = {})
    %view_447 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_37, [2, 10, 1024, 64]), kwargs = {})
    %permute_124 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_447, [0, 2, 1, 3]), kwargs = {})
    %clone_111 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_124,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_187 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_111, [2, 1024, 640]), kwargs = {})
    %_param_constant549 : [#users=1] = get_attr[target=_param_constant549]
    %t_169 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant549,), kwargs = {})
    %view_448 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_187, [2048, 640]), kwargs = {})
    %_param_constant550 : [#users=1] = get_attr[target=_param_constant550]
    %addmm_81 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant550, %view_448, %t_169), kwargs = {})
    %view_449 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_81, [2, 1024, 640]), kwargs = {})
    %add_275 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_449, %add_272), kwargs = {})
    %var_mean_88 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_275, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_176 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_88, 0), kwargs = {})
    %getitem_177 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_88, 1), kwargs = {})
    %add_276 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_176, 1e-05), kwargs = {})
    %rsqrt_88 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_276,), kwargs = {})
    %sub_88 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_275, %getitem_177), kwargs = {})
    %mul_191 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_88, %rsqrt_88), kwargs = {})
    %_param_constant551 : [#users=1] = get_attr[target=_param_constant551]
    %mul_192 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_191, %_param_constant551), kwargs = {})
    %_param_constant552 : [#users=1] = get_attr[target=_param_constant552]
    %add_277 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_192, %_param_constant552), kwargs = {})
    %_param_constant553 : [#users=1] = get_attr[target=_param_constant553]
    %t_170 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant553,), kwargs = {})
    %view_450 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_277, [2048, 640]), kwargs = {})
    %mm_75 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_450, %t_170), kwargs = {})
    %_unsafe_view_188 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_75, [2, 1024, 640]), kwargs = {})
    %view_451 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_188, [2, 1024, 10, 64]), kwargs = {})
    %permute_125 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_451, [0, 2, 1, 3]), kwargs = {})
    %clone_112 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_125,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_189 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_112, [20, 1024, 64]), kwargs = {})
    %_param_constant554 : [#users=1] = get_attr[target=_param_constant554]
    %t_171 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant554,), kwargs = {})
    %view_452 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_76 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_452, %t_171), kwargs = {})
    %_unsafe_view_190 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_76, [2, 64, 640]), kwargs = {})
    %_param_constant555 : [#users=1] = get_attr[target=_param_constant555]
    %t_172 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant555,), kwargs = {})
    %view_453 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_77 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_453, %t_172), kwargs = {})
    %_unsafe_view_191 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_77, [2, 64, 640]), kwargs = {})
    %view_454 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_190, [2, 64, 10, 64]), kwargs = {})
    %permute_126 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_454, [0, 2, 1, 3]), kwargs = {})
    %clone_113 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_126,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_192 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_113, [20, 64, 64]), kwargs = {})
    %view_455 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_191, [2, 64, 10, 64]), kwargs = {})
    %permute_127 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_455, [0, 2, 1, 3]), kwargs = {})
    %clone_114 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_127,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_193 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_114, [20, 64, 64]), kwargs = {})
    %empty_25 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([20, 1024, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_25 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_192, -1, -2), kwargs = {})
    %baddbmm_25 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_25, %_unsafe_view_189, %transpose_25), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_25 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_25, -1, False), kwargs = {})
    %detach_127 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_25,), kwargs = {})
    %bmm_38 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_25, %_unsafe_view_193), kwargs = {})
    %view_456 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_38, [2, 10, 1024, 64]), kwargs = {})
    %permute_128 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_456, [0, 2, 1, 3]), kwargs = {})
    %clone_115 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_128,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_194 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_115, [2, 1024, 640]), kwargs = {})
    %_param_constant556 : [#users=1] = get_attr[target=_param_constant556]
    %t_173 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant556,), kwargs = {})
    %view_457 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_194, [2048, 640]), kwargs = {})
    %_param_constant557 : [#users=1] = get_attr[target=_param_constant557]
    %addmm_82 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant557, %view_457, %t_173), kwargs = {})
    %view_458 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_82, [2, 1024, 640]), kwargs = {})
    %add_278 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_458, %add_275), kwargs = {})
    %var_mean_89 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_278, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_178 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_89, 0), kwargs = {})
    %getitem_179 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_89, 1), kwargs = {})
    %add_279 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_178, 1e-05), kwargs = {})
    %rsqrt_89 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_279,), kwargs = {})
    %sub_89 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_278, %getitem_179), kwargs = {})
    %mul_193 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_89, %rsqrt_89), kwargs = {})
    %_param_constant558 : [#users=1] = get_attr[target=_param_constant558]
    %mul_194 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_193, %_param_constant558), kwargs = {})
    %_param_constant559 : [#users=1] = get_attr[target=_param_constant559]
    %add_280 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_194, %_param_constant559), kwargs = {})
    %_param_constant560 : [#users=1] = get_attr[target=_param_constant560]
    %t_174 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant560,), kwargs = {})
    %view_459 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_280, [2048, 640]), kwargs = {})
    %_param_constant561 : [#users=1] = get_attr[target=_param_constant561]
    %addmm_83 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant561, %view_459, %t_174), kwargs = {})
    %view_460 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_83, [2, 1024, 5120]), kwargs = {})
    %slice_69 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_460, -1, 0, 2560), kwargs = {})
    %slice_70 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_460, -1, 2560, 5120), kwargs = {})
    %gelu_12 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_70,), kwargs = {})
    %mul_195 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_69, %gelu_12), kwargs = {})
    %_param_constant562 : [#users=1] = get_attr[target=_param_constant562]
    %t_175 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant562,), kwargs = {})
    %view_461 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_195, [2048, 2560]), kwargs = {})
    %_param_constant563 : [#users=1] = get_attr[target=_param_constant563]
    %addmm_84 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant563, %view_461, %t_175), kwargs = {})
    %view_462 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_84, [2, 1024, 640]), kwargs = {})
    %add_281 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_462, %add_278), kwargs = {})
    %_param_constant564 : [#users=1] = get_attr[target=_param_constant564]
    %t_176 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant564,), kwargs = {})
    %view_463 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_281, [2048, 640]), kwargs = {})
    %_param_constant565 : [#users=1] = get_attr[target=_param_constant565]
    %addmm_85 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant565, %view_463, %t_176), kwargs = {})
    %view_464 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_85, [2, 1024, 640]), kwargs = {})
    %view_465 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_464, [2, 32, 32, 640]), kwargs = {})
    %permute_129 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_465, [0, 3, 1, 2]), kwargs = {})
    %clone_116 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_129,), kwargs = {memory_format: torch.contiguous_format})
    %add_282 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%clone_116, %div_19), kwargs = {})
    %upsample_nearest2d_2 : [#users=1] = call_function[target=torch.ops.aten.upsample_nearest2d](args = (%add_282, [64, 64], 2.0, 2.0), kwargs = {})
    %_param_constant566 : [#users=1] = get_attr[target=_param_constant566]
    %_param_constant567 : [#users=1] = get_attr[target=_param_constant567]
    %convolution_55 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%upsample_nearest2d_2, %_param_constant566, %_param_constant567, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %cat_12 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%convolution_55, %add_37], 1), kwargs = {})
    %view_466 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_12, [2, 32, 30, 4096]), kwargs = {})
    %var_mean_90 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_466, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_180 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_90, 0), kwargs = {})
    %getitem_181 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_90, 1), kwargs = {})
    %add_283 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_180, 1e-05), kwargs = {})
    %rsqrt_90 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_283,), kwargs = {})
    %sub_90 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_466, %getitem_181), kwargs = {})
    %mul_196 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_90, %rsqrt_90), kwargs = {})
    %view_467 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_196, [2, 960, 64, 64]), kwargs = {})
    %_param_constant568 : [#users=1] = get_attr[target=_param_constant568]
    %unsqueeze_346 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant568, 0), kwargs = {})
    %unsqueeze_347 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_346, 2), kwargs = {})
    %unsqueeze_348 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_347, 3), kwargs = {})
    %_param_constant569 : [#users=1] = get_attr[target=_param_constant569]
    %unsqueeze_349 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant569, 0), kwargs = {})
    %unsqueeze_350 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_349, 2), kwargs = {})
    %unsqueeze_351 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_350, 3), kwargs = {})
    %mul_197 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_467, %unsqueeze_351), kwargs = {})
    %add_284 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_197, %unsqueeze_348), kwargs = {})
    %squeeze_204 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_181, 3), kwargs = {})
    %squeeze_205 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_204, 2), kwargs = {})
    %squeeze_206 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_90, 3), kwargs = {})
    %squeeze_207 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_206, 2), kwargs = {})
    %detach_128 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_205,), kwargs = {})
    %detach_129 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_207,), kwargs = {})
    %silu_58 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_284,), kwargs = {})
    %_param_constant570 : [#users=1] = get_attr[target=_param_constant570]
    %_param_constant571 : [#users=1] = get_attr[target=_param_constant571]
    %convolution_56 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_58, %_param_constant570, %_param_constant571, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_59 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant572 : [#users=1] = get_attr[target=_param_constant572]
    %t_177 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant572,), kwargs = {})
    %_param_constant573 : [#users=1] = get_attr[target=_param_constant573]
    %addmm_86 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant573, %silu_59, %t_177), kwargs = {})
    %slice_71 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_86, 0, 0, 9223372036854775807), kwargs = {})
    %slice_72 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_71, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_352 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_72, 2), kwargs = {})
    %unsqueeze_353 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_352, 3), kwargs = {})
    %add_285 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_56, %unsqueeze_353), kwargs = {})
    %view_468 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_285, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_91 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_468, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_182 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_91, 0), kwargs = {})
    %getitem_183 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_91, 1), kwargs = {})
    %add_286 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_182, 1e-05), kwargs = {})
    %rsqrt_91 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_286,), kwargs = {})
    %sub_91 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_468, %getitem_183), kwargs = {})
    %mul_198 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_91, %rsqrt_91), kwargs = {})
    %view_469 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_198, [2, 320, 64, 64]), kwargs = {})
    %_param_constant574 : [#users=1] = get_attr[target=_param_constant574]
    %unsqueeze_354 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant574, 0), kwargs = {})
    %unsqueeze_355 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_354, 2), kwargs = {})
    %unsqueeze_356 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_355, 3), kwargs = {})
    %_param_constant575 : [#users=1] = get_attr[target=_param_constant575]
    %unsqueeze_357 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant575, 0), kwargs = {})
    %unsqueeze_358 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_357, 2), kwargs = {})
    %unsqueeze_359 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_358, 3), kwargs = {})
    %mul_199 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_469, %unsqueeze_359), kwargs = {})
    %add_287 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_199, %unsqueeze_356), kwargs = {})
    %squeeze_208 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_183, 3), kwargs = {})
    %squeeze_209 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_208, 2), kwargs = {})
    %squeeze_210 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_91, 3), kwargs = {})
    %squeeze_211 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_210, 2), kwargs = {})
    %detach_130 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_209,), kwargs = {})
    %detach_131 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_211,), kwargs = {})
    %silu_60 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_287,), kwargs = {})
    %_param_constant576 : [#users=1] = get_attr[target=_param_constant576]
    %_param_constant577 : [#users=1] = get_attr[target=_param_constant577]
    %convolution_57 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_60, %_param_constant576, %_param_constant577, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant578 : [#users=1] = get_attr[target=_param_constant578]
    %_param_constant579 : [#users=1] = get_attr[target=_param_constant579]
    %convolution_58 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_12, %_param_constant578, %_param_constant579, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_288 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_58, %convolution_57), kwargs = {})
    %div_20 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_288, 1.0), kwargs = {})
    %view_470 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_20, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_92 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_470, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_184 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_92, 0), kwargs = {})
    %getitem_185 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_92, 1), kwargs = {})
    %add_289 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_184, 1e-06), kwargs = {})
    %rsqrt_92 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_289,), kwargs = {})
    %sub_92 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_470, %getitem_185), kwargs = {})
    %mul_200 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_92, %rsqrt_92), kwargs = {})
    %view_471 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_200, [2, 320, 64, 64]), kwargs = {})
    %_param_constant580 : [#users=1] = get_attr[target=_param_constant580]
    %unsqueeze_360 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant580, 0), kwargs = {})
    %unsqueeze_361 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_360, 2), kwargs = {})
    %unsqueeze_362 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_361, 3), kwargs = {})
    %_param_constant581 : [#users=1] = get_attr[target=_param_constant581]
    %unsqueeze_363 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant581, 0), kwargs = {})
    %unsqueeze_364 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_363, 2), kwargs = {})
    %unsqueeze_365 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_364, 3), kwargs = {})
    %mul_201 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_471, %unsqueeze_365), kwargs = {})
    %add_290 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_201, %unsqueeze_362), kwargs = {})
    %squeeze_212 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_185, 3), kwargs = {})
    %squeeze_213 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_212, 2), kwargs = {})
    %squeeze_214 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_92, 3), kwargs = {})
    %squeeze_215 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_214, 2), kwargs = {})
    %detach_132 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_213,), kwargs = {})
    %detach_133 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_215,), kwargs = {})
    %permute_130 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_290, [0, 2, 3, 1]), kwargs = {})
    %view_472 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_130, [2, 4096, 320]), kwargs = {})
    %_param_constant582 : [#users=1] = get_attr[target=_param_constant582]
    %t_178 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant582,), kwargs = {})
    %expand_27 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_472, [2, 4096, 320]), kwargs = {})
    %view_473 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_27, [2, 4096, 320]), kwargs = {})
    %expand_28 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_178, [2, 320, 320]), kwargs = {})
    %view_474 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_28, [2, 320, 320]), kwargs = {})
    %bmm_39 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_473, %view_474), kwargs = {})
    %_unsafe_view_195 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_39, [2, 4096, 320]), kwargs = {})
    %_param_constant583 : [#users=1] = get_attr[target=_param_constant583]
    %add_291 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_195, %_param_constant583), kwargs = {})
    %var_mean_93 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_291, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_186 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_93, 0), kwargs = {})
    %getitem_187 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_93, 1), kwargs = {})
    %add_292 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_186, 1e-05), kwargs = {})
    %rsqrt_93 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_292,), kwargs = {})
    %sub_93 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_291, %getitem_187), kwargs = {})
    %mul_202 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_93, %rsqrt_93), kwargs = {})
    %_param_constant584 : [#users=1] = get_attr[target=_param_constant584]
    %mul_203 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_202, %_param_constant584), kwargs = {})
    %_param_constant585 : [#users=1] = get_attr[target=_param_constant585]
    %add_293 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_203, %_param_constant585), kwargs = {})
    %_param_constant586 : [#users=1] = get_attr[target=_param_constant586]
    %t_179 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant586,), kwargs = {})
    %view_475 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_293, [8192, 320]), kwargs = {})
    %mm_78 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_475, %t_179), kwargs = {})
    %_unsafe_view_196 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_78, [2, 4096, 320]), kwargs = {})
    %view_476 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_196, [2, 4096, 5, 64]), kwargs = {})
    %permute_131 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_476, [0, 2, 1, 3]), kwargs = {})
    %clone_117 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_131,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_197 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_117, [10, 4096, 64]), kwargs = {})
    %_param_constant587 : [#users=1] = get_attr[target=_param_constant587]
    %t_180 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant587,), kwargs = {})
    %view_477 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_293, [8192, 320]), kwargs = {})
    %mm_79 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_477, %t_180), kwargs = {})
    %_unsafe_view_198 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_79, [2, 4096, 320]), kwargs = {})
    %_param_constant588 : [#users=1] = get_attr[target=_param_constant588]
    %t_181 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant588,), kwargs = {})
    %view_478 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_293, [8192, 320]), kwargs = {})
    %mm_80 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_478, %t_181), kwargs = {})
    %_unsafe_view_199 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_80, [2, 4096, 320]), kwargs = {})
    %view_479 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_198, [2, 4096, 5, 64]), kwargs = {})
    %permute_132 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_479, [0, 2, 1, 3]), kwargs = {})
    %clone_118 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_132,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_200 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_118, [10, 4096, 64]), kwargs = {})
    %view_480 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_199, [2, 4096, 5, 64]), kwargs = {})
    %permute_133 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_480, [0, 2, 1, 3]), kwargs = {})
    %clone_119 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_133,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_201 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_119, [10, 4096, 64]), kwargs = {})
    %empty_26 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([10, 4096, 4096],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_26 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_200, -1, -2), kwargs = {})
    %baddbmm_26 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_26, %_unsafe_view_197, %transpose_26), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_26 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_26, -1, False), kwargs = {})
    %detach_134 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_26,), kwargs = {})
    %bmm_40 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_26, %_unsafe_view_201), kwargs = {})
    %view_481 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_40, [2, 5, 4096, 64]), kwargs = {})
    %permute_134 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_481, [0, 2, 1, 3]), kwargs = {})
    %clone_120 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_134,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_202 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_120, [2, 4096, 320]), kwargs = {})
    %_param_constant589 : [#users=1] = get_attr[target=_param_constant589]
    %t_182 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant589,), kwargs = {})
    %view_482 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_202, [8192, 320]), kwargs = {})
    %_param_constant590 : [#users=1] = get_attr[target=_param_constant590]
    %addmm_87 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant590, %view_482, %t_182), kwargs = {})
    %view_483 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_87, [2, 4096, 320]), kwargs = {})
    %add_294 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_483, %add_291), kwargs = {})
    %var_mean_94 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_294, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_188 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_94, 0), kwargs = {})
    %getitem_189 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_94, 1), kwargs = {})
    %add_295 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_188, 1e-05), kwargs = {})
    %rsqrt_94 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_295,), kwargs = {})
    %sub_94 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_294, %getitem_189), kwargs = {})
    %mul_204 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_94, %rsqrt_94), kwargs = {})
    %_param_constant591 : [#users=1] = get_attr[target=_param_constant591]
    %mul_205 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_204, %_param_constant591), kwargs = {})
    %_param_constant592 : [#users=1] = get_attr[target=_param_constant592]
    %add_296 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_205, %_param_constant592), kwargs = {})
    %_param_constant593 : [#users=1] = get_attr[target=_param_constant593]
    %t_183 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant593,), kwargs = {})
    %view_484 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_296, [8192, 320]), kwargs = {})
    %mm_81 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_484, %t_183), kwargs = {})
    %_unsafe_view_203 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_81, [2, 4096, 320]), kwargs = {})
    %view_485 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_203, [2, 4096, 5, 64]), kwargs = {})
    %permute_135 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_485, [0, 2, 1, 3]), kwargs = {})
    %clone_121 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_135,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_204 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_121, [10, 4096, 64]), kwargs = {})
    %_param_constant594 : [#users=1] = get_attr[target=_param_constant594]
    %t_184 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant594,), kwargs = {})
    %view_486 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_82 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_486, %t_184), kwargs = {})
    %_unsafe_view_205 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_82, [2, 64, 320]), kwargs = {})
    %_param_constant595 : [#users=1] = get_attr[target=_param_constant595]
    %t_185 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant595,), kwargs = {})
    %view_487 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_83 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_487, %t_185), kwargs = {})
    %_unsafe_view_206 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_83, [2, 64, 320]), kwargs = {})
    %view_488 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_205, [2, 64, 5, 64]), kwargs = {})
    %permute_136 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_488, [0, 2, 1, 3]), kwargs = {})
    %clone_122 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_136,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_207 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_122, [10, 64, 64]), kwargs = {})
    %view_489 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_206, [2, 64, 5, 64]), kwargs = {})
    %permute_137 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_489, [0, 2, 1, 3]), kwargs = {})
    %clone_123 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_137,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_208 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_123, [10, 64, 64]), kwargs = {})
    %empty_27 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([10, 4096, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_27 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_207, -1, -2), kwargs = {})
    %baddbmm_27 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_27, %_unsafe_view_204, %transpose_27), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_27 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_27, -1, False), kwargs = {})
    %detach_135 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_27,), kwargs = {})
    %bmm_41 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_27, %_unsafe_view_208), kwargs = {})
    %view_490 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_41, [2, 5, 4096, 64]), kwargs = {})
    %permute_138 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_490, [0, 2, 1, 3]), kwargs = {})
    %clone_124 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_138,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_209 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_124, [2, 4096, 320]), kwargs = {})
    %_param_constant596 : [#users=1] = get_attr[target=_param_constant596]
    %t_186 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant596,), kwargs = {})
    %view_491 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_209, [8192, 320]), kwargs = {})
    %_param_constant597 : [#users=1] = get_attr[target=_param_constant597]
    %addmm_88 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant597, %view_491, %t_186), kwargs = {})
    %view_492 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_88, [2, 4096, 320]), kwargs = {})
    %add_297 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_492, %add_294), kwargs = {})
    %var_mean_95 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_297, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_190 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_95, 0), kwargs = {})
    %getitem_191 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_95, 1), kwargs = {})
    %add_298 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_190, 1e-05), kwargs = {})
    %rsqrt_95 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_298,), kwargs = {})
    %sub_95 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_297, %getitem_191), kwargs = {})
    %mul_206 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_95, %rsqrt_95), kwargs = {})
    %_param_constant598 : [#users=1] = get_attr[target=_param_constant598]
    %mul_207 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_206, %_param_constant598), kwargs = {})
    %_param_constant599 : [#users=1] = get_attr[target=_param_constant599]
    %add_299 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_207, %_param_constant599), kwargs = {})
    %_param_constant600 : [#users=1] = get_attr[target=_param_constant600]
    %t_187 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant600,), kwargs = {})
    %view_493 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_299, [8192, 320]), kwargs = {})
    %_param_constant601 : [#users=1] = get_attr[target=_param_constant601]
    %addmm_89 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant601, %view_493, %t_187), kwargs = {})
    %view_494 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_89, [2, 4096, 2560]), kwargs = {})
    %slice_73 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_494, -1, 0, 1280), kwargs = {})
    %slice_74 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_494, -1, 1280, 2560), kwargs = {})
    %gelu_13 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_74,), kwargs = {})
    %mul_208 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_73, %gelu_13), kwargs = {})
    %_param_constant602 : [#users=1] = get_attr[target=_param_constant602]
    %t_188 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant602,), kwargs = {})
    %view_495 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_208, [8192, 1280]), kwargs = {})
    %_param_constant603 : [#users=1] = get_attr[target=_param_constant603]
    %addmm_90 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant603, %view_495, %t_188), kwargs = {})
    %view_496 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_90, [2, 4096, 320]), kwargs = {})
    %add_300 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_496, %add_297), kwargs = {})
    %_param_constant604 : [#users=1] = get_attr[target=_param_constant604]
    %t_189 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant604,), kwargs = {})
    %view_497 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_300, [8192, 320]), kwargs = {})
    %_param_constant605 : [#users=1] = get_attr[target=_param_constant605]
    %addmm_91 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant605, %view_497, %t_189), kwargs = {})
    %view_498 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_91, [2, 4096, 320]), kwargs = {})
    %view_499 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_498, [2, 64, 64, 320]), kwargs = {})
    %permute_139 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_499, [0, 3, 1, 2]), kwargs = {})
    %clone_125 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_139,), kwargs = {memory_format: torch.contiguous_format})
    %add_301 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%clone_125, %div_20), kwargs = {})
    %cat_13 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%add_301, %add_18], 1), kwargs = {})
    %view_500 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_13, [2, 32, 20, 4096]), kwargs = {})
    %var_mean_96 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_500, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_192 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_96, 0), kwargs = {})
    %getitem_193 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_96, 1), kwargs = {})
    %add_302 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_192, 1e-05), kwargs = {})
    %rsqrt_96 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_302,), kwargs = {})
    %sub_96 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_500, %getitem_193), kwargs = {})
    %mul_209 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_96, %rsqrt_96), kwargs = {})
    %view_501 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_209, [2, 640, 64, 64]), kwargs = {})
    %_param_constant606 : [#users=1] = get_attr[target=_param_constant606]
    %unsqueeze_366 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant606, 0), kwargs = {})
    %unsqueeze_367 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_366, 2), kwargs = {})
    %unsqueeze_368 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_367, 3), kwargs = {})
    %_param_constant607 : [#users=1] = get_attr[target=_param_constant607]
    %unsqueeze_369 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant607, 0), kwargs = {})
    %unsqueeze_370 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_369, 2), kwargs = {})
    %unsqueeze_371 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_370, 3), kwargs = {})
    %mul_210 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_501, %unsqueeze_371), kwargs = {})
    %add_303 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_210, %unsqueeze_368), kwargs = {})
    %squeeze_216 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_193, 3), kwargs = {})
    %squeeze_217 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_216, 2), kwargs = {})
    %squeeze_218 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_96, 3), kwargs = {})
    %squeeze_219 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_218, 2), kwargs = {})
    %detach_136 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_217,), kwargs = {})
    %detach_137 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_219,), kwargs = {})
    %silu_61 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_303,), kwargs = {})
    %_param_constant608 : [#users=1] = get_attr[target=_param_constant608]
    %_param_constant609 : [#users=1] = get_attr[target=_param_constant609]
    %convolution_59 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_61, %_param_constant608, %_param_constant609, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_62 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant610 : [#users=1] = get_attr[target=_param_constant610]
    %t_190 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant610,), kwargs = {})
    %_param_constant611 : [#users=1] = get_attr[target=_param_constant611]
    %addmm_92 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant611, %silu_62, %t_190), kwargs = {})
    %slice_75 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_92, 0, 0, 9223372036854775807), kwargs = {})
    %slice_76 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_75, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_372 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_76, 2), kwargs = {})
    %unsqueeze_373 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_372, 3), kwargs = {})
    %add_304 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_59, %unsqueeze_373), kwargs = {})
    %view_502 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_304, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_97 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_502, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_194 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_97, 0), kwargs = {})
    %getitem_195 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_97, 1), kwargs = {})
    %add_305 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_194, 1e-05), kwargs = {})
    %rsqrt_97 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_305,), kwargs = {})
    %sub_97 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_502, %getitem_195), kwargs = {})
    %mul_211 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_97, %rsqrt_97), kwargs = {})
    %view_503 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_211, [2, 320, 64, 64]), kwargs = {})
    %_param_constant612 : [#users=1] = get_attr[target=_param_constant612]
    %unsqueeze_374 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant612, 0), kwargs = {})
    %unsqueeze_375 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_374, 2), kwargs = {})
    %unsqueeze_376 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_375, 3), kwargs = {})
    %_param_constant613 : [#users=1] = get_attr[target=_param_constant613]
    %unsqueeze_377 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant613, 0), kwargs = {})
    %unsqueeze_378 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_377, 2), kwargs = {})
    %unsqueeze_379 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_378, 3), kwargs = {})
    %mul_212 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_503, %unsqueeze_379), kwargs = {})
    %add_306 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_212, %unsqueeze_376), kwargs = {})
    %squeeze_220 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_195, 3), kwargs = {})
    %squeeze_221 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_220, 2), kwargs = {})
    %squeeze_222 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_97, 3), kwargs = {})
    %squeeze_223 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_222, 2), kwargs = {})
    %detach_138 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_221,), kwargs = {})
    %detach_139 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_223,), kwargs = {})
    %silu_63 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_306,), kwargs = {})
    %_param_constant614 : [#users=1] = get_attr[target=_param_constant614]
    %_param_constant615 : [#users=1] = get_attr[target=_param_constant615]
    %convolution_60 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_63, %_param_constant614, %_param_constant615, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant616 : [#users=1] = get_attr[target=_param_constant616]
    %_param_constant617 : [#users=1] = get_attr[target=_param_constant617]
    %convolution_61 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_13, %_param_constant616, %_param_constant617, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_307 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_61, %convolution_60), kwargs = {})
    %div_21 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_307, 1.0), kwargs = {})
    %view_504 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_21, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_98 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_504, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_196 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_98, 0), kwargs = {})
    %getitem_197 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_98, 1), kwargs = {})
    %add_308 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_196, 1e-06), kwargs = {})
    %rsqrt_98 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_308,), kwargs = {})
    %sub_98 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_504, %getitem_197), kwargs = {})
    %mul_213 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_98, %rsqrt_98), kwargs = {})
    %view_505 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_213, [2, 320, 64, 64]), kwargs = {})
    %_param_constant618 : [#users=1] = get_attr[target=_param_constant618]
    %unsqueeze_380 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant618, 0), kwargs = {})
    %unsqueeze_381 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_380, 2), kwargs = {})
    %unsqueeze_382 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_381, 3), kwargs = {})
    %_param_constant619 : [#users=1] = get_attr[target=_param_constant619]
    %unsqueeze_383 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant619, 0), kwargs = {})
    %unsqueeze_384 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_383, 2), kwargs = {})
    %unsqueeze_385 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_384, 3), kwargs = {})
    %mul_214 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_505, %unsqueeze_385), kwargs = {})
    %add_309 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_214, %unsqueeze_382), kwargs = {})
    %squeeze_224 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_197, 3), kwargs = {})
    %squeeze_225 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_224, 2), kwargs = {})
    %squeeze_226 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_98, 3), kwargs = {})
    %squeeze_227 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_226, 2), kwargs = {})
    %detach_140 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_225,), kwargs = {})
    %detach_141 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_227,), kwargs = {})
    %permute_140 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_309, [0, 2, 3, 1]), kwargs = {})
    %view_506 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_140, [2, 4096, 320]), kwargs = {})
    %_param_constant620 : [#users=1] = get_attr[target=_param_constant620]
    %t_191 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant620,), kwargs = {})
    %expand_29 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_506, [2, 4096, 320]), kwargs = {})
    %view_507 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_29, [2, 4096, 320]), kwargs = {})
    %expand_30 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_191, [2, 320, 320]), kwargs = {})
    %view_508 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_30, [2, 320, 320]), kwargs = {})
    %bmm_42 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_507, %view_508), kwargs = {})
    %_unsafe_view_210 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_42, [2, 4096, 320]), kwargs = {})
    %_param_constant621 : [#users=1] = get_attr[target=_param_constant621]
    %add_310 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_210, %_param_constant621), kwargs = {})
    %var_mean_99 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_310, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_198 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_99, 0), kwargs = {})
    %getitem_199 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_99, 1), kwargs = {})
    %add_311 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_198, 1e-05), kwargs = {})
    %rsqrt_99 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_311,), kwargs = {})
    %sub_99 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_310, %getitem_199), kwargs = {})
    %mul_215 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_99, %rsqrt_99), kwargs = {})
    %_param_constant622 : [#users=1] = get_attr[target=_param_constant622]
    %mul_216 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_215, %_param_constant622), kwargs = {})
    %_param_constant623 : [#users=1] = get_attr[target=_param_constant623]
    %add_312 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_216, %_param_constant623), kwargs = {})
    %_param_constant624 : [#users=1] = get_attr[target=_param_constant624]
    %t_192 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant624,), kwargs = {})
    %view_509 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_312, [8192, 320]), kwargs = {})
    %mm_84 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_509, %t_192), kwargs = {})
    %_unsafe_view_211 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_84, [2, 4096, 320]), kwargs = {})
    %view_510 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_211, [2, 4096, 5, 64]), kwargs = {})
    %permute_141 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_510, [0, 2, 1, 3]), kwargs = {})
    %clone_126 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_141,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_212 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_126, [10, 4096, 64]), kwargs = {})
    %_param_constant625 : [#users=1] = get_attr[target=_param_constant625]
    %t_193 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant625,), kwargs = {})
    %view_511 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_312, [8192, 320]), kwargs = {})
    %mm_85 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_511, %t_193), kwargs = {})
    %_unsafe_view_213 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_85, [2, 4096, 320]), kwargs = {})
    %_param_constant626 : [#users=1] = get_attr[target=_param_constant626]
    %t_194 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant626,), kwargs = {})
    %view_512 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_312, [8192, 320]), kwargs = {})
    %mm_86 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_512, %t_194), kwargs = {})
    %_unsafe_view_214 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_86, [2, 4096, 320]), kwargs = {})
    %view_513 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_213, [2, 4096, 5, 64]), kwargs = {})
    %permute_142 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_513, [0, 2, 1, 3]), kwargs = {})
    %clone_127 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_142,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_215 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_127, [10, 4096, 64]), kwargs = {})
    %view_514 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_214, [2, 4096, 5, 64]), kwargs = {})
    %permute_143 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_514, [0, 2, 1, 3]), kwargs = {})
    %clone_128 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_143,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_216 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_128, [10, 4096, 64]), kwargs = {})
    %empty_28 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([10, 4096, 4096],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_28 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_215, -1, -2), kwargs = {})
    %baddbmm_28 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_28, %_unsafe_view_212, %transpose_28), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_28 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_28, -1, False), kwargs = {})
    %detach_142 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_28,), kwargs = {})
    %bmm_43 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_28, %_unsafe_view_216), kwargs = {})
    %view_515 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_43, [2, 5, 4096, 64]), kwargs = {})
    %permute_144 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_515, [0, 2, 1, 3]), kwargs = {})
    %clone_129 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_144,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_217 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_129, [2, 4096, 320]), kwargs = {})
    %_param_constant627 : [#users=1] = get_attr[target=_param_constant627]
    %t_195 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant627,), kwargs = {})
    %view_516 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_217, [8192, 320]), kwargs = {})
    %_param_constant628 : [#users=1] = get_attr[target=_param_constant628]
    %addmm_93 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant628, %view_516, %t_195), kwargs = {})
    %view_517 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_93, [2, 4096, 320]), kwargs = {})
    %add_313 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_517, %add_310), kwargs = {})
    %var_mean_100 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_313, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_200 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_100, 0), kwargs = {})
    %getitem_201 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_100, 1), kwargs = {})
    %add_314 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_200, 1e-05), kwargs = {})
    %rsqrt_100 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_314,), kwargs = {})
    %sub_100 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_313, %getitem_201), kwargs = {})
    %mul_217 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_100, %rsqrt_100), kwargs = {})
    %_param_constant629 : [#users=1] = get_attr[target=_param_constant629]
    %mul_218 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_217, %_param_constant629), kwargs = {})
    %_param_constant630 : [#users=1] = get_attr[target=_param_constant630]
    %add_315 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_218, %_param_constant630), kwargs = {})
    %_param_constant631 : [#users=1] = get_attr[target=_param_constant631]
    %t_196 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant631,), kwargs = {})
    %view_518 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_315, [8192, 320]), kwargs = {})
    %mm_87 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_518, %t_196), kwargs = {})
    %_unsafe_view_218 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_87, [2, 4096, 320]), kwargs = {})
    %view_519 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_218, [2, 4096, 5, 64]), kwargs = {})
    %permute_145 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_519, [0, 2, 1, 3]), kwargs = {})
    %clone_130 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_145,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_219 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_130, [10, 4096, 64]), kwargs = {})
    %_param_constant632 : [#users=1] = get_attr[target=_param_constant632]
    %t_197 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant632,), kwargs = {})
    %view_520 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_88 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_520, %t_197), kwargs = {})
    %_unsafe_view_220 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_88, [2, 64, 320]), kwargs = {})
    %_param_constant633 : [#users=1] = get_attr[target=_param_constant633]
    %t_198 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant633,), kwargs = {})
    %view_521 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_89 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_521, %t_198), kwargs = {})
    %_unsafe_view_221 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_89, [2, 64, 320]), kwargs = {})
    %view_522 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_220, [2, 64, 5, 64]), kwargs = {})
    %permute_146 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_522, [0, 2, 1, 3]), kwargs = {})
    %clone_131 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_146,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_222 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_131, [10, 64, 64]), kwargs = {})
    %view_523 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_221, [2, 64, 5, 64]), kwargs = {})
    %permute_147 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_523, [0, 2, 1, 3]), kwargs = {})
    %clone_132 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_147,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_223 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_132, [10, 64, 64]), kwargs = {})
    %empty_29 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([10, 4096, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_29 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_222, -1, -2), kwargs = {})
    %baddbmm_29 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_29, %_unsafe_view_219, %transpose_29), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_29 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_29, -1, False), kwargs = {})
    %detach_143 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_29,), kwargs = {})
    %bmm_44 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_29, %_unsafe_view_223), kwargs = {})
    %view_524 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_44, [2, 5, 4096, 64]), kwargs = {})
    %permute_148 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_524, [0, 2, 1, 3]), kwargs = {})
    %clone_133 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_148,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_224 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_133, [2, 4096, 320]), kwargs = {})
    %_param_constant634 : [#users=1] = get_attr[target=_param_constant634]
    %t_199 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant634,), kwargs = {})
    %view_525 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_224, [8192, 320]), kwargs = {})
    %_param_constant635 : [#users=1] = get_attr[target=_param_constant635]
    %addmm_94 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant635, %view_525, %t_199), kwargs = {})
    %view_526 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_94, [2, 4096, 320]), kwargs = {})
    %add_316 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_526, %add_313), kwargs = {})
    %var_mean_101 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_316, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_202 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_101, 0), kwargs = {})
    %getitem_203 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_101, 1), kwargs = {})
    %add_317 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_202, 1e-05), kwargs = {})
    %rsqrt_101 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_317,), kwargs = {})
    %sub_101 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_316, %getitem_203), kwargs = {})
    %mul_219 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_101, %rsqrt_101), kwargs = {})
    %_param_constant636 : [#users=1] = get_attr[target=_param_constant636]
    %mul_220 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_219, %_param_constant636), kwargs = {})
    %_param_constant637 : [#users=1] = get_attr[target=_param_constant637]
    %add_318 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_220, %_param_constant637), kwargs = {})
    %_param_constant638 : [#users=1] = get_attr[target=_param_constant638]
    %t_200 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant638,), kwargs = {})
    %view_527 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_318, [8192, 320]), kwargs = {})
    %_param_constant639 : [#users=1] = get_attr[target=_param_constant639]
    %addmm_95 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant639, %view_527, %t_200), kwargs = {})
    %view_528 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_95, [2, 4096, 2560]), kwargs = {})
    %slice_77 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_528, -1, 0, 1280), kwargs = {})
    %slice_78 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_528, -1, 1280, 2560), kwargs = {})
    %gelu_14 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_78,), kwargs = {})
    %mul_221 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_77, %gelu_14), kwargs = {})
    %_param_constant640 : [#users=1] = get_attr[target=_param_constant640]
    %t_201 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant640,), kwargs = {})
    %view_529 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_221, [8192, 1280]), kwargs = {})
    %_param_constant641 : [#users=1] = get_attr[target=_param_constant641]
    %addmm_96 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant641, %view_529, %t_201), kwargs = {})
    %view_530 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_96, [2, 4096, 320]), kwargs = {})
    %add_319 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_530, %add_316), kwargs = {})
    %_param_constant642 : [#users=1] = get_attr[target=_param_constant642]
    %t_202 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant642,), kwargs = {})
    %view_531 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_319, [8192, 320]), kwargs = {})
    %_param_constant643 : [#users=1] = get_attr[target=_param_constant643]
    %addmm_97 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant643, %view_531, %t_202), kwargs = {})
    %view_532 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_97, [2, 4096, 320]), kwargs = {})
    %view_533 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_532, [2, 64, 64, 320]), kwargs = {})
    %permute_149 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_533, [0, 3, 1, 2]), kwargs = {})
    %clone_134 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_149,), kwargs = {memory_format: torch.contiguous_format})
    %add_320 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%clone_134, %div_21), kwargs = {})
    %cat_14 : [#users=2] = call_function[target=torch.ops.aten.cat](args = ([%add_320, %convolution], 1), kwargs = {})
    %view_534 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%cat_14, [2, 32, 20, 4096]), kwargs = {})
    %var_mean_102 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_534, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_204 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_102, 0), kwargs = {})
    %getitem_205 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_102, 1), kwargs = {})
    %add_321 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_204, 1e-05), kwargs = {})
    %rsqrt_102 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_321,), kwargs = {})
    %sub_102 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_534, %getitem_205), kwargs = {})
    %mul_222 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_102, %rsqrt_102), kwargs = {})
    %view_535 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_222, [2, 640, 64, 64]), kwargs = {})
    %_param_constant644 : [#users=1] = get_attr[target=_param_constant644]
    %unsqueeze_386 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant644, 0), kwargs = {})
    %unsqueeze_387 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_386, 2), kwargs = {})
    %unsqueeze_388 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_387, 3), kwargs = {})
    %_param_constant645 : [#users=1] = get_attr[target=_param_constant645]
    %unsqueeze_389 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant645, 0), kwargs = {})
    %unsqueeze_390 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_389, 2), kwargs = {})
    %unsqueeze_391 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_390, 3), kwargs = {})
    %mul_223 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_535, %unsqueeze_391), kwargs = {})
    %add_322 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_223, %unsqueeze_388), kwargs = {})
    %squeeze_228 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_205, 3), kwargs = {})
    %squeeze_229 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_228, 2), kwargs = {})
    %squeeze_230 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_102, 3), kwargs = {})
    %squeeze_231 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_230, 2), kwargs = {})
    %detach_144 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_229,), kwargs = {})
    %detach_145 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_231,), kwargs = {})
    %silu_64 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_322,), kwargs = {})
    %_param_constant646 : [#users=1] = get_attr[target=_param_constant646]
    %_param_constant647 : [#users=1] = get_attr[target=_param_constant647]
    %convolution_62 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_64, %_param_constant646, %_param_constant647, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %silu_65 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%addmm_1,), kwargs = {})
    %_param_constant648 : [#users=1] = get_attr[target=_param_constant648]
    %t_203 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant648,), kwargs = {})
    %_param_constant649 : [#users=1] = get_attr[target=_param_constant649]
    %addmm_98 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant649, %silu_65, %t_203), kwargs = {})
    %slice_79 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%addmm_98, 0, 0, 9223372036854775807), kwargs = {})
    %slice_80 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%slice_79, 1, 0, 9223372036854775807), kwargs = {})
    %unsqueeze_392 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%slice_80, 2), kwargs = {})
    %unsqueeze_393 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_392, 3), kwargs = {})
    %add_323 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_62, %unsqueeze_393), kwargs = {})
    %view_536 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_323, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_103 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_536, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_206 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_103, 0), kwargs = {})
    %getitem_207 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_103, 1), kwargs = {})
    %add_324 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_206, 1e-05), kwargs = {})
    %rsqrt_103 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_324,), kwargs = {})
    %sub_103 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_536, %getitem_207), kwargs = {})
    %mul_224 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_103, %rsqrt_103), kwargs = {})
    %view_537 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_224, [2, 320, 64, 64]), kwargs = {})
    %_param_constant650 : [#users=1] = get_attr[target=_param_constant650]
    %unsqueeze_394 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant650, 0), kwargs = {})
    %unsqueeze_395 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_394, 2), kwargs = {})
    %unsqueeze_396 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_395, 3), kwargs = {})
    %_param_constant651 : [#users=1] = get_attr[target=_param_constant651]
    %unsqueeze_397 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant651, 0), kwargs = {})
    %unsqueeze_398 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_397, 2), kwargs = {})
    %unsqueeze_399 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_398, 3), kwargs = {})
    %mul_225 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_537, %unsqueeze_399), kwargs = {})
    %add_325 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_225, %unsqueeze_396), kwargs = {})
    %squeeze_232 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_207, 3), kwargs = {})
    %squeeze_233 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_232, 2), kwargs = {})
    %squeeze_234 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_103, 3), kwargs = {})
    %squeeze_235 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_234, 2), kwargs = {})
    %detach_146 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_233,), kwargs = {})
    %detach_147 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_235,), kwargs = {})
    %silu_66 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_325,), kwargs = {})
    %_param_constant652 : [#users=1] = get_attr[target=_param_constant652]
    %_param_constant653 : [#users=1] = get_attr[target=_param_constant653]
    %convolution_63 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%silu_66, %_param_constant652, %_param_constant653, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %_param_constant654 : [#users=1] = get_attr[target=_param_constant654]
    %_param_constant655 : [#users=1] = get_attr[target=_param_constant655]
    %convolution_64 : [#users=1] = call_function[target=torch.ops.aten.convolution](args = (%cat_14, %_param_constant654, %_param_constant655, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
    %add_326 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%convolution_64, %convolution_63), kwargs = {})
    %div_22 : [#users=2] = call_function[target=torch.ops.aten.div](args = (%add_326, 1.0), kwargs = {})
    %view_538 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%div_22, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_104 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_538, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_208 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_104, 0), kwargs = {})
    %getitem_209 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_104, 1), kwargs = {})
    %add_327 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_208, 1e-06), kwargs = {})
    %rsqrt_104 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_327,), kwargs = {})
    %sub_104 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_538, %getitem_209), kwargs = {})
    %mul_226 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_104, %rsqrt_104), kwargs = {})
    %view_539 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_226, [2, 320, 64, 64]), kwargs = {})
    %_param_constant656 : [#users=1] = get_attr[target=_param_constant656]
    %unsqueeze_400 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant656, 0), kwargs = {})
    %unsqueeze_401 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_400, 2), kwargs = {})
    %unsqueeze_402 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_401, 3), kwargs = {})
    %_param_constant657 : [#users=1] = get_attr[target=_param_constant657]
    %unsqueeze_403 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant657, 0), kwargs = {})
    %unsqueeze_404 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_403, 2), kwargs = {})
    %unsqueeze_405 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_404, 3), kwargs = {})
    %mul_227 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_539, %unsqueeze_405), kwargs = {})
    %add_328 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_227, %unsqueeze_402), kwargs = {})
    %squeeze_236 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_209, 3), kwargs = {})
    %squeeze_237 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_236, 2), kwargs = {})
    %squeeze_238 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_104, 3), kwargs = {})
    %squeeze_239 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_238, 2), kwargs = {})
    %detach_148 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_237,), kwargs = {})
    %detach_149 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_239,), kwargs = {})
    %permute_150 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%add_328, [0, 2, 3, 1]), kwargs = {})
    %view_540 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%permute_150, [2, 4096, 320]), kwargs = {})
    %_param_constant658 : [#users=1] = get_attr[target=_param_constant658]
    %t_204 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant658,), kwargs = {})
    %expand_31 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%view_540, [2, 4096, 320]), kwargs = {})
    %view_541 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_31, [2, 4096, 320]), kwargs = {})
    %expand_32 : [#users=1] = call_function[target=torch.ops.aten.expand](args = (%t_204, [2, 320, 320]), kwargs = {})
    %view_542 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%expand_32, [2, 320, 320]), kwargs = {})
    %bmm_45 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%view_541, %view_542), kwargs = {})
    %_unsafe_view_225 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%bmm_45, [2, 4096, 320]), kwargs = {})
    %_param_constant659 : [#users=1] = get_attr[target=_param_constant659]
    %add_329 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%_unsafe_view_225, %_param_constant659), kwargs = {})
    %var_mean_105 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_329, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_210 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_105, 0), kwargs = {})
    %getitem_211 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_105, 1), kwargs = {})
    %add_330 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_210, 1e-05), kwargs = {})
    %rsqrt_105 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_330,), kwargs = {})
    %sub_105 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_329, %getitem_211), kwargs = {})
    %mul_228 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_105, %rsqrt_105), kwargs = {})
    %_param_constant660 : [#users=1] = get_attr[target=_param_constant660]
    %mul_229 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_228, %_param_constant660), kwargs = {})
    %_param_constant661 : [#users=1] = get_attr[target=_param_constant661]
    %add_331 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%mul_229, %_param_constant661), kwargs = {})
    %_param_constant662 : [#users=1] = get_attr[target=_param_constant662]
    %t_205 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant662,), kwargs = {})
    %view_543 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_331, [8192, 320]), kwargs = {})
    %mm_90 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_543, %t_205), kwargs = {})
    %_unsafe_view_226 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_90, [2, 4096, 320]), kwargs = {})
    %view_544 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_226, [2, 4096, 5, 64]), kwargs = {})
    %permute_151 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_544, [0, 2, 1, 3]), kwargs = {})
    %clone_135 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_151,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_227 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_135, [10, 4096, 64]), kwargs = {})
    %_param_constant663 : [#users=1] = get_attr[target=_param_constant663]
    %t_206 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant663,), kwargs = {})
    %view_545 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_331, [8192, 320]), kwargs = {})
    %mm_91 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_545, %t_206), kwargs = {})
    %_unsafe_view_228 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_91, [2, 4096, 320]), kwargs = {})
    %_param_constant664 : [#users=1] = get_attr[target=_param_constant664]
    %t_207 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant664,), kwargs = {})
    %view_546 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_331, [8192, 320]), kwargs = {})
    %mm_92 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_546, %t_207), kwargs = {})
    %_unsafe_view_229 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_92, [2, 4096, 320]), kwargs = {})
    %view_547 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_228, [2, 4096, 5, 64]), kwargs = {})
    %permute_152 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_547, [0, 2, 1, 3]), kwargs = {})
    %clone_136 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_152,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_230 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_136, [10, 4096, 64]), kwargs = {})
    %view_548 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_229, [2, 4096, 5, 64]), kwargs = {})
    %permute_153 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_548, [0, 2, 1, 3]), kwargs = {})
    %clone_137 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_153,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_231 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_137, [10, 4096, 64]), kwargs = {})
    %empty_30 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([10, 4096, 4096],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_30 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_230, -1, -2), kwargs = {})
    %baddbmm_30 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_30, %_unsafe_view_227, %transpose_30), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_30 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_30, -1, False), kwargs = {})
    %detach_150 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_30,), kwargs = {})
    %bmm_46 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_30, %_unsafe_view_231), kwargs = {})
    %view_549 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_46, [2, 5, 4096, 64]), kwargs = {})
    %permute_154 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_549, [0, 2, 1, 3]), kwargs = {})
    %clone_138 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_154,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_232 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_138, [2, 4096, 320]), kwargs = {})
    %_param_constant665 : [#users=1] = get_attr[target=_param_constant665]
    %t_208 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant665,), kwargs = {})
    %view_550 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_232, [8192, 320]), kwargs = {})
    %_param_constant666 : [#users=1] = get_attr[target=_param_constant666]
    %addmm_99 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant666, %view_550, %t_208), kwargs = {})
    %view_551 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_99, [2, 4096, 320]), kwargs = {})
    %add_332 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_551, %add_329), kwargs = {})
    %var_mean_106 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_332, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_212 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_106, 0), kwargs = {})
    %getitem_213 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_106, 1), kwargs = {})
    %add_333 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_212, 1e-05), kwargs = {})
    %rsqrt_106 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_333,), kwargs = {})
    %sub_106 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_332, %getitem_213), kwargs = {})
    %mul_230 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_106, %rsqrt_106), kwargs = {})
    %_param_constant667 : [#users=1] = get_attr[target=_param_constant667]
    %mul_231 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_230, %_param_constant667), kwargs = {})
    %_param_constant668 : [#users=1] = get_attr[target=_param_constant668]
    %add_334 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_231, %_param_constant668), kwargs = {})
    %_param_constant669 : [#users=1] = get_attr[target=_param_constant669]
    %t_209 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant669,), kwargs = {})
    %view_552 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_334, [8192, 320]), kwargs = {})
    %mm_93 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_552, %t_209), kwargs = {})
    %_unsafe_view_233 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_93, [2, 4096, 320]), kwargs = {})
    %view_553 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_233, [2, 4096, 5, 64]), kwargs = {})
    %permute_155 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_553, [0, 2, 1, 3]), kwargs = {})
    %clone_139 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_155,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_234 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_139, [10, 4096, 64]), kwargs = {})
    %_param_constant670 : [#users=1] = get_attr[target=_param_constant670]
    %t_210 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant670,), kwargs = {})
    %view_554 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_94 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_554, %t_210), kwargs = {})
    %_unsafe_view_235 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_94, [2, 64, 320]), kwargs = {})
    %_param_constant671 : [#users=1] = get_attr[target=_param_constant671]
    %t_211 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant671,), kwargs = {})
    %view_555 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%arg2_1, [128, 1024]), kwargs = {})
    %mm_95 : [#users=1] = call_function[target=torch.ops.aten.mm](args = (%view_555, %t_211), kwargs = {})
    %_unsafe_view_236 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%mm_95, [2, 64, 320]), kwargs = {})
    %view_556 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_235, [2, 64, 5, 64]), kwargs = {})
    %permute_156 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_556, [0, 2, 1, 3]), kwargs = {})
    %clone_140 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_156,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_237 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_140, [10, 64, 64]), kwargs = {})
    %view_557 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_236, [2, 64, 5, 64]), kwargs = {})
    %permute_157 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_557, [0, 2, 1, 3]), kwargs = {})
    %clone_141 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_157,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_238 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_141, [10, 64, 64]), kwargs = {})
    %empty_31 : [#users=1] = call_function[target=torch.ops.aten.empty](args = ([10, 4096, 64],), kwargs = {dtype: torch.float16, device: cpu, pin_memory: False})
    %transpose_31 : [#users=1] = call_function[target=torch.ops.aten.transpose](args = (%_unsafe_view_237, -1, -2), kwargs = {})
    %baddbmm_31 : [#users=1] = call_function[target=torch.ops.aten.baddbmm](args = (%empty_31, %_unsafe_view_234, %transpose_31), kwargs = {beta: 0, alpha: 0.125})
    %_softmax_31 : [#users=2] = call_function[target=torch.ops.aten._softmax](args = (%baddbmm_31, -1, False), kwargs = {})
    %detach_151 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%_softmax_31,), kwargs = {})
    %bmm_47 : [#users=1] = call_function[target=torch.ops.aten.bmm](args = (%_softmax_31, %_unsafe_view_238), kwargs = {})
    %view_558 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%bmm_47, [2, 5, 4096, 64]), kwargs = {})
    %permute_158 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_558, [0, 2, 1, 3]), kwargs = {})
    %clone_142 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_158,), kwargs = {memory_format: torch.contiguous_format})
    %_unsafe_view_239 : [#users=1] = call_function[target=torch.ops.aten._unsafe_view](args = (%clone_142, [2, 4096, 320]), kwargs = {})
    %_param_constant672 : [#users=1] = get_attr[target=_param_constant672]
    %t_212 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant672,), kwargs = {})
    %view_559 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%_unsafe_view_239, [8192, 320]), kwargs = {})
    %_param_constant673 : [#users=1] = get_attr[target=_param_constant673]
    %addmm_100 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant673, %view_559, %t_212), kwargs = {})
    %view_560 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_100, [2, 4096, 320]), kwargs = {})
    %add_335 : [#users=3] = call_function[target=torch.ops.aten.add](args = (%view_560, %add_332), kwargs = {})
    %var_mean_107 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%add_335, [2]), kwargs = {correction: 0, keepdim: True})
    %getitem_214 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_107, 0), kwargs = {})
    %getitem_215 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_107, 1), kwargs = {})
    %add_336 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_214, 1e-05), kwargs = {})
    %rsqrt_107 : [#users=1] = call_function[target=torch.ops.aten.rsqrt](args = (%add_336,), kwargs = {})
    %sub_107 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%add_335, %getitem_215), kwargs = {})
    %mul_232 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_107, %rsqrt_107), kwargs = {})
    %_param_constant674 : [#users=1] = get_attr[target=_param_constant674]
    %mul_233 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%mul_232, %_param_constant674), kwargs = {})
    %_param_constant675 : [#users=1] = get_attr[target=_param_constant675]
    %add_337 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_233, %_param_constant675), kwargs = {})
    %_param_constant676 : [#users=1] = get_attr[target=_param_constant676]
    %t_213 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant676,), kwargs = {})
    %view_561 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_337, [8192, 320]), kwargs = {})
    %_param_constant677 : [#users=1] = get_attr[target=_param_constant677]
    %addmm_101 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant677, %view_561, %t_213), kwargs = {})
    %view_562 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%addmm_101, [2, 4096, 2560]), kwargs = {})
    %slice_81 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_562, -1, 0, 1280), kwargs = {})
    %slice_82 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%view_562, -1, 1280, 2560), kwargs = {})
    %gelu_15 : [#users=1] = call_function[target=torch.ops.aten.gelu](args = (%slice_82,), kwargs = {})
    %mul_234 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%slice_81, %gelu_15), kwargs = {})
    %_param_constant678 : [#users=1] = get_attr[target=_param_constant678]
    %t_214 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant678,), kwargs = {})
    %view_563 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_234, [8192, 1280]), kwargs = {})
    %_param_constant679 : [#users=1] = get_attr[target=_param_constant679]
    %addmm_102 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant679, %view_563, %t_214), kwargs = {})
    %view_564 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_102, [2, 4096, 320]), kwargs = {})
    %add_338 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%view_564, %add_335), kwargs = {})
    %_param_constant680 : [#users=1] = get_attr[target=_param_constant680]
    %t_215 : [#users=1] = call_function[target=torch.ops.aten.t](args = (%_param_constant680,), kwargs = {})
    %view_565 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%add_338, [8192, 320]), kwargs = {})
    %_param_constant681 : [#users=1] = get_attr[target=_param_constant681]
    %addmm_103 : [#users=1] = call_function[target=torch.ops.aten.addmm](args = (%_param_constant681, %view_565, %t_215), kwargs = {})
    %view_566 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%addmm_103, [2, 4096, 320]), kwargs = {})
    %view_567 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%view_566, [2, 64, 64, 320]), kwargs = {})
    %permute_159 : [#users=1] = call_function[target=torch.ops.aten.permute](args = (%view_567, [0, 3, 1, 2]), kwargs = {})
    %clone_143 : [#users=1] = call_function[target=torch.ops.aten.clone](args = (%permute_159,), kwargs = {memory_format: torch.contiguous_format})
    %add_339 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%clone_143, %div_22), kwargs = {})
    %view_568 : [#users=2] = call_function[target=torch.ops.aten.view](args = (%add_339, [2, 32, 10, 4096]), kwargs = {})
    %var_mean_108 : [#users=2] = call_function[target=torch.ops.aten.var_mean](args = (%view_568, [2, 3]), kwargs = {correction: 0, keepdim: True})
    %getitem_216 : [#users=1] = call_function[target=operator.getitem](args = (%var_mean_108, 0), kwargs = {})
    %getitem_217 : [#users=2] = call_function[target=operator.getitem](args = (%var_mean_108, 1), kwargs = {})
    %add_340 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%getitem_216, 1e-05), kwargs = {})
    %rsqrt_108 : [#users=2] = call_function[target=torch.ops.aten.rsqrt](args = (%add_340,), kwargs = {})
    %sub_108 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%view_568, %getitem_217), kwargs = {})
    %mul_235 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%sub_108, %rsqrt_108), kwargs = {})
    %view_569 : [#users=1] = call_function[target=torch.ops.aten.view](args = (%mul_235, [2, 320, 64, 64]), kwargs = {})
    %_param_constant682 : [#users=1] = get_attr[target=_param_constant682]
    %unsqueeze_406 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant682, 0), kwargs = {})
    %unsqueeze_407 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_406, 2), kwargs = {})
    %unsqueeze_408 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_407, 3), kwargs = {})
    %_param_constant683 : [#users=1] = get_attr[target=_param_constant683]
    %unsqueeze_409 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%_param_constant683, 0), kwargs = {})
    %unsqueeze_410 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_409, 2), kwargs = {})
    %unsqueeze_411 : [#users=1] = call_function[target=torch.ops.aten.unsqueeze](args = (%unsqueeze_410, 3), kwargs = {})
    %mul_236 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%view_569, %unsqueeze_411), kwargs = {})
    %add_341 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%mul_236, %unsqueeze_408), kwargs = {})
    %squeeze_240 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%getitem_217, 3), kwargs = {})
    %squeeze_241 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_240, 2), kwargs = {})
    %squeeze_242 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%rsqrt_108, 3), kwargs = {})
    %squeeze_243 : [#users=1] = call_function[target=torch.ops.aten.squeeze](args = (%squeeze_242, 2), kwargs = {})
    %detach_152 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_241,), kwargs = {})
    %detach_153 : [#users=0] = call_function[target=torch.ops.aten.detach](args = (%squeeze_243,), kwargs = {})
    %silu_67 : [#users=1] = call_function[target=torch.ops.aten.silu](args = (%add_341,), kwargs = {})
    %_param_constant684 : [#users=1] = get_attr[target=_param_constant684]
    %_param_constant685 : [#users=1] = get_attr[target=_param_constant685]
    %convolution_65 : [#users=2] = call_function[target=torch.ops.aten.convolution](args = (%silu_67, %_param_constant684, %_param_constant685, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
    %slice_83 : [#users=2] = call_function[target=torch.ops.aten.slice](args = (%convolution_65, 0, 0, 1), kwargs = {})
    %slice_84 : [#users=1] = call_function[target=torch.ops.aten.slice](args = (%convolution_65, 0, 1, 2), kwargs = {})
    %sub_109 : [#users=1] = call_function[target=torch.ops.aten.sub](args = (%slice_84, %slice_83), kwargs = {})
    %mul_237 : [#users=1] = call_function[target=torch.ops.aten.mul](args = (%arg3_1, %sub_109), kwargs = {})
    %add_342 : [#users=1] = call_function[target=torch.ops.aten.add](args = (%slice_83, %mul_237), kwargs = {})
    return add_342

cpu.mlir

module attributes {torch.debug_module_name = "_lambda"} {
  func.func @forward(%arg0: !torch.vtensor<[1,4,64,64],f16>, %arg1: !torch.vtensor<[1],f16>, %arg2: !torch.vtensor<[2,64,1024],f16>, %arg3: !torch.vtensor<[],f32>) -> !torch.vtensor<[1,4,64,64],f16> {
    %none = torch.constant.none
    %int7 = torch.constant.int 7
    %false = torch.constant.bool false
    %true = torch.constant.bool true
    %int40960 = torch.constant.int 40960
    %int10 = torch.constant.int 10
    %int4096 = torch.constant.int 4096
    %float1.000000e00 = torch.constant.float 1.000000e+00
    %int5 = torch.constant.int 5
    %int0 = torch.constant.int 0
    %int1 = torch.constant.int 1
    %int320 = torch.constant.int 320
    %int81920 = torch.constant.int 81920
    %int20 = torch.constant.int 20
    %int122880 = torch.constant.int 122880
    %int30 = torch.constant.int 30
    %int640 = torch.constant.int 640
    %int20480 = torch.constant.int 20480
    %int1024 = torch.constant.int 1024
    %int30720 = torch.constant.int 30720
    %int40 = torch.constant.int 40
    %int61440 = torch.constant.int 61440
    %int60 = torch.constant.int 60
    %int1280 = torch.constant.int 1280
    %int10240 = torch.constant.int 10240
    %int256 = torch.constant.int 256
    %int15360 = torch.constant.int 15360
    %int80 = torch.constant.int 80
    %int2560 = torch.constant.int 2560
    %int64 = torch.constant.int 64
    %int5120 = torch.constant.int 5120
    %int160 = torch.constant.int 160
    %0 = torch.vtensor.literal(dense<[-2.388000e-03, 3.170010e-03, 4.332070e-04, 1.795770e-03]> : tensor<4xf16>) : !torch.vtensor<[4],f16>
    %1 = torch.vtensor.literal(dense_resource<__elided__> : tensor<4x320x3x3xf16>) : !torch.vtensor<[4,320,3,3],f16>
    %2 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %3 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %4 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %5 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %6 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %7 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1280xf16>) : !torch.vtensor<[320,1280],f16>
    %8 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %9 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560x320xf16>) : !torch.vtensor<[2560,320],f16>
    %10 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %11 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %12 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %13 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %14 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1024xf16>) : !torch.vtensor<[320,1024],f16>
    %15 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1024xf16>) : !torch.vtensor<[320,1024],f16>
    %16 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %17 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %18 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %19 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %20 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %21 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %22 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %23 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %24 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %25 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %26 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %27 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %28 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %29 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %30 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %31 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x640x1x1xf16>) : !torch.vtensor<[320,640,1,1],f16>
    %32 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %33 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320x3x3xf16>) : !torch.vtensor<[320,320,3,3],f16>
    %34 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %35 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %36 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %37 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1280xf16>) : !torch.vtensor<[320,1280],f16>
    %38 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %39 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x640x3x3xf16>) : !torch.vtensor<[320,640,3,3],f16>
    %40 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %41 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %42 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %43 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %44 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %45 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1280xf16>) : !torch.vtensor<[320,1280],f16>
    %46 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %47 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560x320xf16>) : !torch.vtensor<[2560,320],f16>
    %48 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %49 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %50 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %51 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %52 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1024xf16>) : !torch.vtensor<[320,1024],f16>
    %53 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1024xf16>) : !torch.vtensor<[320,1024],f16>
    %54 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %55 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %56 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %57 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %58 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %59 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %60 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %61 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %62 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %63 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %64 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %65 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %66 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %67 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %68 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %69 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x640x1x1xf16>) : !torch.vtensor<[320,640,1,1],f16>
    %70 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %71 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320x3x3xf16>) : !torch.vtensor<[320,320,3,3],f16>
    %72 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %73 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %74 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %75 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1280xf16>) : !torch.vtensor<[320,1280],f16>
    %76 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %77 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x640x3x3xf16>) : !torch.vtensor<[320,640,3,3],f16>
    %78 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %79 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %80 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %81 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %82 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %83 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1280xf16>) : !torch.vtensor<[320,1280],f16>
    %84 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %85 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560x320xf16>) : !torch.vtensor<[2560,320],f16>
    %86 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %87 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %88 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %89 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %90 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1024xf16>) : !torch.vtensor<[320,1024],f16>
    %91 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1024xf16>) : !torch.vtensor<[320,1024],f16>
    %92 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %93 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %94 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %95 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %96 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %97 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %98 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %99 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %100 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %101 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %102 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %103 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %104 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %105 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %106 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %107 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x960x1x1xf16>) : !torch.vtensor<[320,960,1,1],f16>
    %108 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %109 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320x3x3xf16>) : !torch.vtensor<[320,320,3,3],f16>
    %110 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %111 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %112 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %113 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1280xf16>) : !torch.vtensor<[320,1280],f16>
    %114 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %115 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x960x3x3xf16>) : !torch.vtensor<[320,960,3,3],f16>
    %116 = torch.vtensor.literal(dense_resource<__elided__> : tensor<960xf16>) : !torch.vtensor<[960],f16>
    %117 = torch.vtensor.literal(dense_resource<__elided__> : tensor<960xf16>) : !torch.vtensor<[960],f16>
    %118 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %119 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640x3x3xf16>) : !torch.vtensor<[640,640,3,3],f16>
    %120 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %121 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %122 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %123 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x2560xf16>) : !torch.vtensor<[640,2560],f16>
    %124 = torch.vtensor.literal(dense_resource<__elided__> : tensor<5120xf16>) : !torch.vtensor<[5120],f16>
    %125 = torch.vtensor.literal(dense_resource<__elided__> : tensor<5120x640xf16>) : !torch.vtensor<[5120,640],f16>
    %126 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %127 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %128 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %129 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %130 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1024xf16>) : !torch.vtensor<[640,1024],f16>
    %131 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1024xf16>) : !torch.vtensor<[640,1024],f16>
    %132 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %133 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %134 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %135 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %136 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %137 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %138 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %139 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %140 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %141 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %142 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %143 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %144 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %145 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %146 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %147 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x960x1x1xf16>) : !torch.vtensor<[640,960,1,1],f16>
    %148 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %149 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640x3x3xf16>) : !torch.vtensor<[640,640,3,3],f16>
    %150 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %151 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %152 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %153 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1280xf16>) : !torch.vtensor<[640,1280],f16>
    %154 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %155 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x960x3x3xf16>) : !torch.vtensor<[640,960,3,3],f16>
    %156 = torch.vtensor.literal(dense_resource<__elided__> : tensor<960xf16>) : !torch.vtensor<[960],f16>
    %157 = torch.vtensor.literal(dense_resource<__elided__> : tensor<960xf16>) : !torch.vtensor<[960],f16>
    %158 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %159 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %160 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %161 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x2560xf16>) : !torch.vtensor<[640,2560],f16>
    %162 = torch.vtensor.literal(dense_resource<__elided__> : tensor<5120xf16>) : !torch.vtensor<[5120],f16>
    %163 = torch.vtensor.literal(dense_resource<__elided__> : tensor<5120x640xf16>) : !torch.vtensor<[5120,640],f16>
    %164 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %165 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %166 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %167 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %168 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1024xf16>) : !torch.vtensor<[640,1024],f16>
    %169 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1024xf16>) : !torch.vtensor<[640,1024],f16>
    %170 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %171 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %172 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %173 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %174 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %175 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %176 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %177 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %178 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %179 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %180 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %181 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %182 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %183 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %184 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %185 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1280x1x1xf16>) : !torch.vtensor<[640,1280,1,1],f16>
    %186 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %187 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640x3x3xf16>) : !torch.vtensor<[640,640,3,3],f16>
    %188 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %189 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %190 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %191 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1280xf16>) : !torch.vtensor<[640,1280],f16>
    %192 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %193 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1280x3x3xf16>) : !torch.vtensor<[640,1280,3,3],f16>
    %194 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %195 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %196 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %197 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %198 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %199 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x2560xf16>) : !torch.vtensor<[640,2560],f16>
    %200 = torch.vtensor.literal(dense_resource<__elided__> : tensor<5120xf16>) : !torch.vtensor<[5120],f16>
    %201 = torch.vtensor.literal(dense_resource<__elided__> : tensor<5120x640xf16>) : !torch.vtensor<[5120,640],f16>
    %202 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %203 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %204 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %205 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %206 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1024xf16>) : !torch.vtensor<[640,1024],f16>
    %207 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1024xf16>) : !torch.vtensor<[640,1024],f16>
    %208 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %209 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %210 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %211 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %212 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %213 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %214 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %215 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %216 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %217 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %218 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %219 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %220 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %221 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %222 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %223 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1920x1x1xf16>) : !torch.vtensor<[640,1920,1,1],f16>
    %224 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %225 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640x3x3xf16>) : !torch.vtensor<[640,640,3,3],f16>
    %226 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %227 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %228 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %229 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1280xf16>) : !torch.vtensor<[640,1280],f16>
    %230 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %231 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1920x3x3xf16>) : !torch.vtensor<[640,1920,3,3],f16>
    %232 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1920xf16>) : !torch.vtensor<[1920],f16>
    %233 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1920xf16>) : !torch.vtensor<[1920],f16>
    %234 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %235 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %236 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %237 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %238 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %239 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x5120xf16>) : !torch.vtensor<[1280,5120],f16>
    %240 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240xf16>) : !torch.vtensor<[10240],f16>
    %241 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240x1280xf16>) : !torch.vtensor<[10240,1280],f16>
    %242 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %243 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %244 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %245 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %246 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %247 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %248 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %249 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %250 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %251 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %252 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %253 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %254 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %255 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %256 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %257 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %258 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %259 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %260 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %261 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %262 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %263 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1920x1x1xf16>) : !torch.vtensor<[1280,1920,1,1],f16>
    %264 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %265 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %266 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %267 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %268 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %269 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %270 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %271 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1920x3x3xf16>) : !torch.vtensor<[1280,1920,3,3],f16>
    %272 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1920xf16>) : !torch.vtensor<[1920],f16>
    %273 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1920xf16>) : !torch.vtensor<[1920],f16>
    %274 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %275 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %276 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %277 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x5120xf16>) : !torch.vtensor<[1280,5120],f16>
    %278 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240xf16>) : !torch.vtensor<[10240],f16>
    %279 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240x1280xf16>) : !torch.vtensor<[10240,1280],f16>
    %280 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %281 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %282 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %283 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %284 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %285 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %286 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %287 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %288 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %289 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %290 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %291 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %292 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %293 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %294 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %295 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %296 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %297 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %298 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %299 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %300 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %301 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x2560x1x1xf16>) : !torch.vtensor<[1280,2560,1,1],f16>
    %302 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %303 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %304 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %305 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %306 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %307 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %308 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %309 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x2560x3x3xf16>) : !torch.vtensor<[1280,2560,3,3],f16>
    %310 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %311 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %312 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %313 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %314 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %315 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x5120xf16>) : !torch.vtensor<[1280,5120],f16>
    %316 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240xf16>) : !torch.vtensor<[10240],f16>
    %317 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240x1280xf16>) : !torch.vtensor<[10240,1280],f16>
    %318 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %319 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %320 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %321 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %322 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %323 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %324 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %325 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %326 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %327 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %328 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %329 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %330 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %331 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %332 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %333 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %334 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %335 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %336 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %337 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %338 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %339 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x2560x1x1xf16>) : !torch.vtensor<[1280,2560,1,1],f16>
    %340 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %341 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %342 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %343 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %344 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %345 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %346 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %347 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x2560x3x3xf16>) : !torch.vtensor<[1280,2560,3,3],f16>
    %348 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %349 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %350 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %351 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %352 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %353 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x2560x1x1xf16>) : !torch.vtensor<[1280,2560,1,1],f16>
    %354 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %355 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %356 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %357 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %358 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %359 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %360 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %361 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x2560x3x3xf16>) : !torch.vtensor<[1280,2560,3,3],f16>
    %362 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %363 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %364 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %365 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x2560x1x1xf16>) : !torch.vtensor<[1280,2560,1,1],f16>
    %366 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %367 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %368 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %369 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %370 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %371 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %372 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %373 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x2560x3x3xf16>) : !torch.vtensor<[1280,2560,3,3],f16>
    %374 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %375 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %376 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %377 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x2560x1x1xf16>) : !torch.vtensor<[1280,2560,1,1],f16>
    %378 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %379 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %380 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %381 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %382 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %383 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %384 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %385 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x2560x3x3xf16>) : !torch.vtensor<[1280,2560,3,3],f16>
    %386 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %387 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %388 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %389 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %390 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %391 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %392 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %393 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %394 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %395 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %396 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %397 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %398 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %399 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %400 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %401 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x5120xf16>) : !torch.vtensor<[1280,5120],f16>
    %402 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240xf16>) : !torch.vtensor<[10240],f16>
    %403 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240x1280xf16>) : !torch.vtensor<[10240,1280],f16>
    %404 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %405 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %406 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %407 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %408 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %409 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %410 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %411 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %412 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %413 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %414 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %415 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %416 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %417 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %418 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %419 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %420 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %421 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %422 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %423 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %424 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %425 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %426 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %427 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %428 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %429 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %430 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %431 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %432 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %433 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %434 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %435 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %436 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %437 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %438 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %439 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %440 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %441 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %442 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %443 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %444 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %445 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %446 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %447 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %448 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %449 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %450 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %451 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %452 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %453 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %454 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %455 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %456 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %457 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %458 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %459 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x5120xf16>) : !torch.vtensor<[1280,5120],f16>
    %460 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240xf16>) : !torch.vtensor<[10240],f16>
    %461 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240x1280xf16>) : !torch.vtensor<[10240,1280],f16>
    %462 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %463 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %464 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %465 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %466 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %467 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %468 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %469 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %470 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %471 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %472 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %473 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %474 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %475 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %476 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %477 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %478 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %479 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %480 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %481 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %482 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %483 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %484 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %485 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %486 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %487 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %488 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %489 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %490 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %491 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %492 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %493 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %494 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %495 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x5120xf16>) : !torch.vtensor<[1280,5120],f16>
    %496 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240xf16>) : !torch.vtensor<[10240],f16>
    %497 = torch.vtensor.literal(dense_resource<__elided__> : tensor<10240x1280xf16>) : !torch.vtensor<[10240,1280],f16>
    %498 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %499 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %500 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %501 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %502 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %503 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1024xf16>) : !torch.vtensor<[1280,1024],f16>
    %504 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %505 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %506 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %507 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %508 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %509 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %510 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %511 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %512 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %513 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %514 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %515 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %516 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %517 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %518 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %519 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x640x1x1xf16>) : !torch.vtensor<[1280,640,1,1],f16>
    %520 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %521 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280x3x3xf16>) : !torch.vtensor<[1280,1280,3,3],f16>
    %522 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %523 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %524 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %525 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %526 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %527 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x640x3x3xf16>) : !torch.vtensor<[1280,640,3,3],f16>
    %528 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %529 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %530 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %531 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640x3x3xf16>) : !torch.vtensor<[640,640,3,3],f16>
    %532 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %533 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %534 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %535 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x2560xf16>) : !torch.vtensor<[640,2560],f16>
    %536 = torch.vtensor.literal(dense_resource<__elided__> : tensor<5120xf16>) : !torch.vtensor<[5120],f16>
    %537 = torch.vtensor.literal(dense_resource<__elided__> : tensor<5120x640xf16>) : !torch.vtensor<[5120,640],f16>
    %538 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %539 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %540 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %541 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %542 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1024xf16>) : !torch.vtensor<[640,1024],f16>
    %543 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1024xf16>) : !torch.vtensor<[640,1024],f16>
    %544 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %545 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %546 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %547 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %548 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %549 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %550 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %551 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %552 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %553 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %554 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %555 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %556 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %557 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %558 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %559 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640x3x3xf16>) : !torch.vtensor<[640,640,3,3],f16>
    %560 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %561 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %562 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %563 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1280xf16>) : !torch.vtensor<[640,1280],f16>
    %564 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %565 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640x3x3xf16>) : !torch.vtensor<[640,640,3,3],f16>
    %566 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %567 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %568 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %569 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %570 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %571 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x2560xf16>) : !torch.vtensor<[640,2560],f16>
    %572 = torch.vtensor.literal(dense_resource<__elided__> : tensor<5120xf16>) : !torch.vtensor<[5120],f16>
    %573 = torch.vtensor.literal(dense_resource<__elided__> : tensor<5120x640xf16>) : !torch.vtensor<[5120,640],f16>
    %574 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %575 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %576 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %577 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %578 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1024xf16>) : !torch.vtensor<[640,1024],f16>
    %579 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1024xf16>) : !torch.vtensor<[640,1024],f16>
    %580 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %581 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %582 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %583 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %584 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %585 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %586 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %587 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %588 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %589 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %590 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %591 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640xf16>) : !torch.vtensor<[640,640],f16>
    %592 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %593 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %594 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %595 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x320x1x1xf16>) : !torch.vtensor<[640,320,1,1],f16>
    %596 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %597 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x640x3x3xf16>) : !torch.vtensor<[640,640,3,3],f16>
    %598 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %599 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %600 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %601 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x1280xf16>) : !torch.vtensor<[640,1280],f16>
    %602 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640xf16>) : !torch.vtensor<[640],f16>
    %603 = torch.vtensor.literal(dense_resource<__elided__> : tensor<640x320x3x3xf16>) : !torch.vtensor<[640,320,3,3],f16>
    %604 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %605 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %606 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %607 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320x3x3xf16>) : !torch.vtensor<[320,320,3,3],f16>
    %608 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %609 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %610 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %611 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1280xf16>) : !torch.vtensor<[320,1280],f16>
    %612 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %613 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560x320xf16>) : !torch.vtensor<[2560,320],f16>
    %614 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %615 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %616 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %617 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %618 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1024xf16>) : !torch.vtensor<[320,1024],f16>
    %619 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1024xf16>) : !torch.vtensor<[320,1024],f16>
    %620 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %621 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %622 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %623 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %624 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %625 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %626 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %627 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %628 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %629 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %630 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %631 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %632 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %633 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %634 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %635 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320x3x3xf16>) : !torch.vtensor<[320,320,3,3],f16>
    %636 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %637 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %638 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %639 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1280xf16>) : !torch.vtensor<[320,1280],f16>
    %640 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %641 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320x3x3xf16>) : !torch.vtensor<[320,320,3,3],f16>
    %642 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %643 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %644 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %645 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %646 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %647 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1280xf16>) : !torch.vtensor<[320,1280],f16>
    %648 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560xf16>) : !torch.vtensor<[2560],f16>
    %649 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2560x320xf16>) : !torch.vtensor<[2560,320],f16>
    %650 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %651 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %652 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %653 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %654 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1024xf16>) : !torch.vtensor<[320,1024],f16>
    %655 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1024xf16>) : !torch.vtensor<[320,1024],f16>
    %656 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %657 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %658 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %659 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %660 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %661 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %662 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %663 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %664 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %665 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %666 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %667 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320xf16>) : !torch.vtensor<[320,320],f16>
    %668 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %669 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %670 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %671 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320x3x3xf16>) : !torch.vtensor<[320,320,3,3],f16>
    %672 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %673 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %674 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %675 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x1280xf16>) : !torch.vtensor<[320,1280],f16>
    %676 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %677 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x320x3x3xf16>) : !torch.vtensor<[320,320,3,3],f16>
    %678 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %679 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %680 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320xf16>) : !torch.vtensor<[320],f16>
    %681 = torch.vtensor.literal(dense_resource<__elided__> : tensor<320x4x3x3xf16>) : !torch.vtensor<[320,4,3,3],f16>
    %682 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %683 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x1280xf16>) : !torch.vtensor<[1280,1280],f16>
    %684 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280xf16>) : !torch.vtensor<[1280],f16>
    %685 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1280x320xf16>) : !torch.vtensor<[1280,320],f16>
    %float-9.210340e00 = torch.constant.float -9.2103403719761836
    %int-1 = torch.constant.int -1
    %int6 = torch.constant.int 6
    %int-2 = torch.constant.int -2
    %str = torch.constant.str "none"
    %int2 = torch.constant.int 2
    %int9223372036854775807 = torch.constant.int 9223372036854775807
    %int32 = torch.constant.int 32
    %int3 = torch.constant.int 3
    %float1.000000e-05 = torch.constant.float 1.000000e-05
    %float9.999990e-07 = torch.constant.float 9.9999999999999995E-7
    %int8192 = torch.constant.int 8192
    %float1.250000e-01 = torch.constant.float 1.250000e-01
    %int128 = torch.constant.int 128
    %int2048 = torch.constant.int 2048
    %int16 = torch.constant.int 16
    %int512 = torch.constant.int 512
    %int8 = torch.constant.int 8
    %float2.000000e00 = torch.constant.float 2.000000e+00
    %int1920 = torch.constant.int 1920
    %int960 = torch.constant.int 960
    %cpu = torch.constant.device "cpu"
    %686 = torch.prim.ListConstruct %arg0, %arg0 : (!torch.vtensor<[1,4,64,64],f16>, !torch.vtensor<[1,4,64,64],f16>) -> !torch.list<vtensor>
    %687 = torch.aten.cat %686, %int0 : !torch.list<vtensor>, !torch.int -> !torch.vtensor<[2,4,64,64],f16>
    %688 = torch.prim.ListConstruct %int2 : (!torch.int) -> !torch.list<int>
    %689 = torch.aten.broadcast_to %arg1, %688 : !torch.vtensor<[1],f16>, !torch.list<int> -> !torch.vtensor<[2],f16>
    %690 = torch.aten.arange.start_step %int0, %int160, %int1, %int5, %none, %cpu, %false : !torch.int, !torch.int, !torch.int, !torch.int, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[160],f16>
    %691 = torch.aten.mul.Scalar %690, %float-9.210340e00 : !torch.vtensor<[160],f16>, !torch.float -> !torch.vtensor<[160],f16>
    %692 = torch.aten.div.Scalar %691, %int160 : !torch.vtensor<[160],f16>, !torch.int -> !torch.vtensor<[160],f16>
    %693 = torch.aten.exp %692 : !torch.vtensor<[160],f16> -> !torch.vtensor<[160],f16>
    %694 = torch.aten.unsqueeze %689, %int1 : !torch.vtensor<[2],f16>, !torch.int -> !torch.vtensor<[2,1],f16>
    %695 = torch.aten.unsqueeze %693, %int0 : !torch.vtensor<[160],f16>, !torch.int -> !torch.vtensor<[1,160],f16>
    %696 = torch.aten.mul.Tensor %694, %695 : !torch.vtensor<[2,1],f16>, !torch.vtensor<[1,160],f16> -> !torch.vtensor<[2,160],f16>
    %697 = torch.aten.mul.Scalar %696, %int1 : !torch.vtensor<[2,160],f16>, !torch.int -> !torch.vtensor<[2,160],f16>
    %698 = torch.aten.sin %697 : !torch.vtensor<[2,160],f16> -> !torch.vtensor<[2,160],f16>
    %699 = torch.aten.cos %697 : !torch.vtensor<[2,160],f16> -> !torch.vtensor<[2,160],f16>
    %700 = torch.prim.ListConstruct %698, %699 : (!torch.vtensor<[2,160],f16>, !torch.vtensor<[2,160],f16>) -> !torch.list<vtensor>
    %701 = torch.aten.cat %700, %int-1 : !torch.list<vtensor>, !torch.int -> !torch.vtensor<[2,320],f16>
    %702 = torch.aten.slice.Tensor %701, %int1, %int160, %int9223372036854775807, %int1 : !torch.vtensor<[2,320],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,160],f16>
    %703 = torch.aten.slice.Tensor %701, %int1, %int0, %int160, %int1 : !torch.vtensor<[2,320],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,160],f16>
    %704 = torch.prim.ListConstruct %702, %703 : (!torch.vtensor<[2,160],f16>, !torch.vtensor<[2,160],f16>) -> !torch.list<vtensor>
    %705 = torch.aten.cat %704, %int-1 : !torch.list<vtensor>, !torch.int -> !torch.vtensor<[2,320],f16>
    %706 = torch.aten.transpose.int %685, %int0, %int1 : !torch.vtensor<[1280,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,1280],f16>
    %707 = torch.aten.mm %705, %706 : !torch.vtensor<[2,320],f16>, !torch.vtensor<[320,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %708 = torch.aten.mul.Scalar %684, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %709 = torch.aten.add.Tensor %708, %707, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280],f16>
    %710 = torch.aten.sigmoid %709 : !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %711 = torch.aten.mul.Tensor %710, %709 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %712 = torch.aten.transpose.int %683, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %713 = torch.aten.mm %711, %712 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %714 = torch.aten.mul.Scalar %682, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %715 = torch.aten.add.Tensor %714, %713, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280],f16>
    %716 = torch.prim.ListConstruct %int1, %int1 : (!torch.int, !torch.int) -> !torch.list<int>
    %717 = torch.prim.ListConstruct %int0, %int0 : (!torch.int, !torch.int) -> !torch.list<int>
    %718 = torch.aten.convolution %687, %681, %680, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,4,64,64],f16>, !torch.vtensor<[320,4,3,3],f16>, !torch.vtensor<[320],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %719 = torch.prim.ListConstruct %int2, %int32, %int10, %int4096 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %720 = torch.aten.view %718, %719 : !torch.vtensor<[2,320,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,32,10,4096],f16>
    %721 = torch.prim.ListConstruct %int2, %int3 : (!torch.int, !torch.int) -> !torch.list<int>
    %722 = torch.aten.to.dtype %720, %int7, %false, %false, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,10,4096],f64>
    %723 = torch.aten.sum.dim_IntList %722, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %724 = torch.aten.div.Scalar %723, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %725 = torch.aten.sub.Tensor %722, %724, %float1.000000e00 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,10,4096],f64>
    %726 = torch.aten.mul.Tensor %725, %725 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,10,4096],f64> -> !torch.vtensor<[2,32,10,4096],f64>
    %727 = torch.aten.sum.dim_IntList %726, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %728 = torch.aten.div.Scalar %727, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %729 = torch.aten.to.dtype %728, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %730 = torch.aten.sum.dim_IntList %720, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %731 = torch.aten.div.Scalar %730, %int40960 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %732 = torch.aten.add.Scalar %729, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %733 = torch.aten.rsqrt %732 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %734 = torch.aten.sub.Tensor %720, %731, %int1 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,10,4096],f16>
    %735 = torch.aten.mul.Tensor %734, %733 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,10,4096],f16>
    %736 = torch.prim.ListConstruct %int2, %int320, %int64, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %737 = torch.aten.view %735, %736 : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int> -> !torch.vtensor<[2,320,64,64],f16>
    %738 = torch.aten.unsqueeze %679, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %739 = torch.aten.unsqueeze %738, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %740 = torch.aten.unsqueeze %739, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %741 = torch.aten.unsqueeze %678, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %742 = torch.aten.unsqueeze %741, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %743 = torch.aten.unsqueeze %742, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %744 = torch.aten.mul.Tensor %737, %743 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %745 = torch.aten.add.Tensor %744, %740, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %746 = torch.aten.sigmoid %745 : !torch.vtensor<[2,320,64,64],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %747 = torch.aten.mul.Tensor %746, %745 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[2,320,64,64],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %748 = torch.aten.convolution %747, %677, %676, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[320,320,3,3],f16>, !torch.vtensor<[320],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %749 = torch.aten.sigmoid %715 : !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %750 = torch.aten.mul.Tensor %749, %715 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %751 = torch.aten.transpose.int %675, %int0, %int1 : !torch.vtensor<[320,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,320],f16>
    %752 = torch.aten.mm %750, %751 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[1280,320],f16> -> !torch.vtensor<[2,320],f16>
    %753 = torch.aten.mul.Scalar %674, %int1 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[320],f16>
    %754 = torch.aten.add.Tensor %753, %752, %int1 : !torch.vtensor<[320],f16>, !torch.vtensor<[2,320],f16>, !torch.int -> !torch.vtensor<[2,320],f16>
    %755 = torch.aten.unsqueeze %754, %int2 : !torch.vtensor<[2,320],f16>, !torch.int -> !torch.vtensor<[2,320,1],f16>
    %756 = torch.aten.unsqueeze %755, %int3 : !torch.vtensor<[2,320,1],f16>, !torch.int -> !torch.vtensor<[2,320,1,1],f16>
    %757 = torch.aten.add.Tensor %748, %756, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[2,320,1,1],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %758 = torch.aten.view %757, %719 : !torch.vtensor<[2,320,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,32,10,4096],f16>
    %759 = torch.aten.to.dtype %758, %int7, %false, %false, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,10,4096],f64>
    %760 = torch.aten.sum.dim_IntList %759, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %761 = torch.aten.div.Scalar %760, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %762 = torch.aten.sub.Tensor %759, %761, %float1.000000e00 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,10,4096],f64>
    %763 = torch.aten.mul.Tensor %762, %762 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,10,4096],f64> -> !torch.vtensor<[2,32,10,4096],f64>
    %764 = torch.aten.sum.dim_IntList %763, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %765 = torch.aten.div.Scalar %764, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %766 = torch.aten.to.dtype %765, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %767 = torch.aten.sum.dim_IntList %758, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %768 = torch.aten.div.Scalar %767, %int40960 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %769 = torch.aten.add.Scalar %766, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %770 = torch.aten.rsqrt %769 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %771 = torch.aten.sub.Tensor %758, %768, %int1 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,10,4096],f16>
    %772 = torch.aten.mul.Tensor %771, %770 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,10,4096],f16>
    %773 = torch.aten.view %772, %736 : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int> -> !torch.vtensor<[2,320,64,64],f16>
    %774 = torch.aten.unsqueeze %673, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %775 = torch.aten.unsqueeze %774, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %776 = torch.aten.unsqueeze %775, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %777 = torch.aten.unsqueeze %672, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %778 = torch.aten.unsqueeze %777, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %779 = torch.aten.unsqueeze %778, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %780 = torch.aten.mul.Tensor %773, %779 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %781 = torch.aten.add.Tensor %780, %776, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %782 = torch.aten.sigmoid %781 : !torch.vtensor<[2,320,64,64],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %783 = torch.aten.mul.Tensor %782, %781 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[2,320,64,64],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %784 = torch.aten.convolution %783, %671, %670, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[320,320,3,3],f16>, !torch.vtensor<[320],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %785 = torch.aten.add.Tensor %718, %784, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[2,320,64,64],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %786 = torch.aten.div.Scalar %785, %float1.000000e00 : !torch.vtensor<[2,320,64,64],f16>, !torch.float -> !torch.vtensor<[2,320,64,64],f16>
    %787 = torch.aten.view %786, %719 : !torch.vtensor<[2,320,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,32,10,4096],f16>
    %788 = torch.aten.to.dtype %787, %int7, %false, %false, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,10,4096],f64>
    %789 = torch.aten.sum.dim_IntList %788, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %790 = torch.aten.div.Scalar %789, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %791 = torch.aten.sub.Tensor %788, %790, %float1.000000e00 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,10,4096],f64>
    %792 = torch.aten.mul.Tensor %791, %791 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,10,4096],f64> -> !torch.vtensor<[2,32,10,4096],f64>
    %793 = torch.aten.sum.dim_IntList %792, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %794 = torch.aten.div.Scalar %793, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %795 = torch.aten.to.dtype %794, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %796 = torch.aten.sum.dim_IntList %787, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %797 = torch.aten.div.Scalar %796, %int40960 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %798 = torch.aten.add.Scalar %795, %float9.999990e-07, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %799 = torch.aten.rsqrt %798 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %800 = torch.aten.sub.Tensor %787, %797, %int1 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,10,4096],f16>
    %801 = torch.aten.mul.Tensor %800, %799 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,10,4096],f16>
    %802 = torch.aten.view %801, %736 : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int> -> !torch.vtensor<[2,320,64,64],f16>
    %803 = torch.aten.unsqueeze %669, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %804 = torch.aten.unsqueeze %803, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %805 = torch.aten.unsqueeze %804, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %806 = torch.aten.unsqueeze %668, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %807 = torch.aten.unsqueeze %806, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %808 = torch.aten.unsqueeze %807, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %809 = torch.aten.mul.Tensor %802, %808 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %810 = torch.aten.add.Tensor %809, %805, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %811 = torch.prim.ListConstruct %int0, %int2, %int3, %int1 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %812 = torch.aten.permute %810, %811 : !torch.vtensor<[2,320,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,64,64,320],f16>
    %813 = torch.prim.ListConstruct %int2, %int4096, %int320 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %814 = torch.aten.view %812, %813 : !torch.vtensor<[2,64,64,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %815 = torch.aten.transpose.int %667, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %816 = torch.aten.broadcast_to %814, %813 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %817 = torch.aten.view %816, %813 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %818 = torch.prim.ListConstruct %int2, %int320, %int320 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %819 = torch.aten.broadcast_to %815, %818 : !torch.vtensor<[320,320],f16>, !torch.list<int> -> !torch.vtensor<[2,320,320],f16>
    %820 = torch.aten.view %819, %818 : !torch.vtensor<[2,320,320],f16>, !torch.list<int> -> !torch.vtensor<[2,320,320],f16>
    %821 = torch.aten.bmm %817, %820 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,320,320],f16> -> !torch.vtensor<[2,4096,320],f16>
    %822 = torch.aten.view %821, %813 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %823 = torch.aten.add.Tensor %822, %666, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %824 = torch.aten.to.dtype %823, %int7, %false, %false, %none : !torch.vtensor<[2,4096,320],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,320],f64>
    %825 = torch.aten.sum.dim_IntList %824, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %826 = torch.aten.div.Scalar %825, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %827 = torch.aten.sub.Tensor %824, %826, %float1.000000e00 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,1],f64>, !torch.float -> !torch.vtensor<[2,4096,320],f64>
    %828 = torch.aten.mul.Tensor %827, %827 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,320],f64> -> !torch.vtensor<[2,4096,320],f64>
    %829 = torch.aten.sum.dim_IntList %828, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %830 = torch.aten.div.Scalar %829, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %831 = torch.aten.to.dtype %830, %int5, %false, %false, %none : !torch.vtensor<[2,4096,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %832 = torch.aten.sum.dim_IntList %823, %688, %true, %none : !torch.vtensor<[2,4096,320],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %833 = torch.aten.div.Scalar %832, %int320 : !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %834 = torch.aten.add.Scalar %831, %float1.000000e-05, %int1 : !torch.vtensor<[2,4096,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %835 = torch.aten.rsqrt %834 : !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,1],f16>
    %836 = torch.aten.sub.Tensor %823, %833, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %837 = torch.aten.mul.Tensor %836, %835 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,320],f16>
    %838 = torch.aten.mul.Tensor %837, %665 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16> -> !torch.vtensor<[2,4096,320],f16>
    %839 = torch.aten.add.Tensor %838, %664, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %840 = torch.aten.transpose.int %663, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %841 = torch.prim.ListConstruct %int8192, %int320 : (!torch.int, !torch.int) -> !torch.list<int>
    %842 = torch.aten.view %839, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %843 = torch.aten.mm %842, %840 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %844 = torch.aten.view %843, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %845 = torch.prim.ListConstruct %int2, %int4096, %int5, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %846 = torch.aten.view %844, %845 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %847 = torch.prim.ListConstruct %int0, %int2, %int1, %int3 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %848 = torch.aten.permute %846, %847 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %849 = torch.aten.clone %848, %int0 : !torch.vtensor<[2,5,4096,64],f16>, !torch.int -> !torch.vtensor<[2,5,4096,64],f16>
    %850 = torch.prim.ListConstruct %int10, %int4096, %int64 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %851 = torch.aten.view %849, %850 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[10,4096,64],f16>
    %852 = torch.aten.transpose.int %662, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %853 = torch.aten.view %839, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %854 = torch.aten.mm %853, %852 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %855 = torch.aten.view %854, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %856 = torch.aten.transpose.int %661, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %857 = torch.aten.view %839, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %858 = torch.aten.mm %857, %856 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %859 = torch.aten.view %858, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %860 = torch.aten.view %855, %845 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %861 = torch.aten.permute %860, %847 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %862 = torch.aten.clone %861, %int0 : !torch.vtensor<[2,5,4096,64],f16>, !torch.int -> !torch.vtensor<[2,5,4096,64],f16>
    %863 = torch.aten.view %862, %850 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[10,4096,64],f16>
    %864 = torch.aten.view %859, %845 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %865 = torch.aten.permute %864, %847 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %866 = torch.aten.clone %865, %int0 : !torch.vtensor<[2,5,4096,64],f16>, !torch.int -> !torch.vtensor<[2,5,4096,64],f16>
    %867 = torch.aten.view %866, %850 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[10,4096,64],f16>
    %868 = torch.prim.ListConstruct %int10, %int4096, %int4096 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %869 = torch.aten.empty.memory_format %868, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,4096],f32>
    %870 = torch.aten.transpose.int %863, %int-1, %int-2 : !torch.vtensor<[10,4096,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[10,64,4096],f16>
    %871 = torch.aten.bmm %851, %870 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,64,4096],f16> -> !torch.vtensor<[10,4096,4096],f16>
    %872 = torch.aten.mul.Scalar %871, %float1.250000e-01 : !torch.vtensor<[10,4096,4096],f16>, !torch.float -> !torch.vtensor<[10,4096,4096],f16>
    %873 = torch.aten.to.dtype %869, %int5, %false, %false, %none : !torch.vtensor<[10,4096,4096],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,4096],f16>
    %874 = torch.aten.add.Tensor %872, %869, %int0 : !torch.vtensor<[10,4096,4096],f16>, !torch.vtensor<[10,4096,4096],f32>, !torch.int -> !torch.vtensor<[10,4096,4096],f16>
    %values, %indices = torch.aten.max.dim %874, %int-1, %true : !torch.vtensor<[10,4096,4096],f16>, !torch.int, !torch.bool -> !torch.vtensor<[10,4096,1],f16>, !torch.vtensor<[10,4096,1],si64>
    %875 = torch.aten.sub.Tensor %874, %values, %float1.000000e00 : !torch.vtensor<[10,4096,4096],f16>, !torch.vtensor<[10,4096,1],f16>, !torch.float -> !torch.vtensor<[10,4096,4096],f16>
    %876 = torch.aten.exp %875 : !torch.vtensor<[10,4096,4096],f16> -> !torch.vtensor<[10,4096,4096],f16>
    %877 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %878 = torch.aten.sum.dim_IntList %876, %877, %true, %none : !torch.vtensor<[10,4096,4096],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,1],f16>
    %879 = torch.aten.div.Tensor %876, %878 : !torch.vtensor<[10,4096,4096],f16>, !torch.vtensor<[10,4096,1],f16> -> !torch.vtensor<[10,4096,4096],f16>
    %880 = torch.aten.bmm %879, %867 : !torch.vtensor<[10,4096,4096],f16>, !torch.vtensor<[10,4096,64],f16> -> !torch.vtensor<[10,4096,64],f16>
    %881 = torch.prim.ListConstruct %int2, %int5, %int4096, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %882 = torch.aten.view %880, %881 : !torch.vtensor<[10,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %883 = torch.aten.permute %882, %847 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %884 = torch.aten.clone %883, %int0 : !torch.vtensor<[2,4096,5,64],f16>, !torch.int -> !torch.vtensor<[2,4096,5,64],f16>
    %885 = torch.aten.view %884, %813 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %886 = torch.aten.transpose.int %660, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %887 = torch.aten.view %885, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %888 = torch.aten.mm %887, %886 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %889 = torch.aten.mul.Scalar %659, %int1 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[320],f16>
    %890 = torch.aten.add.Tensor %889, %888, %int1 : !torch.vtensor<[320],f16>, !torch.vtensor<[8192,320],f16>, !torch.int -> !torch.vtensor<[8192,320],f16>
    %891 = torch.aten.view %890, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %892 = torch.aten.add.Tensor %891, %823, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %893 = torch.aten.to.dtype %892, %int7, %false, %false, %none : !torch.vtensor<[2,4096,320],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,320],f64>
    %894 = torch.aten.sum.dim_IntList %893, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %895 = torch.aten.div.Scalar %894, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %896 = torch.aten.sub.Tensor %893, %895, %float1.000000e00 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,1],f64>, !torch.float -> !torch.vtensor<[2,4096,320],f64>
    %897 = torch.aten.mul.Tensor %896, %896 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,320],f64> -> !torch.vtensor<[2,4096,320],f64>
    %898 = torch.aten.sum.dim_IntList %897, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %899 = torch.aten.div.Scalar %898, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %900 = torch.aten.to.dtype %899, %int5, %false, %false, %none : !torch.vtensor<[2,4096,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %901 = torch.aten.sum.dim_IntList %892, %688, %true, %none : !torch.vtensor<[2,4096,320],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %902 = torch.aten.div.Scalar %901, %int320 : !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %903 = torch.aten.add.Scalar %900, %float1.000000e-05, %int1 : !torch.vtensor<[2,4096,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %904 = torch.aten.rsqrt %903 : !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,1],f16>
    %905 = torch.aten.sub.Tensor %892, %902, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %906 = torch.aten.mul.Tensor %905, %904 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,320],f16>
    %907 = torch.aten.mul.Tensor %906, %658 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16> -> !torch.vtensor<[2,4096,320],f16>
    %908 = torch.aten.add.Tensor %907, %657, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %909 = torch.aten.transpose.int %656, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %910 = torch.aten.view %908, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %911 = torch.aten.mm %910, %909 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %912 = torch.aten.view %911, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %913 = torch.aten.view %912, %845 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %914 = torch.aten.permute %913, %847 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %915 = torch.aten.clone %914, %int0 : !torch.vtensor<[2,5,4096,64],f16>, !torch.int -> !torch.vtensor<[2,5,4096,64],f16>
    %916 = torch.aten.view %915, %850 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[10,4096,64],f16>
    %917 = torch.aten.transpose.int %655, %int0, %int1 : !torch.vtensor<[320,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,320],f16>
    %918 = torch.prim.ListConstruct %int128, %int1024 : (!torch.int, !torch.int) -> !torch.list<int>
    %919 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %920 = torch.aten.mm %919, %917 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,320],f16> -> !torch.vtensor<[128,320],f16>
    %921 = torch.prim.ListConstruct %int2, %int64, %int320 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %922 = torch.aten.view %920, %921 : !torch.vtensor<[128,320],f16>, !torch.list<int> -> !torch.vtensor<[2,64,320],f16>
    %923 = torch.aten.transpose.int %654, %int0, %int1 : !torch.vtensor<[320,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,320],f16>
    %924 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %925 = torch.aten.mm %924, %923 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,320],f16> -> !torch.vtensor<[128,320],f16>
    %926 = torch.aten.view %925, %921 : !torch.vtensor<[128,320],f16>, !torch.list<int> -> !torch.vtensor<[2,64,320],f16>
    %927 = torch.prim.ListConstruct %int2, %int64, %int5, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %928 = torch.aten.view %922, %927 : !torch.vtensor<[2,64,320],f16>, !torch.list<int> -> !torch.vtensor<[2,64,5,64],f16>
    %929 = torch.aten.permute %928, %847 : !torch.vtensor<[2,64,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,64,64],f16>
    %930 = torch.aten.clone %929, %int0 : !torch.vtensor<[2,5,64,64],f16>, !torch.int -> !torch.vtensor<[2,5,64,64],f16>
    %931 = torch.prim.ListConstruct %int10, %int64, %int64 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %932 = torch.aten.view %930, %931 : !torch.vtensor<[2,5,64,64],f16>, !torch.list<int> -> !torch.vtensor<[10,64,64],f16>
    %933 = torch.aten.view %926, %927 : !torch.vtensor<[2,64,320],f16>, !torch.list<int> -> !torch.vtensor<[2,64,5,64],f16>
    %934 = torch.aten.permute %933, %847 : !torch.vtensor<[2,64,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,64,64],f16>
    %935 = torch.aten.clone %934, %int0 : !torch.vtensor<[2,5,64,64],f16>, !torch.int -> !torch.vtensor<[2,5,64,64],f16>
    %936 = torch.aten.view %935, %931 : !torch.vtensor<[2,5,64,64],f16>, !torch.list<int> -> !torch.vtensor<[10,64,64],f16>
    %937 = torch.aten.empty.memory_format %850, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,64],f32>
    %938 = torch.aten.transpose.int %932, %int-1, %int-2 : !torch.vtensor<[10,64,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[10,64,64],f16>
    %939 = torch.aten.bmm %916, %938 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,64,64],f16> -> !torch.vtensor<[10,4096,64],f16>
    %940 = torch.aten.mul.Scalar %939, %float1.250000e-01 : !torch.vtensor<[10,4096,64],f16>, !torch.float -> !torch.vtensor<[10,4096,64],f16>
    %941 = torch.aten.to.dtype %937, %int5, %false, %false, %none : !torch.vtensor<[10,4096,64],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,64],f16>
    %942 = torch.aten.add.Tensor %940, %937, %int0 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,4096,64],f32>, !torch.int -> !torch.vtensor<[10,4096,64],f16>
    %values_0, %indices_1 = torch.aten.max.dim %942, %int-1, %true : !torch.vtensor<[10,4096,64],f16>, !torch.int, !torch.bool -> !torch.vtensor<[10,4096,1],f16>, !torch.vtensor<[10,4096,1],si64>
    %943 = torch.aten.sub.Tensor %942, %values_0, %float1.000000e00 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,4096,1],f16>, !torch.float -> !torch.vtensor<[10,4096,64],f16>
    %944 = torch.aten.exp %943 : !torch.vtensor<[10,4096,64],f16> -> !torch.vtensor<[10,4096,64],f16>
    %945 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %946 = torch.aten.sum.dim_IntList %944, %945, %true, %none : !torch.vtensor<[10,4096,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,1],f16>
    %947 = torch.aten.div.Tensor %944, %946 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,4096,1],f16> -> !torch.vtensor<[10,4096,64],f16>
    %948 = torch.aten.bmm %947, %936 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,64,64],f16> -> !torch.vtensor<[10,4096,64],f16>
    %949 = torch.aten.view %948, %881 : !torch.vtensor<[10,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %950 = torch.aten.permute %949, %847 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %951 = torch.aten.clone %950, %int0 : !torch.vtensor<[2,4096,5,64],f16>, !torch.int -> !torch.vtensor<[2,4096,5,64],f16>
    %952 = torch.aten.view %951, %813 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %953 = torch.aten.transpose.int %653, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %954 = torch.aten.view %952, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %955 = torch.aten.mm %954, %953 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %956 = torch.aten.mul.Scalar %652, %int1 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[320],f16>
    %957 = torch.aten.add.Tensor %956, %955, %int1 : !torch.vtensor<[320],f16>, !torch.vtensor<[8192,320],f16>, !torch.int -> !torch.vtensor<[8192,320],f16>
    %958 = torch.aten.view %957, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %959 = torch.aten.add.Tensor %958, %892, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %960 = torch.aten.to.dtype %959, %int7, %false, %false, %none : !torch.vtensor<[2,4096,320],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,320],f64>
    %961 = torch.aten.sum.dim_IntList %960, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %962 = torch.aten.div.Scalar %961, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %963 = torch.aten.sub.Tensor %960, %962, %float1.000000e00 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,1],f64>, !torch.float -> !torch.vtensor<[2,4096,320],f64>
    %964 = torch.aten.mul.Tensor %963, %963 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,320],f64> -> !torch.vtensor<[2,4096,320],f64>
    %965 = torch.aten.sum.dim_IntList %964, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %966 = torch.aten.div.Scalar %965, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %967 = torch.aten.to.dtype %966, %int5, %false, %false, %none : !torch.vtensor<[2,4096,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %968 = torch.aten.sum.dim_IntList %959, %688, %true, %none : !torch.vtensor<[2,4096,320],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %969 = torch.aten.div.Scalar %968, %int320 : !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %970 = torch.aten.add.Scalar %967, %float1.000000e-05, %int1 : !torch.vtensor<[2,4096,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %971 = torch.aten.rsqrt %970 : !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,1],f16>
    %972 = torch.aten.sub.Tensor %959, %969, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %973 = torch.aten.mul.Tensor %972, %971 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,320],f16>
    %974 = torch.aten.mul.Tensor %973, %651 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16> -> !torch.vtensor<[2,4096,320],f16>
    %975 = torch.aten.add.Tensor %974, %650, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %976 = torch.aten.transpose.int %649, %int0, %int1 : !torch.vtensor<[2560,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,2560],f16>
    %977 = torch.aten.view %975, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %978 = torch.aten.mm %977, %976 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,2560],f16> -> !torch.vtensor<[8192,2560],f16>
    %979 = torch.aten.mul.Scalar %648, %int1 : !torch.vtensor<[2560],f16>, !torch.int -> !torch.vtensor<[2560],f16>
    %980 = torch.aten.add.Tensor %979, %978, %int1 : !torch.vtensor<[2560],f16>, !torch.vtensor<[8192,2560],f16>, !torch.int -> !torch.vtensor<[8192,2560],f16>
    %981 = torch.prim.ListConstruct %int2, %int4096, %int2560 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %982 = torch.aten.view %980, %981 : !torch.vtensor<[8192,2560],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,2560],f16>
    %983 = torch.aten.slice.Tensor %982, %int-1, %int0, %int1280, %int1 : !torch.vtensor<[2,4096,2560],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,4096,1280],f16>
    %984 = torch.aten.slice.Tensor %982, %int-1, %int1280, %int2560, %int1 : !torch.vtensor<[2,4096,2560],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,4096,1280],f16>
    %985 = torch.aten.gelu %984, %str : !torch.vtensor<[2,4096,1280],f16>, !torch.str -> !torch.vtensor<[2,4096,1280],f16>
    %986 = torch.aten.mul.Tensor %983, %985 : !torch.vtensor<[2,4096,1280],f16>, !torch.vtensor<[2,4096,1280],f16> -> !torch.vtensor<[2,4096,1280],f16>
    %987 = torch.aten.transpose.int %647, %int0, %int1 : !torch.vtensor<[320,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,320],f16>
    %988 = torch.prim.ListConstruct %int8192, %int1280 : (!torch.int, !torch.int) -> !torch.list<int>
    %989 = torch.aten.view %986, %988 : !torch.vtensor<[2,4096,1280],f16>, !torch.list<int> -> !torch.vtensor<[8192,1280],f16>
    %990 = torch.aten.mm %989, %987 : !torch.vtensor<[8192,1280],f16>, !torch.vtensor<[1280,320],f16> -> !torch.vtensor<[8192,320],f16>
    %991 = torch.aten.mul.Scalar %646, %int1 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[320],f16>
    %992 = torch.aten.add.Tensor %991, %990, %int1 : !torch.vtensor<[320],f16>, !torch.vtensor<[8192,320],f16>, !torch.int -> !torch.vtensor<[8192,320],f16>
    %993 = torch.aten.view %992, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %994 = torch.aten.add.Tensor %993, %959, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %995 = torch.aten.transpose.int %645, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %996 = torch.aten.view %994, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %997 = torch.aten.mm %996, %995 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %998 = torch.aten.mul.Scalar %644, %int1 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[320],f16>
    %999 = torch.aten.add.Tensor %998, %997, %int1 : !torch.vtensor<[320],f16>, !torch.vtensor<[8192,320],f16>, !torch.int -> !torch.vtensor<[8192,320],f16>
    %1000 = torch.aten.view %999, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1001 = torch.prim.ListConstruct %int2, %int64, %int64, %int320 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1002 = torch.aten.view %1000, %1001 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,64,64,320],f16>
    %1003 = torch.prim.ListConstruct %int0, %int3, %int1, %int2 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1004 = torch.aten.permute %1002, %1003 : !torch.vtensor<[2,64,64,320],f16>, !torch.list<int> -> !torch.vtensor<[2,320,64,64],f16>
    %1005 = torch.aten.clone %1004, %int0 : !torch.vtensor<[2,320,64,64],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1006 = torch.aten.add.Tensor %1005, %786, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[2,320,64,64],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1007 = torch.aten.view %1006, %719 : !torch.vtensor<[2,320,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,32,10,4096],f16>
    %1008 = torch.aten.to.dtype %1007, %int7, %false, %false, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,10,4096],f64>
    %1009 = torch.aten.sum.dim_IntList %1008, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1010 = torch.aten.div.Scalar %1009, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1011 = torch.aten.sub.Tensor %1008, %1010, %float1.000000e00 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,10,4096],f64>
    %1012 = torch.aten.mul.Tensor %1011, %1011 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,10,4096],f64> -> !torch.vtensor<[2,32,10,4096],f64>
    %1013 = torch.aten.sum.dim_IntList %1012, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1014 = torch.aten.div.Scalar %1013, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1015 = torch.aten.to.dtype %1014, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1016 = torch.aten.sum.dim_IntList %1007, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1017 = torch.aten.div.Scalar %1016, %int40960 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1018 = torch.aten.add.Scalar %1015, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1019 = torch.aten.rsqrt %1018 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1020 = torch.aten.sub.Tensor %1007, %1017, %int1 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,10,4096],f16>
    %1021 = torch.aten.mul.Tensor %1020, %1019 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,10,4096],f16>
    %1022 = torch.aten.view %1021, %736 : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int> -> !torch.vtensor<[2,320,64,64],f16>
    %1023 = torch.aten.unsqueeze %643, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %1024 = torch.aten.unsqueeze %1023, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %1025 = torch.aten.unsqueeze %1024, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %1026 = torch.aten.unsqueeze %642, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %1027 = torch.aten.unsqueeze %1026, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %1028 = torch.aten.unsqueeze %1027, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %1029 = torch.aten.mul.Tensor %1022, %1028 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %1030 = torch.aten.add.Tensor %1029, %1025, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1031 = torch.aten.sigmoid %1030 : !torch.vtensor<[2,320,64,64],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %1032 = torch.aten.mul.Tensor %1031, %1030 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[2,320,64,64],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %1033 = torch.aten.convolution %1032, %641, %640, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[320,320,3,3],f16>, !torch.vtensor<[320],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1034 = torch.aten.sigmoid %715 : !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %1035 = torch.aten.mul.Tensor %1034, %715 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %1036 = torch.aten.transpose.int %639, %int0, %int1 : !torch.vtensor<[320,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,320],f16>
    %1037 = torch.aten.mm %1035, %1036 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[1280,320],f16> -> !torch.vtensor<[2,320],f16>
    %1038 = torch.aten.mul.Scalar %638, %int1 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[320],f16>
    %1039 = torch.aten.add.Tensor %1038, %1037, %int1 : !torch.vtensor<[320],f16>, !torch.vtensor<[2,320],f16>, !torch.int -> !torch.vtensor<[2,320],f16>
    %1040 = torch.aten.unsqueeze %1039, %int2 : !torch.vtensor<[2,320],f16>, !torch.int -> !torch.vtensor<[2,320,1],f16>
    %1041 = torch.aten.unsqueeze %1040, %int3 : !torch.vtensor<[2,320,1],f16>, !torch.int -> !torch.vtensor<[2,320,1,1],f16>
    %1042 = torch.aten.add.Tensor %1033, %1041, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[2,320,1,1],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1043 = torch.aten.view %1042, %719 : !torch.vtensor<[2,320,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,32,10,4096],f16>
    %1044 = torch.aten.to.dtype %1043, %int7, %false, %false, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,10,4096],f64>
    %1045 = torch.aten.sum.dim_IntList %1044, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1046 = torch.aten.div.Scalar %1045, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1047 = torch.aten.sub.Tensor %1044, %1046, %float1.000000e00 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,10,4096],f64>
    %1048 = torch.aten.mul.Tensor %1047, %1047 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,10,4096],f64> -> !torch.vtensor<[2,32,10,4096],f64>
    %1049 = torch.aten.sum.dim_IntList %1048, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1050 = torch.aten.div.Scalar %1049, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1051 = torch.aten.to.dtype %1050, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1052 = torch.aten.sum.dim_IntList %1043, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1053 = torch.aten.div.Scalar %1052, %int40960 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1054 = torch.aten.add.Scalar %1051, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1055 = torch.aten.rsqrt %1054 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1056 = torch.aten.sub.Tensor %1043, %1053, %int1 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,10,4096],f16>
    %1057 = torch.aten.mul.Tensor %1056, %1055 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,10,4096],f16>
    %1058 = torch.aten.view %1057, %736 : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int> -> !torch.vtensor<[2,320,64,64],f16>
    %1059 = torch.aten.unsqueeze %637, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %1060 = torch.aten.unsqueeze %1059, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %1061 = torch.aten.unsqueeze %1060, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %1062 = torch.aten.unsqueeze %636, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %1063 = torch.aten.unsqueeze %1062, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %1064 = torch.aten.unsqueeze %1063, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %1065 = torch.aten.mul.Tensor %1058, %1064 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %1066 = torch.aten.add.Tensor %1065, %1061, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1067 = torch.aten.sigmoid %1066 : !torch.vtensor<[2,320,64,64],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %1068 = torch.aten.mul.Tensor %1067, %1066 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[2,320,64,64],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %1069 = torch.aten.convolution %1068, %635, %634, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[320,320,3,3],f16>, !torch.vtensor<[320],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1070 = torch.aten.add.Tensor %1006, %1069, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[2,320,64,64],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1071 = torch.aten.div.Scalar %1070, %float1.000000e00 : !torch.vtensor<[2,320,64,64],f16>, !torch.float -> !torch.vtensor<[2,320,64,64],f16>
    %1072 = torch.aten.view %1071, %719 : !torch.vtensor<[2,320,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,32,10,4096],f16>
    %1073 = torch.aten.to.dtype %1072, %int7, %false, %false, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,10,4096],f64>
    %1074 = torch.aten.sum.dim_IntList %1073, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1075 = torch.aten.div.Scalar %1074, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1076 = torch.aten.sub.Tensor %1073, %1075, %float1.000000e00 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,10,4096],f64>
    %1077 = torch.aten.mul.Tensor %1076, %1076 : !torch.vtensor<[2,32,10,4096],f64>, !torch.vtensor<[2,32,10,4096],f64> -> !torch.vtensor<[2,32,10,4096],f64>
    %1078 = torch.aten.sum.dim_IntList %1077, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1079 = torch.aten.div.Scalar %1078, %int40960 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1080 = torch.aten.to.dtype %1079, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1081 = torch.aten.sum.dim_IntList %1072, %721, %true, %none : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1082 = torch.aten.div.Scalar %1081, %int40960 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1083 = torch.aten.add.Scalar %1080, %float9.999990e-07, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1084 = torch.aten.rsqrt %1083 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1085 = torch.aten.sub.Tensor %1072, %1082, %int1 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,10,4096],f16>
    %1086 = torch.aten.mul.Tensor %1085, %1084 : !torch.vtensor<[2,32,10,4096],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,10,4096],f16>
    %1087 = torch.aten.view %1086, %736 : !torch.vtensor<[2,32,10,4096],f16>, !torch.list<int> -> !torch.vtensor<[2,320,64,64],f16>
    %1088 = torch.aten.unsqueeze %633, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %1089 = torch.aten.unsqueeze %1088, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %1090 = torch.aten.unsqueeze %1089, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %1091 = torch.aten.unsqueeze %632, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %1092 = torch.aten.unsqueeze %1091, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %1093 = torch.aten.unsqueeze %1092, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %1094 = torch.aten.mul.Tensor %1087, %1093 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16> -> !torch.vtensor<[2,320,64,64],f16>
    %1095 = torch.aten.add.Tensor %1094, %1090, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[1,320,1,1],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1096 = torch.aten.permute %1095, %811 : !torch.vtensor<[2,320,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,64,64,320],f16>
    %1097 = torch.aten.view %1096, %813 : !torch.vtensor<[2,64,64,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1098 = torch.aten.transpose.int %631, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %1099 = torch.aten.broadcast_to %1097, %813 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1100 = torch.aten.view %1099, %813 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1101 = torch.aten.broadcast_to %1098, %818 : !torch.vtensor<[320,320],f16>, !torch.list<int> -> !torch.vtensor<[2,320,320],f16>
    %1102 = torch.aten.view %1101, %818 : !torch.vtensor<[2,320,320],f16>, !torch.list<int> -> !torch.vtensor<[2,320,320],f16>
    %1103 = torch.aten.bmm %1100, %1102 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,320,320],f16> -> !torch.vtensor<[2,4096,320],f16>
    %1104 = torch.aten.view %1103, %813 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1105 = torch.aten.add.Tensor %1104, %630, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %1106 = torch.aten.to.dtype %1105, %int7, %false, %false, %none : !torch.vtensor<[2,4096,320],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,320],f64>
    %1107 = torch.aten.sum.dim_IntList %1106, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %1108 = torch.aten.div.Scalar %1107, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %1109 = torch.aten.sub.Tensor %1106, %1108, %float1.000000e00 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,1],f64>, !torch.float -> !torch.vtensor<[2,4096,320],f64>
    %1110 = torch.aten.mul.Tensor %1109, %1109 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,320],f64> -> !torch.vtensor<[2,4096,320],f64>
    %1111 = torch.aten.sum.dim_IntList %1110, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %1112 = torch.aten.div.Scalar %1111, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %1113 = torch.aten.to.dtype %1112, %int5, %false, %false, %none : !torch.vtensor<[2,4096,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %1114 = torch.aten.sum.dim_IntList %1105, %688, %true, %none : !torch.vtensor<[2,4096,320],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %1115 = torch.aten.div.Scalar %1114, %int320 : !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %1116 = torch.aten.add.Scalar %1113, %float1.000000e-05, %int1 : !torch.vtensor<[2,4096,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %1117 = torch.aten.rsqrt %1116 : !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,1],f16>
    %1118 = torch.aten.sub.Tensor %1105, %1115, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %1119 = torch.aten.mul.Tensor %1118, %1117 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,320],f16>
    %1120 = torch.aten.mul.Tensor %1119, %629 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16> -> !torch.vtensor<[2,4096,320],f16>
    %1121 = torch.aten.add.Tensor %1120, %628, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %1122 = torch.aten.transpose.int %627, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %1123 = torch.aten.view %1121, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %1124 = torch.aten.mm %1123, %1122 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %1125 = torch.aten.view %1124, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1126 = torch.aten.view %1125, %845 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %1127 = torch.aten.permute %1126, %847 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %1128 = torch.aten.clone %1127, %int0 : !torch.vtensor<[2,5,4096,64],f16>, !torch.int -> !torch.vtensor<[2,5,4096,64],f16>
    %1129 = torch.aten.view %1128, %850 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[10,4096,64],f16>
    %1130 = torch.aten.transpose.int %626, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %1131 = torch.aten.view %1121, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %1132 = torch.aten.mm %1131, %1130 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %1133 = torch.aten.view %1132, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1134 = torch.aten.transpose.int %625, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %1135 = torch.aten.view %1121, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %1136 = torch.aten.mm %1135, %1134 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %1137 = torch.aten.view %1136, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1138 = torch.aten.view %1133, %845 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %1139 = torch.aten.permute %1138, %847 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %1140 = torch.aten.clone %1139, %int0 : !torch.vtensor<[2,5,4096,64],f16>, !torch.int -> !torch.vtensor<[2,5,4096,64],f16>
    %1141 = torch.aten.view %1140, %850 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[10,4096,64],f16>
    %1142 = torch.aten.view %1137, %845 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %1143 = torch.aten.permute %1142, %847 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %1144 = torch.aten.clone %1143, %int0 : !torch.vtensor<[2,5,4096,64],f16>, !torch.int -> !torch.vtensor<[2,5,4096,64],f16>
    %1145 = torch.aten.view %1144, %850 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[10,4096,64],f16>
    %1146 = torch.aten.empty.memory_format %868, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,4096],f32>
    %1147 = torch.aten.transpose.int %1141, %int-1, %int-2 : !torch.vtensor<[10,4096,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[10,64,4096],f16>
    %1148 = torch.aten.bmm %1129, %1147 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,64,4096],f16> -> !torch.vtensor<[10,4096,4096],f16>
    %1149 = torch.aten.mul.Scalar %1148, %float1.250000e-01 : !torch.vtensor<[10,4096,4096],f16>, !torch.float -> !torch.vtensor<[10,4096,4096],f16>
    %1150 = torch.aten.to.dtype %1146, %int5, %false, %false, %none : !torch.vtensor<[10,4096,4096],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,4096],f16>
    %1151 = torch.aten.add.Tensor %1149, %1146, %int0 : !torch.vtensor<[10,4096,4096],f16>, !torch.vtensor<[10,4096,4096],f32>, !torch.int -> !torch.vtensor<[10,4096,4096],f16>
    %values_2, %indices_3 = torch.aten.max.dim %1151, %int-1, %true : !torch.vtensor<[10,4096,4096],f16>, !torch.int, !torch.bool -> !torch.vtensor<[10,4096,1],f16>, !torch.vtensor<[10,4096,1],si64>
    %1152 = torch.aten.sub.Tensor %1151, %values_2, %float1.000000e00 : !torch.vtensor<[10,4096,4096],f16>, !torch.vtensor<[10,4096,1],f16>, !torch.float -> !torch.vtensor<[10,4096,4096],f16>
    %1153 = torch.aten.exp %1152 : !torch.vtensor<[10,4096,4096],f16> -> !torch.vtensor<[10,4096,4096],f16>
    %1154 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %1155 = torch.aten.sum.dim_IntList %1153, %1154, %true, %none : !torch.vtensor<[10,4096,4096],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,1],f16>
    %1156 = torch.aten.div.Tensor %1153, %1155 : !torch.vtensor<[10,4096,4096],f16>, !torch.vtensor<[10,4096,1],f16> -> !torch.vtensor<[10,4096,4096],f16>
    %1157 = torch.aten.bmm %1156, %1145 : !torch.vtensor<[10,4096,4096],f16>, !torch.vtensor<[10,4096,64],f16> -> !torch.vtensor<[10,4096,64],f16>
    %1158 = torch.aten.view %1157, %881 : !torch.vtensor<[10,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %1159 = torch.aten.permute %1158, %847 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %1160 = torch.aten.clone %1159, %int0 : !torch.vtensor<[2,4096,5,64],f16>, !torch.int -> !torch.vtensor<[2,4096,5,64],f16>
    %1161 = torch.aten.view %1160, %813 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1162 = torch.aten.transpose.int %624, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %1163 = torch.aten.view %1161, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %1164 = torch.aten.mm %1163, %1162 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %1165 = torch.aten.mul.Scalar %623, %int1 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[320],f16>
    %1166 = torch.aten.add.Tensor %1165, %1164, %int1 : !torch.vtensor<[320],f16>, !torch.vtensor<[8192,320],f16>, !torch.int -> !torch.vtensor<[8192,320],f16>
    %1167 = torch.aten.view %1166, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1168 = torch.aten.add.Tensor %1167, %1105, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %1169 = torch.aten.to.dtype %1168, %int7, %false, %false, %none : !torch.vtensor<[2,4096,320],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,320],f64>
    %1170 = torch.aten.sum.dim_IntList %1169, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %1171 = torch.aten.div.Scalar %1170, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %1172 = torch.aten.sub.Tensor %1169, %1171, %float1.000000e00 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,1],f64>, !torch.float -> !torch.vtensor<[2,4096,320],f64>
    %1173 = torch.aten.mul.Tensor %1172, %1172 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,320],f64> -> !torch.vtensor<[2,4096,320],f64>
    %1174 = torch.aten.sum.dim_IntList %1173, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %1175 = torch.aten.div.Scalar %1174, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %1176 = torch.aten.to.dtype %1175, %int5, %false, %false, %none : !torch.vtensor<[2,4096,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %1177 = torch.aten.sum.dim_IntList %1168, %688, %true, %none : !torch.vtensor<[2,4096,320],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %1178 = torch.aten.div.Scalar %1177, %int320 : !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %1179 = torch.aten.add.Scalar %1176, %float1.000000e-05, %int1 : !torch.vtensor<[2,4096,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %1180 = torch.aten.rsqrt %1179 : !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,1],f16>
    %1181 = torch.aten.sub.Tensor %1168, %1178, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %1182 = torch.aten.mul.Tensor %1181, %1180 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,320],f16>
    %1183 = torch.aten.mul.Tensor %1182, %622 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16> -> !torch.vtensor<[2,4096,320],f16>
    %1184 = torch.aten.add.Tensor %1183, %621, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %1185 = torch.aten.transpose.int %620, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %1186 = torch.aten.view %1184, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %1187 = torch.aten.mm %1186, %1185 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %1188 = torch.aten.view %1187, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1189 = torch.aten.view %1188, %845 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %1190 = torch.aten.permute %1189, %847 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %1191 = torch.aten.clone %1190, %int0 : !torch.vtensor<[2,5,4096,64],f16>, !torch.int -> !torch.vtensor<[2,5,4096,64],f16>
    %1192 = torch.aten.view %1191, %850 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[10,4096,64],f16>
    %1193 = torch.aten.transpose.int %619, %int0, %int1 : !torch.vtensor<[320,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,320],f16>
    %1194 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %1195 = torch.aten.mm %1194, %1193 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,320],f16> -> !torch.vtensor<[128,320],f16>
    %1196 = torch.aten.view %1195, %921 : !torch.vtensor<[128,320],f16>, !torch.list<int> -> !torch.vtensor<[2,64,320],f16>
    %1197 = torch.aten.transpose.int %618, %int0, %int1 : !torch.vtensor<[320,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,320],f16>
    %1198 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %1199 = torch.aten.mm %1198, %1197 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,320],f16> -> !torch.vtensor<[128,320],f16>
    %1200 = torch.aten.view %1199, %921 : !torch.vtensor<[128,320],f16>, !torch.list<int> -> !torch.vtensor<[2,64,320],f16>
    %1201 = torch.aten.view %1196, %927 : !torch.vtensor<[2,64,320],f16>, !torch.list<int> -> !torch.vtensor<[2,64,5,64],f16>
    %1202 = torch.aten.permute %1201, %847 : !torch.vtensor<[2,64,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,64,64],f16>
    %1203 = torch.aten.clone %1202, %int0 : !torch.vtensor<[2,5,64,64],f16>, !torch.int -> !torch.vtensor<[2,5,64,64],f16>
    %1204 = torch.aten.view %1203, %931 : !torch.vtensor<[2,5,64,64],f16>, !torch.list<int> -> !torch.vtensor<[10,64,64],f16>
    %1205 = torch.aten.view %1200, %927 : !torch.vtensor<[2,64,320],f16>, !torch.list<int> -> !torch.vtensor<[2,64,5,64],f16>
    %1206 = torch.aten.permute %1205, %847 : !torch.vtensor<[2,64,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,64,64],f16>
    %1207 = torch.aten.clone %1206, %int0 : !torch.vtensor<[2,5,64,64],f16>, !torch.int -> !torch.vtensor<[2,5,64,64],f16>
    %1208 = torch.aten.view %1207, %931 : !torch.vtensor<[2,5,64,64],f16>, !torch.list<int> -> !torch.vtensor<[10,64,64],f16>
    %1209 = torch.aten.empty.memory_format %850, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,64],f32>
    %1210 = torch.aten.transpose.int %1204, %int-1, %int-2 : !torch.vtensor<[10,64,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[10,64,64],f16>
    %1211 = torch.aten.bmm %1192, %1210 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,64,64],f16> -> !torch.vtensor<[10,4096,64],f16>
    %1212 = torch.aten.mul.Scalar %1211, %float1.250000e-01 : !torch.vtensor<[10,4096,64],f16>, !torch.float -> !torch.vtensor<[10,4096,64],f16>
    %1213 = torch.aten.to.dtype %1209, %int5, %false, %false, %none : !torch.vtensor<[10,4096,64],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,64],f16>
    %1214 = torch.aten.add.Tensor %1212, %1209, %int0 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,4096,64],f32>, !torch.int -> !torch.vtensor<[10,4096,64],f16>
    %values_4, %indices_5 = torch.aten.max.dim %1214, %int-1, %true : !torch.vtensor<[10,4096,64],f16>, !torch.int, !torch.bool -> !torch.vtensor<[10,4096,1],f16>, !torch.vtensor<[10,4096,1],si64>
    %1215 = torch.aten.sub.Tensor %1214, %values_4, %float1.000000e00 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,4096,1],f16>, !torch.float -> !torch.vtensor<[10,4096,64],f16>
    %1216 = torch.aten.exp %1215 : !torch.vtensor<[10,4096,64],f16> -> !torch.vtensor<[10,4096,64],f16>
    %1217 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %1218 = torch.aten.sum.dim_IntList %1216, %1217, %true, %none : !torch.vtensor<[10,4096,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[10,4096,1],f16>
    %1219 = torch.aten.div.Tensor %1216, %1218 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,4096,1],f16> -> !torch.vtensor<[10,4096,64],f16>
    %1220 = torch.aten.bmm %1219, %1208 : !torch.vtensor<[10,4096,64],f16>, !torch.vtensor<[10,64,64],f16> -> !torch.vtensor<[10,4096,64],f16>
    %1221 = torch.aten.view %1220, %881 : !torch.vtensor<[10,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[2,5,4096,64],f16>
    %1222 = torch.aten.permute %1221, %847 : !torch.vtensor<[2,5,4096,64],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,5,64],f16>
    %1223 = torch.aten.clone %1222, %int0 : !torch.vtensor<[2,4096,5,64],f16>, !torch.int -> !torch.vtensor<[2,4096,5,64],f16>
    %1224 = torch.aten.view %1223, %813 : !torch.vtensor<[2,4096,5,64],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1225 = torch.aten.transpose.int %617, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %1226 = torch.aten.view %1224, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %1227 = torch.aten.mm %1226, %1225 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %1228 = torch.aten.mul.Scalar %616, %int1 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[320],f16>
    %1229 = torch.aten.add.Tensor %1228, %1227, %int1 : !torch.vtensor<[320],f16>, !torch.vtensor<[8192,320],f16>, !torch.int -> !torch.vtensor<[8192,320],f16>
    %1230 = torch.aten.view %1229, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1231 = torch.aten.add.Tensor %1230, %1168, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %1232 = torch.aten.to.dtype %1231, %int7, %false, %false, %none : !torch.vtensor<[2,4096,320],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,320],f64>
    %1233 = torch.aten.sum.dim_IntList %1232, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %1234 = torch.aten.div.Scalar %1233, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %1235 = torch.aten.sub.Tensor %1232, %1234, %float1.000000e00 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,1],f64>, !torch.float -> !torch.vtensor<[2,4096,320],f64>
    %1236 = torch.aten.mul.Tensor %1235, %1235 : !torch.vtensor<[2,4096,320],f64>, !torch.vtensor<[2,4096,320],f64> -> !torch.vtensor<[2,4096,320],f64>
    %1237 = torch.aten.sum.dim_IntList %1236, %688, %true, %none : !torch.vtensor<[2,4096,320],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f64>
    %1238 = torch.aten.div.Scalar %1237, %int320 : !torch.vtensor<[2,4096,1],f64>, !torch.int -> !torch.vtensor<[2,4096,1],f64>
    %1239 = torch.aten.to.dtype %1238, %int5, %false, %false, %none : !torch.vtensor<[2,4096,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %1240 = torch.aten.sum.dim_IntList %1231, %688, %true, %none : !torch.vtensor<[2,4096,320],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,4096,1],f16>
    %1241 = torch.aten.div.Scalar %1240, %int320 : !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %1242 = torch.aten.add.Scalar %1239, %float1.000000e-05, %int1 : !torch.vtensor<[2,4096,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,4096,1],f16>
    %1243 = torch.aten.rsqrt %1242 : !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,1],f16>
    %1244 = torch.aten.sub.Tensor %1231, %1241, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %1245 = torch.aten.mul.Tensor %1244, %1243 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,1],f16> -> !torch.vtensor<[2,4096,320],f16>
    %1246 = torch.aten.mul.Tensor %1245, %615 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16> -> !torch.vtensor<[2,4096,320],f16>
    %1247 = torch.aten.add.Tensor %1246, %614, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %1248 = torch.aten.transpose.int %613, %int0, %int1 : !torch.vtensor<[2560,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,2560],f16>
    %1249 = torch.aten.view %1247, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %1250 = torch.aten.mm %1249, %1248 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,2560],f16> -> !torch.vtensor<[8192,2560],f16>
    %1251 = torch.aten.mul.Scalar %612, %int1 : !torch.vtensor<[2560],f16>, !torch.int -> !torch.vtensor<[2560],f16>
    %1252 = torch.aten.add.Tensor %1251, %1250, %int1 : !torch.vtensor<[2560],f16>, !torch.vtensor<[8192,2560],f16>, !torch.int -> !torch.vtensor<[8192,2560],f16>
    %1253 = torch.aten.view %1252, %981 : !torch.vtensor<[8192,2560],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,2560],f16>
    %1254 = torch.aten.slice.Tensor %1253, %int-1, %int0, %int1280, %int1 : !torch.vtensor<[2,4096,2560],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,4096,1280],f16>
    %1255 = torch.aten.slice.Tensor %1253, %int-1, %int1280, %int2560, %int1 : !torch.vtensor<[2,4096,2560],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,4096,1280],f16>
    %1256 = torch.aten.gelu %1255, %str : !torch.vtensor<[2,4096,1280],f16>, !torch.str -> !torch.vtensor<[2,4096,1280],f16>
    %1257 = torch.aten.mul.Tensor %1254, %1256 : !torch.vtensor<[2,4096,1280],f16>, !torch.vtensor<[2,4096,1280],f16> -> !torch.vtensor<[2,4096,1280],f16>
    %1258 = torch.aten.transpose.int %611, %int0, %int1 : !torch.vtensor<[320,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,320],f16>
    %1259 = torch.aten.view %1257, %988 : !torch.vtensor<[2,4096,1280],f16>, !torch.list<int> -> !torch.vtensor<[8192,1280],f16>
    %1260 = torch.aten.mm %1259, %1258 : !torch.vtensor<[8192,1280],f16>, !torch.vtensor<[1280,320],f16> -> !torch.vtensor<[8192,320],f16>
    %1261 = torch.aten.mul.Scalar %610, %int1 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[320],f16>
    %1262 = torch.aten.add.Tensor %1261, %1260, %int1 : !torch.vtensor<[320],f16>, !torch.vtensor<[8192,320],f16>, !torch.int -> !torch.vtensor<[8192,320],f16>
    %1263 = torch.aten.view %1262, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1264 = torch.aten.add.Tensor %1263, %1231, %int1 : !torch.vtensor<[2,4096,320],f16>, !torch.vtensor<[2,4096,320],f16>, !torch.int -> !torch.vtensor<[2,4096,320],f16>
    %1265 = torch.aten.transpose.int %609, %int0, %int1 : !torch.vtensor<[320,320],f16>, !torch.int, !torch.int -> !torch.vtensor<[320,320],f16>
    %1266 = torch.aten.view %1264, %841 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[8192,320],f16>
    %1267 = torch.aten.mm %1266, %1265 : !torch.vtensor<[8192,320],f16>, !torch.vtensor<[320,320],f16> -> !torch.vtensor<[8192,320],f16>
    %1268 = torch.aten.mul.Scalar %608, %int1 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[320],f16>
    %1269 = torch.aten.add.Tensor %1268, %1267, %int1 : !torch.vtensor<[320],f16>, !torch.vtensor<[8192,320],f16>, !torch.int -> !torch.vtensor<[8192,320],f16>
    %1270 = torch.aten.view %1269, %813 : !torch.vtensor<[8192,320],f16>, !torch.list<int> -> !torch.vtensor<[2,4096,320],f16>
    %1271 = torch.aten.view %1270, %1001 : !torch.vtensor<[2,4096,320],f16>, !torch.list<int> -> !torch.vtensor<[2,64,64,320],f16>
    %1272 = torch.aten.permute %1271, %1003 : !torch.vtensor<[2,64,64,320],f16>, !torch.list<int> -> !torch.vtensor<[2,320,64,64],f16>
    %1273 = torch.aten.clone %1272, %int0 : !torch.vtensor<[2,320,64,64],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1274 = torch.aten.add.Tensor %1273, %1071, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[2,320,64,64],f16>, !torch.int -> !torch.vtensor<[2,320,64,64],f16>
    %1275 = torch.prim.ListConstruct %int2, %int2 : (!torch.int, !torch.int) -> !torch.list<int>
    %1276 = torch.aten.convolution %1274, %607, %606, %1275, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,320,64,64],f16>, !torch.vtensor<[320,320,3,3],f16>, !torch.vtensor<[320],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,320,32,32],f16>
    %1277 = torch.prim.ListConstruct %int2, %int32, %int10, %int1024 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1278 = torch.aten.view %1276, %1277 : !torch.vtensor<[2,320,32,32],f16>, !torch.list<int> -> !torch.vtensor<[2,32,10,1024],f16>
    %1279 = torch.aten.to.dtype %1278, %int7, %false, %false, %none : !torch.vtensor<[2,32,10,1024],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,10,1024],f64>
    %1280 = torch.aten.sum.dim_IntList %1279, %721, %true, %none : !torch.vtensor<[2,32,10,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1281 = torch.aten.div.Scalar %1280, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1282 = torch.aten.sub.Tensor %1279, %1281, %float1.000000e00 : !torch.vtensor<[2,32,10,1024],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,10,1024],f64>
    %1283 = torch.aten.mul.Tensor %1282, %1282 : !torch.vtensor<[2,32,10,1024],f64>, !torch.vtensor<[2,32,10,1024],f64> -> !torch.vtensor<[2,32,10,1024],f64>
    %1284 = torch.aten.sum.dim_IntList %1283, %721, %true, %none : !torch.vtensor<[2,32,10,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1285 = torch.aten.div.Scalar %1284, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1286 = torch.aten.to.dtype %1285, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1287 = torch.aten.sum.dim_IntList %1278, %721, %true, %none : !torch.vtensor<[2,32,10,1024],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1288 = torch.aten.div.Scalar %1287, %int10240 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1289 = torch.aten.add.Scalar %1286, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1290 = torch.aten.rsqrt %1289 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1291 = torch.aten.sub.Tensor %1278, %1288, %int1 : !torch.vtensor<[2,32,10,1024],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,10,1024],f16>
    %1292 = torch.aten.mul.Tensor %1291, %1290 : !torch.vtensor<[2,32,10,1024],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,10,1024],f16>
    %1293 = torch.prim.ListConstruct %int2, %int320, %int32, %int32 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1294 = torch.aten.view %1292, %1293 : !torch.vtensor<[2,32,10,1024],f16>, !torch.list<int> -> !torch.vtensor<[2,320,32,32],f16>
    %1295 = torch.aten.unsqueeze %605, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %1296 = torch.aten.unsqueeze %1295, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %1297 = torch.aten.unsqueeze %1296, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %1298 = torch.aten.unsqueeze %604, %int0 : !torch.vtensor<[320],f16>, !torch.int -> !torch.vtensor<[1,320],f16>
    %1299 = torch.aten.unsqueeze %1298, %int2 : !torch.vtensor<[1,320],f16>, !torch.int -> !torch.vtensor<[1,320,1],f16>
    %1300 = torch.aten.unsqueeze %1299, %int3 : !torch.vtensor<[1,320,1],f16>, !torch.int -> !torch.vtensor<[1,320,1,1],f16>
    %1301 = torch.aten.mul.Tensor %1294, %1300 : !torch.vtensor<[2,320,32,32],f16>, !torch.vtensor<[1,320,1,1],f16> -> !torch.vtensor<[2,320,32,32],f16>
    %1302 = torch.aten.add.Tensor %1301, %1297, %int1 : !torch.vtensor<[2,320,32,32],f16>, !torch.vtensor<[1,320,1,1],f16>, !torch.int -> !torch.vtensor<[2,320,32,32],f16>
    %1303 = torch.aten.sigmoid %1302 : !torch.vtensor<[2,320,32,32],f16> -> !torch.vtensor<[2,320,32,32],f16>
    %1304 = torch.aten.mul.Tensor %1303, %1302 : !torch.vtensor<[2,320,32,32],f16>, !torch.vtensor<[2,320,32,32],f16> -> !torch.vtensor<[2,320,32,32],f16>
    %1305 = torch.aten.convolution %1304, %603, %602, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,320,32,32],f16>, !torch.vtensor<[640,320,3,3],f16>, !torch.vtensor<[640],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1306 = torch.aten.sigmoid %715 : !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %1307 = torch.aten.mul.Tensor %1306, %715 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %1308 = torch.aten.transpose.int %601, %int0, %int1 : !torch.vtensor<[640,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,640],f16>
    %1309 = torch.aten.mm %1307, %1308 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[1280,640],f16> -> !torch.vtensor<[2,640],f16>
    %1310 = torch.aten.mul.Scalar %600, %int1 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[640],f16>
    %1311 = torch.aten.add.Tensor %1310, %1309, %int1 : !torch.vtensor<[640],f16>, !torch.vtensor<[2,640],f16>, !torch.int -> !torch.vtensor<[2,640],f16>
    %1312 = torch.aten.unsqueeze %1311, %int2 : !torch.vtensor<[2,640],f16>, !torch.int -> !torch.vtensor<[2,640,1],f16>
    %1313 = torch.aten.unsqueeze %1312, %int3 : !torch.vtensor<[2,640,1],f16>, !torch.int -> !torch.vtensor<[2,640,1,1],f16>
    %1314 = torch.aten.add.Tensor %1305, %1313, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[2,640,1,1],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1315 = torch.prim.ListConstruct %int2, %int32, %int20, %int1024 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1316 = torch.aten.view %1314, %1315 : !torch.vtensor<[2,640,32,32],f16>, !torch.list<int> -> !torch.vtensor<[2,32,20,1024],f16>
    %1317 = torch.aten.to.dtype %1316, %int7, %false, %false, %none : !torch.vtensor<[2,32,20,1024],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,20,1024],f64>
    %1318 = torch.aten.sum.dim_IntList %1317, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1319 = torch.aten.div.Scalar %1318, %int20480 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1320 = torch.aten.sub.Tensor %1317, %1319, %float1.000000e00 : !torch.vtensor<[2,32,20,1024],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,20,1024],f64>
    %1321 = torch.aten.mul.Tensor %1320, %1320 : !torch.vtensor<[2,32,20,1024],f64>, !torch.vtensor<[2,32,20,1024],f64> -> !torch.vtensor<[2,32,20,1024],f64>
    %1322 = torch.aten.sum.dim_IntList %1321, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1323 = torch.aten.div.Scalar %1322, %int20480 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1324 = torch.aten.to.dtype %1323, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1325 = torch.aten.sum.dim_IntList %1316, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1326 = torch.aten.div.Scalar %1325, %int20480 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1327 = torch.aten.add.Scalar %1324, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1328 = torch.aten.rsqrt %1327 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1329 = torch.aten.sub.Tensor %1316, %1326, %int1 : !torch.vtensor<[2,32,20,1024],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,20,1024],f16>
    %1330 = torch.aten.mul.Tensor %1329, %1328 : !torch.vtensor<[2,32,20,1024],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,20,1024],f16>
    %1331 = torch.prim.ListConstruct %int2, %int640, %int32, %int32 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1332 = torch.aten.view %1330, %1331 : !torch.vtensor<[2,32,20,1024],f16>, !torch.list<int> -> !torch.vtensor<[2,640,32,32],f16>
    %1333 = torch.aten.unsqueeze %599, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1334 = torch.aten.unsqueeze %1333, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1335 = torch.aten.unsqueeze %1334, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1336 = torch.aten.unsqueeze %598, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1337 = torch.aten.unsqueeze %1336, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1338 = torch.aten.unsqueeze %1337, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1339 = torch.aten.mul.Tensor %1332, %1338 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[1,640,1,1],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1340 = torch.aten.add.Tensor %1339, %1335, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[1,640,1,1],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1341 = torch.aten.sigmoid %1340 : !torch.vtensor<[2,640,32,32],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1342 = torch.aten.mul.Tensor %1341, %1340 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[2,640,32,32],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1343 = torch.aten.convolution %1342, %597, %596, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[640,640,3,3],f16>, !torch.vtensor<[640],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1344 = torch.aten.convolution %1276, %595, %594, %716, %717, %716, %false, %717, %int1 : !torch.vtensor<[2,320,32,32],f16>, !torch.vtensor<[640,320,1,1],f16>, !torch.vtensor<[640],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1345 = torch.aten.add.Tensor %1344, %1343, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[2,640,32,32],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1346 = torch.aten.div.Scalar %1345, %float1.000000e00 : !torch.vtensor<[2,640,32,32],f16>, !torch.float -> !torch.vtensor<[2,640,32,32],f16>
    %1347 = torch.aten.view %1346, %1315 : !torch.vtensor<[2,640,32,32],f16>, !torch.list<int> -> !torch.vtensor<[2,32,20,1024],f16>
    %1348 = torch.aten.to.dtype %1347, %int7, %false, %false, %none : !torch.vtensor<[2,32,20,1024],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,20,1024],f64>
    %1349 = torch.aten.sum.dim_IntList %1348, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1350 = torch.aten.div.Scalar %1349, %int20480 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1351 = torch.aten.sub.Tensor %1348, %1350, %float1.000000e00 : !torch.vtensor<[2,32,20,1024],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,20,1024],f64>
    %1352 = torch.aten.mul.Tensor %1351, %1351 : !torch.vtensor<[2,32,20,1024],f64>, !torch.vtensor<[2,32,20,1024],f64> -> !torch.vtensor<[2,32,20,1024],f64>
    %1353 = torch.aten.sum.dim_IntList %1352, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1354 = torch.aten.div.Scalar %1353, %int20480 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1355 = torch.aten.to.dtype %1354, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1356 = torch.aten.sum.dim_IntList %1347, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1357 = torch.aten.div.Scalar %1356, %int20480 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1358 = torch.aten.add.Scalar %1355, %float9.999990e-07, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1359 = torch.aten.rsqrt %1358 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1360 = torch.aten.sub.Tensor %1347, %1357, %int1 : !torch.vtensor<[2,32,20,1024],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,20,1024],f16>
    %1361 = torch.aten.mul.Tensor %1360, %1359 : !torch.vtensor<[2,32,20,1024],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,20,1024],f16>
    %1362 = torch.aten.view %1361, %1331 : !torch.vtensor<[2,32,20,1024],f16>, !torch.list<int> -> !torch.vtensor<[2,640,32,32],f16>
    %1363 = torch.aten.unsqueeze %593, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1364 = torch.aten.unsqueeze %1363, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1365 = torch.aten.unsqueeze %1364, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1366 = torch.aten.unsqueeze %592, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1367 = torch.aten.unsqueeze %1366, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1368 = torch.aten.unsqueeze %1367, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1369 = torch.aten.mul.Tensor %1362, %1368 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[1,640,1,1],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1370 = torch.aten.add.Tensor %1369, %1365, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[1,640,1,1],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1371 = torch.aten.permute %1370, %811 : !torch.vtensor<[2,640,32,32],f16>, !torch.list<int> -> !torch.vtensor<[2,32,32,640],f16>
    %1372 = torch.prim.ListConstruct %int2, %int1024, %int640 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1373 = torch.aten.view %1371, %1372 : !torch.vtensor<[2,32,32,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1374 = torch.aten.transpose.int %591, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1375 = torch.aten.broadcast_to %1373, %1372 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1376 = torch.aten.view %1375, %1372 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1377 = torch.prim.ListConstruct %int2, %int640, %int640 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1378 = torch.aten.broadcast_to %1374, %1377 : !torch.vtensor<[640,640],f16>, !torch.list<int> -> !torch.vtensor<[2,640,640],f16>
    %1379 = torch.aten.view %1378, %1377 : !torch.vtensor<[2,640,640],f16>, !torch.list<int> -> !torch.vtensor<[2,640,640],f16>
    %1380 = torch.aten.bmm %1376, %1379 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,640,640],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1381 = torch.aten.view %1380, %1372 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1382 = torch.aten.add.Tensor %1381, %590, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1383 = torch.aten.to.dtype %1382, %int7, %false, %false, %none : !torch.vtensor<[2,1024,640],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,640],f64>
    %1384 = torch.aten.sum.dim_IntList %1383, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1385 = torch.aten.div.Scalar %1384, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1386 = torch.aten.sub.Tensor %1383, %1385, %float1.000000e00 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,1],f64>, !torch.float -> !torch.vtensor<[2,1024,640],f64>
    %1387 = torch.aten.mul.Tensor %1386, %1386 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,640],f64> -> !torch.vtensor<[2,1024,640],f64>
    %1388 = torch.aten.sum.dim_IntList %1387, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1389 = torch.aten.div.Scalar %1388, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1390 = torch.aten.to.dtype %1389, %int5, %false, %false, %none : !torch.vtensor<[2,1024,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1391 = torch.aten.sum.dim_IntList %1382, %688, %true, %none : !torch.vtensor<[2,1024,640],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1392 = torch.aten.div.Scalar %1391, %int640 : !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1393 = torch.aten.add.Scalar %1390, %float1.000000e-05, %int1 : !torch.vtensor<[2,1024,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1394 = torch.aten.rsqrt %1393 : !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,1],f16>
    %1395 = torch.aten.sub.Tensor %1382, %1392, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1396 = torch.aten.mul.Tensor %1395, %1394 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1397 = torch.aten.mul.Tensor %1396, %589 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1398 = torch.aten.add.Tensor %1397, %588, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1399 = torch.aten.transpose.int %587, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1400 = torch.prim.ListConstruct %int2048, %int640 : (!torch.int, !torch.int) -> !torch.list<int>
    %1401 = torch.aten.view %1398, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1402 = torch.aten.mm %1401, %1399 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1403 = torch.aten.view %1402, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1404 = torch.prim.ListConstruct %int2, %int1024, %int10, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1405 = torch.aten.view %1403, %1404 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1406 = torch.aten.permute %1405, %847 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1407 = torch.aten.clone %1406, %int0 : !torch.vtensor<[2,10,1024,64],f16>, !torch.int -> !torch.vtensor<[2,10,1024,64],f16>
    %1408 = torch.prim.ListConstruct %int20, %int1024, %int64 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1409 = torch.aten.view %1407, %1408 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[20,1024,64],f16>
    %1410 = torch.aten.transpose.int %586, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1411 = torch.aten.view %1398, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1412 = torch.aten.mm %1411, %1410 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1413 = torch.aten.view %1412, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1414 = torch.aten.transpose.int %585, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1415 = torch.aten.view %1398, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1416 = torch.aten.mm %1415, %1414 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1417 = torch.aten.view %1416, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1418 = torch.aten.view %1413, %1404 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1419 = torch.aten.permute %1418, %847 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1420 = torch.aten.clone %1419, %int0 : !torch.vtensor<[2,10,1024,64],f16>, !torch.int -> !torch.vtensor<[2,10,1024,64],f16>
    %1421 = torch.aten.view %1420, %1408 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[20,1024,64],f16>
    %1422 = torch.aten.view %1417, %1404 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1423 = torch.aten.permute %1422, %847 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1424 = torch.aten.clone %1423, %int0 : !torch.vtensor<[2,10,1024,64],f16>, !torch.int -> !torch.vtensor<[2,10,1024,64],f16>
    %1425 = torch.aten.view %1424, %1408 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[20,1024,64],f16>
    %1426 = torch.prim.ListConstruct %int20, %int1024, %int1024 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1427 = torch.aten.empty.memory_format %1426, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,1024],f32>
    %1428 = torch.aten.transpose.int %1421, %int-1, %int-2 : !torch.vtensor<[20,1024,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[20,64,1024],f16>
    %1429 = torch.aten.bmm %1409, %1428 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,64,1024],f16> -> !torch.vtensor<[20,1024,1024],f16>
    %1430 = torch.aten.mul.Scalar %1429, %float1.250000e-01 : !torch.vtensor<[20,1024,1024],f16>, !torch.float -> !torch.vtensor<[20,1024,1024],f16>
    %1431 = torch.aten.to.dtype %1427, %int5, %false, %false, %none : !torch.vtensor<[20,1024,1024],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,1024],f16>
    %1432 = torch.aten.add.Tensor %1430, %1427, %int0 : !torch.vtensor<[20,1024,1024],f16>, !torch.vtensor<[20,1024,1024],f32>, !torch.int -> !torch.vtensor<[20,1024,1024],f16>
    %values_6, %indices_7 = torch.aten.max.dim %1432, %int-1, %true : !torch.vtensor<[20,1024,1024],f16>, !torch.int, !torch.bool -> !torch.vtensor<[20,1024,1],f16>, !torch.vtensor<[20,1024,1],si64>
    %1433 = torch.aten.sub.Tensor %1432, %values_6, %float1.000000e00 : !torch.vtensor<[20,1024,1024],f16>, !torch.vtensor<[20,1024,1],f16>, !torch.float -> !torch.vtensor<[20,1024,1024],f16>
    %1434 = torch.aten.exp %1433 : !torch.vtensor<[20,1024,1024],f16> -> !torch.vtensor<[20,1024,1024],f16>
    %1435 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %1436 = torch.aten.sum.dim_IntList %1434, %1435, %true, %none : !torch.vtensor<[20,1024,1024],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,1],f16>
    %1437 = torch.aten.div.Tensor %1434, %1436 : !torch.vtensor<[20,1024,1024],f16>, !torch.vtensor<[20,1024,1],f16> -> !torch.vtensor<[20,1024,1024],f16>
    %1438 = torch.aten.bmm %1437, %1425 : !torch.vtensor<[20,1024,1024],f16>, !torch.vtensor<[20,1024,64],f16> -> !torch.vtensor<[20,1024,64],f16>
    %1439 = torch.prim.ListConstruct %int2, %int10, %int1024, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1440 = torch.aten.view %1438, %1439 : !torch.vtensor<[20,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1441 = torch.aten.permute %1440, %847 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1442 = torch.aten.clone %1441, %int0 : !torch.vtensor<[2,1024,10,64],f16>, !torch.int -> !torch.vtensor<[2,1024,10,64],f16>
    %1443 = torch.aten.view %1442, %1372 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1444 = torch.aten.transpose.int %584, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1445 = torch.aten.view %1443, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1446 = torch.aten.mm %1445, %1444 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1447 = torch.aten.mul.Scalar %583, %int1 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[640],f16>
    %1448 = torch.aten.add.Tensor %1447, %1446, %int1 : !torch.vtensor<[640],f16>, !torch.vtensor<[2048,640],f16>, !torch.int -> !torch.vtensor<[2048,640],f16>
    %1449 = torch.aten.view %1448, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1450 = torch.aten.add.Tensor %1449, %1382, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1451 = torch.aten.to.dtype %1450, %int7, %false, %false, %none : !torch.vtensor<[2,1024,640],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,640],f64>
    %1452 = torch.aten.sum.dim_IntList %1451, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1453 = torch.aten.div.Scalar %1452, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1454 = torch.aten.sub.Tensor %1451, %1453, %float1.000000e00 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,1],f64>, !torch.float -> !torch.vtensor<[2,1024,640],f64>
    %1455 = torch.aten.mul.Tensor %1454, %1454 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,640],f64> -> !torch.vtensor<[2,1024,640],f64>
    %1456 = torch.aten.sum.dim_IntList %1455, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1457 = torch.aten.div.Scalar %1456, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1458 = torch.aten.to.dtype %1457, %int5, %false, %false, %none : !torch.vtensor<[2,1024,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1459 = torch.aten.sum.dim_IntList %1450, %688, %true, %none : !torch.vtensor<[2,1024,640],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1460 = torch.aten.div.Scalar %1459, %int640 : !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1461 = torch.aten.add.Scalar %1458, %float1.000000e-05, %int1 : !torch.vtensor<[2,1024,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1462 = torch.aten.rsqrt %1461 : !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,1],f16>
    %1463 = torch.aten.sub.Tensor %1450, %1460, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1464 = torch.aten.mul.Tensor %1463, %1462 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1465 = torch.aten.mul.Tensor %1464, %582 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1466 = torch.aten.add.Tensor %1465, %581, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1467 = torch.aten.transpose.int %580, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1468 = torch.aten.view %1466, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1469 = torch.aten.mm %1468, %1467 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1470 = torch.aten.view %1469, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1471 = torch.aten.view %1470, %1404 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1472 = torch.aten.permute %1471, %847 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1473 = torch.aten.clone %1472, %int0 : !torch.vtensor<[2,10,1024,64],f16>, !torch.int -> !torch.vtensor<[2,10,1024,64],f16>
    %1474 = torch.aten.view %1473, %1408 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[20,1024,64],f16>
    %1475 = torch.aten.transpose.int %579, %int0, %int1 : !torch.vtensor<[640,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,640],f16>
    %1476 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %1477 = torch.aten.mm %1476, %1475 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,640],f16> -> !torch.vtensor<[128,640],f16>
    %1478 = torch.prim.ListConstruct %int2, %int64, %int640 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1479 = torch.aten.view %1477, %1478 : !torch.vtensor<[128,640],f16>, !torch.list<int> -> !torch.vtensor<[2,64,640],f16>
    %1480 = torch.aten.transpose.int %578, %int0, %int1 : !torch.vtensor<[640,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,640],f16>
    %1481 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %1482 = torch.aten.mm %1481, %1480 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,640],f16> -> !torch.vtensor<[128,640],f16>
    %1483 = torch.aten.view %1482, %1478 : !torch.vtensor<[128,640],f16>, !torch.list<int> -> !torch.vtensor<[2,64,640],f16>
    %1484 = torch.prim.ListConstruct %int2, %int64, %int10, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1485 = torch.aten.view %1479, %1484 : !torch.vtensor<[2,64,640],f16>, !torch.list<int> -> !torch.vtensor<[2,64,10,64],f16>
    %1486 = torch.aten.permute %1485, %847 : !torch.vtensor<[2,64,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,64,64],f16>
    %1487 = torch.aten.clone %1486, %int0 : !torch.vtensor<[2,10,64,64],f16>, !torch.int -> !torch.vtensor<[2,10,64,64],f16>
    %1488 = torch.prim.ListConstruct %int20, %int64, %int64 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1489 = torch.aten.view %1487, %1488 : !torch.vtensor<[2,10,64,64],f16>, !torch.list<int> -> !torch.vtensor<[20,64,64],f16>
    %1490 = torch.aten.view %1483, %1484 : !torch.vtensor<[2,64,640],f16>, !torch.list<int> -> !torch.vtensor<[2,64,10,64],f16>
    %1491 = torch.aten.permute %1490, %847 : !torch.vtensor<[2,64,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,64,64],f16>
    %1492 = torch.aten.clone %1491, %int0 : !torch.vtensor<[2,10,64,64],f16>, !torch.int -> !torch.vtensor<[2,10,64,64],f16>
    %1493 = torch.aten.view %1492, %1488 : !torch.vtensor<[2,10,64,64],f16>, !torch.list<int> -> !torch.vtensor<[20,64,64],f16>
    %1494 = torch.aten.empty.memory_format %1408, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,64],f32>
    %1495 = torch.aten.transpose.int %1489, %int-1, %int-2 : !torch.vtensor<[20,64,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[20,64,64],f16>
    %1496 = torch.aten.bmm %1474, %1495 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,64,64],f16> -> !torch.vtensor<[20,1024,64],f16>
    %1497 = torch.aten.mul.Scalar %1496, %float1.250000e-01 : !torch.vtensor<[20,1024,64],f16>, !torch.float -> !torch.vtensor<[20,1024,64],f16>
    %1498 = torch.aten.to.dtype %1494, %int5, %false, %false, %none : !torch.vtensor<[20,1024,64],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,64],f16>
    %1499 = torch.aten.add.Tensor %1497, %1494, %int0 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,1024,64],f32>, !torch.int -> !torch.vtensor<[20,1024,64],f16>
    %values_8, %indices_9 = torch.aten.max.dim %1499, %int-1, %true : !torch.vtensor<[20,1024,64],f16>, !torch.int, !torch.bool -> !torch.vtensor<[20,1024,1],f16>, !torch.vtensor<[20,1024,1],si64>
    %1500 = torch.aten.sub.Tensor %1499, %values_8, %float1.000000e00 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,1024,1],f16>, !torch.float -> !torch.vtensor<[20,1024,64],f16>
    %1501 = torch.aten.exp %1500 : !torch.vtensor<[20,1024,64],f16> -> !torch.vtensor<[20,1024,64],f16>
    %1502 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %1503 = torch.aten.sum.dim_IntList %1501, %1502, %true, %none : !torch.vtensor<[20,1024,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,1],f16>
    %1504 = torch.aten.div.Tensor %1501, %1503 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,1024,1],f16> -> !torch.vtensor<[20,1024,64],f16>
    %1505 = torch.aten.bmm %1504, %1493 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,64,64],f16> -> !torch.vtensor<[20,1024,64],f16>
    %1506 = torch.aten.view %1505, %1439 : !torch.vtensor<[20,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1507 = torch.aten.permute %1506, %847 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1508 = torch.aten.clone %1507, %int0 : !torch.vtensor<[2,1024,10,64],f16>, !torch.int -> !torch.vtensor<[2,1024,10,64],f16>
    %1509 = torch.aten.view %1508, %1372 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1510 = torch.aten.transpose.int %577, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1511 = torch.aten.view %1509, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1512 = torch.aten.mm %1511, %1510 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1513 = torch.aten.mul.Scalar %576, %int1 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[640],f16>
    %1514 = torch.aten.add.Tensor %1513, %1512, %int1 : !torch.vtensor<[640],f16>, !torch.vtensor<[2048,640],f16>, !torch.int -> !torch.vtensor<[2048,640],f16>
    %1515 = torch.aten.view %1514, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1516 = torch.aten.add.Tensor %1515, %1450, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1517 = torch.aten.to.dtype %1516, %int7, %false, %false, %none : !torch.vtensor<[2,1024,640],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,640],f64>
    %1518 = torch.aten.sum.dim_IntList %1517, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1519 = torch.aten.div.Scalar %1518, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1520 = torch.aten.sub.Tensor %1517, %1519, %float1.000000e00 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,1],f64>, !torch.float -> !torch.vtensor<[2,1024,640],f64>
    %1521 = torch.aten.mul.Tensor %1520, %1520 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,640],f64> -> !torch.vtensor<[2,1024,640],f64>
    %1522 = torch.aten.sum.dim_IntList %1521, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1523 = torch.aten.div.Scalar %1522, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1524 = torch.aten.to.dtype %1523, %int5, %false, %false, %none : !torch.vtensor<[2,1024,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1525 = torch.aten.sum.dim_IntList %1516, %688, %true, %none : !torch.vtensor<[2,1024,640],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1526 = torch.aten.div.Scalar %1525, %int640 : !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1527 = torch.aten.add.Scalar %1524, %float1.000000e-05, %int1 : !torch.vtensor<[2,1024,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1528 = torch.aten.rsqrt %1527 : !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,1],f16>
    %1529 = torch.aten.sub.Tensor %1516, %1526, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1530 = torch.aten.mul.Tensor %1529, %1528 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1531 = torch.aten.mul.Tensor %1530, %575 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1532 = torch.aten.add.Tensor %1531, %574, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1533 = torch.aten.transpose.int %573, %int0, %int1 : !torch.vtensor<[5120,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,5120],f16>
    %1534 = torch.aten.view %1532, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1535 = torch.aten.mm %1534, %1533 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,5120],f16> -> !torch.vtensor<[2048,5120],f16>
    %1536 = torch.aten.mul.Scalar %572, %int1 : !torch.vtensor<[5120],f16>, !torch.int -> !torch.vtensor<[5120],f16>
    %1537 = torch.aten.add.Tensor %1536, %1535, %int1 : !torch.vtensor<[5120],f16>, !torch.vtensor<[2048,5120],f16>, !torch.int -> !torch.vtensor<[2048,5120],f16>
    %1538 = torch.prim.ListConstruct %int2, %int1024, %int5120 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1539 = torch.aten.view %1537, %1538 : !torch.vtensor<[2048,5120],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,5120],f16>
    %1540 = torch.aten.slice.Tensor %1539, %int-1, %int0, %int2560, %int1 : !torch.vtensor<[2,1024,5120],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,1024,2560],f16>
    %1541 = torch.aten.slice.Tensor %1539, %int-1, %int2560, %int5120, %int1 : !torch.vtensor<[2,1024,5120],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,1024,2560],f16>
    %1542 = torch.aten.gelu %1541, %str : !torch.vtensor<[2,1024,2560],f16>, !torch.str -> !torch.vtensor<[2,1024,2560],f16>
    %1543 = torch.aten.mul.Tensor %1540, %1542 : !torch.vtensor<[2,1024,2560],f16>, !torch.vtensor<[2,1024,2560],f16> -> !torch.vtensor<[2,1024,2560],f16>
    %1544 = torch.aten.transpose.int %571, %int0, %int1 : !torch.vtensor<[640,2560],f16>, !torch.int, !torch.int -> !torch.vtensor<[2560,640],f16>
    %1545 = torch.prim.ListConstruct %int2048, %int2560 : (!torch.int, !torch.int) -> !torch.list<int>
    %1546 = torch.aten.view %1543, %1545 : !torch.vtensor<[2,1024,2560],f16>, !torch.list<int> -> !torch.vtensor<[2048,2560],f16>
    %1547 = torch.aten.mm %1546, %1544 : !torch.vtensor<[2048,2560],f16>, !torch.vtensor<[2560,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1548 = torch.aten.mul.Scalar %570, %int1 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[640],f16>
    %1549 = torch.aten.add.Tensor %1548, %1547, %int1 : !torch.vtensor<[640],f16>, !torch.vtensor<[2048,640],f16>, !torch.int -> !torch.vtensor<[2048,640],f16>
    %1550 = torch.aten.view %1549, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1551 = torch.aten.add.Tensor %1550, %1516, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1552 = torch.aten.transpose.int %569, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1553 = torch.aten.view %1551, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1554 = torch.aten.mm %1553, %1552 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1555 = torch.aten.mul.Scalar %568, %int1 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[640],f16>
    %1556 = torch.aten.add.Tensor %1555, %1554, %int1 : !torch.vtensor<[640],f16>, !torch.vtensor<[2048,640],f16>, !torch.int -> !torch.vtensor<[2048,640],f16>
    %1557 = torch.aten.view %1556, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1558 = torch.prim.ListConstruct %int2, %int32, %int32, %int640 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1559 = torch.aten.view %1557, %1558 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,32,32,640],f16>
    %1560 = torch.aten.permute %1559, %1003 : !torch.vtensor<[2,32,32,640],f16>, !torch.list<int> -> !torch.vtensor<[2,640,32,32],f16>
    %1561 = torch.aten.clone %1560, %int0 : !torch.vtensor<[2,640,32,32],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1562 = torch.aten.add.Tensor %1561, %1346, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[2,640,32,32],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1563 = torch.aten.view %1562, %1315 : !torch.vtensor<[2,640,32,32],f16>, !torch.list<int> -> !torch.vtensor<[2,32,20,1024],f16>
    %1564 = torch.aten.to.dtype %1563, %int7, %false, %false, %none : !torch.vtensor<[2,32,20,1024],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,20,1024],f64>
    %1565 = torch.aten.sum.dim_IntList %1564, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1566 = torch.aten.div.Scalar %1565, %int20480 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1567 = torch.aten.sub.Tensor %1564, %1566, %float1.000000e00 : !torch.vtensor<[2,32,20,1024],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,20,1024],f64>
    %1568 = torch.aten.mul.Tensor %1567, %1567 : !torch.vtensor<[2,32,20,1024],f64>, !torch.vtensor<[2,32,20,1024],f64> -> !torch.vtensor<[2,32,20,1024],f64>
    %1569 = torch.aten.sum.dim_IntList %1568, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1570 = torch.aten.div.Scalar %1569, %int20480 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1571 = torch.aten.to.dtype %1570, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1572 = torch.aten.sum.dim_IntList %1563, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1573 = torch.aten.div.Scalar %1572, %int20480 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1574 = torch.aten.add.Scalar %1571, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1575 = torch.aten.rsqrt %1574 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1576 = torch.aten.sub.Tensor %1563, %1573, %int1 : !torch.vtensor<[2,32,20,1024],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,20,1024],f16>
    %1577 = torch.aten.mul.Tensor %1576, %1575 : !torch.vtensor<[2,32,20,1024],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,20,1024],f16>
    %1578 = torch.aten.view %1577, %1331 : !torch.vtensor<[2,32,20,1024],f16>, !torch.list<int> -> !torch.vtensor<[2,640,32,32],f16>
    %1579 = torch.aten.unsqueeze %567, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1580 = torch.aten.unsqueeze %1579, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1581 = torch.aten.unsqueeze %1580, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1582 = torch.aten.unsqueeze %566, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1583 = torch.aten.unsqueeze %1582, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1584 = torch.aten.unsqueeze %1583, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1585 = torch.aten.mul.Tensor %1578, %1584 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[1,640,1,1],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1586 = torch.aten.add.Tensor %1585, %1581, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[1,640,1,1],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1587 = torch.aten.sigmoid %1586 : !torch.vtensor<[2,640,32,32],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1588 = torch.aten.mul.Tensor %1587, %1586 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[2,640,32,32],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1589 = torch.aten.convolution %1588, %565, %564, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[640,640,3,3],f16>, !torch.vtensor<[640],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1590 = torch.aten.sigmoid %715 : !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %1591 = torch.aten.mul.Tensor %1590, %715 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %1592 = torch.aten.transpose.int %563, %int0, %int1 : !torch.vtensor<[640,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,640],f16>
    %1593 = torch.aten.mm %1591, %1592 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[1280,640],f16> -> !torch.vtensor<[2,640],f16>
    %1594 = torch.aten.mul.Scalar %562, %int1 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[640],f16>
    %1595 = torch.aten.add.Tensor %1594, %1593, %int1 : !torch.vtensor<[640],f16>, !torch.vtensor<[2,640],f16>, !torch.int -> !torch.vtensor<[2,640],f16>
    %1596 = torch.aten.unsqueeze %1595, %int2 : !torch.vtensor<[2,640],f16>, !torch.int -> !torch.vtensor<[2,640,1],f16>
    %1597 = torch.aten.unsqueeze %1596, %int3 : !torch.vtensor<[2,640,1],f16>, !torch.int -> !torch.vtensor<[2,640,1,1],f16>
    %1598 = torch.aten.add.Tensor %1589, %1597, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[2,640,1,1],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1599 = torch.aten.view %1598, %1315 : !torch.vtensor<[2,640,32,32],f16>, !torch.list<int> -> !torch.vtensor<[2,32,20,1024],f16>
    %1600 = torch.aten.to.dtype %1599, %int7, %false, %false, %none : !torch.vtensor<[2,32,20,1024],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,20,1024],f64>
    %1601 = torch.aten.sum.dim_IntList %1600, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1602 = torch.aten.div.Scalar %1601, %int20480 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1603 = torch.aten.sub.Tensor %1600, %1602, %float1.000000e00 : !torch.vtensor<[2,32,20,1024],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,20,1024],f64>
    %1604 = torch.aten.mul.Tensor %1603, %1603 : !torch.vtensor<[2,32,20,1024],f64>, !torch.vtensor<[2,32,20,1024],f64> -> !torch.vtensor<[2,32,20,1024],f64>
    %1605 = torch.aten.sum.dim_IntList %1604, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1606 = torch.aten.div.Scalar %1605, %int20480 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1607 = torch.aten.to.dtype %1606, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1608 = torch.aten.sum.dim_IntList %1599, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1609 = torch.aten.div.Scalar %1608, %int20480 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1610 = torch.aten.add.Scalar %1607, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1611 = torch.aten.rsqrt %1610 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1612 = torch.aten.sub.Tensor %1599, %1609, %int1 : !torch.vtensor<[2,32,20,1024],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,20,1024],f16>
    %1613 = torch.aten.mul.Tensor %1612, %1611 : !torch.vtensor<[2,32,20,1024],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,20,1024],f16>
    %1614 = torch.aten.view %1613, %1331 : !torch.vtensor<[2,32,20,1024],f16>, !torch.list<int> -> !torch.vtensor<[2,640,32,32],f16>
    %1615 = torch.aten.unsqueeze %561, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1616 = torch.aten.unsqueeze %1615, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1617 = torch.aten.unsqueeze %1616, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1618 = torch.aten.unsqueeze %560, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1619 = torch.aten.unsqueeze %1618, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1620 = torch.aten.unsqueeze %1619, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1621 = torch.aten.mul.Tensor %1614, %1620 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[1,640,1,1],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1622 = torch.aten.add.Tensor %1621, %1617, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[1,640,1,1],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1623 = torch.aten.sigmoid %1622 : !torch.vtensor<[2,640,32,32],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1624 = torch.aten.mul.Tensor %1623, %1622 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[2,640,32,32],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1625 = torch.aten.convolution %1624, %559, %558, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[640,640,3,3],f16>, !torch.vtensor<[640],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1626 = torch.aten.add.Tensor %1562, %1625, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[2,640,32,32],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1627 = torch.aten.div.Scalar %1626, %float1.000000e00 : !torch.vtensor<[2,640,32,32],f16>, !torch.float -> !torch.vtensor<[2,640,32,32],f16>
    %1628 = torch.aten.view %1627, %1315 : !torch.vtensor<[2,640,32,32],f16>, !torch.list<int> -> !torch.vtensor<[2,32,20,1024],f16>
    %1629 = torch.aten.to.dtype %1628, %int7, %false, %false, %none : !torch.vtensor<[2,32,20,1024],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,20,1024],f64>
    %1630 = torch.aten.sum.dim_IntList %1629, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1631 = torch.aten.div.Scalar %1630, %int20480 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1632 = torch.aten.sub.Tensor %1629, %1631, %float1.000000e00 : !torch.vtensor<[2,32,20,1024],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,20,1024],f64>
    %1633 = torch.aten.mul.Tensor %1632, %1632 : !torch.vtensor<[2,32,20,1024],f64>, !torch.vtensor<[2,32,20,1024],f64> -> !torch.vtensor<[2,32,20,1024],f64>
    %1634 = torch.aten.sum.dim_IntList %1633, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1635 = torch.aten.div.Scalar %1634, %int20480 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1636 = torch.aten.to.dtype %1635, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1637 = torch.aten.sum.dim_IntList %1628, %721, %true, %none : !torch.vtensor<[2,32,20,1024],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1638 = torch.aten.div.Scalar %1637, %int20480 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1639 = torch.aten.add.Scalar %1636, %float9.999990e-07, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1640 = torch.aten.rsqrt %1639 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1641 = torch.aten.sub.Tensor %1628, %1638, %int1 : !torch.vtensor<[2,32,20,1024],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,20,1024],f16>
    %1642 = torch.aten.mul.Tensor %1641, %1640 : !torch.vtensor<[2,32,20,1024],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,20,1024],f16>
    %1643 = torch.aten.view %1642, %1331 : !torch.vtensor<[2,32,20,1024],f16>, !torch.list<int> -> !torch.vtensor<[2,640,32,32],f16>
    %1644 = torch.aten.unsqueeze %557, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1645 = torch.aten.unsqueeze %1644, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1646 = torch.aten.unsqueeze %1645, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1647 = torch.aten.unsqueeze %556, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1648 = torch.aten.unsqueeze %1647, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1649 = torch.aten.unsqueeze %1648, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1650 = torch.aten.mul.Tensor %1643, %1649 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[1,640,1,1],f16> -> !torch.vtensor<[2,640,32,32],f16>
    %1651 = torch.aten.add.Tensor %1650, %1646, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[1,640,1,1],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1652 = torch.aten.permute %1651, %811 : !torch.vtensor<[2,640,32,32],f16>, !torch.list<int> -> !torch.vtensor<[2,32,32,640],f16>
    %1653 = torch.aten.view %1652, %1372 : !torch.vtensor<[2,32,32,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1654 = torch.aten.transpose.int %555, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1655 = torch.aten.broadcast_to %1653, %1372 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1656 = torch.aten.view %1655, %1372 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1657 = torch.aten.broadcast_to %1654, %1377 : !torch.vtensor<[640,640],f16>, !torch.list<int> -> !torch.vtensor<[2,640,640],f16>
    %1658 = torch.aten.view %1657, %1377 : !torch.vtensor<[2,640,640],f16>, !torch.list<int> -> !torch.vtensor<[2,640,640],f16>
    %1659 = torch.aten.bmm %1656, %1658 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,640,640],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1660 = torch.aten.view %1659, %1372 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1661 = torch.aten.add.Tensor %1660, %554, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1662 = torch.aten.to.dtype %1661, %int7, %false, %false, %none : !torch.vtensor<[2,1024,640],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,640],f64>
    %1663 = torch.aten.sum.dim_IntList %1662, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1664 = torch.aten.div.Scalar %1663, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1665 = torch.aten.sub.Tensor %1662, %1664, %float1.000000e00 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,1],f64>, !torch.float -> !torch.vtensor<[2,1024,640],f64>
    %1666 = torch.aten.mul.Tensor %1665, %1665 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,640],f64> -> !torch.vtensor<[2,1024,640],f64>
    %1667 = torch.aten.sum.dim_IntList %1666, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1668 = torch.aten.div.Scalar %1667, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1669 = torch.aten.to.dtype %1668, %int5, %false, %false, %none : !torch.vtensor<[2,1024,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1670 = torch.aten.sum.dim_IntList %1661, %688, %true, %none : !torch.vtensor<[2,1024,640],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1671 = torch.aten.div.Scalar %1670, %int640 : !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1672 = torch.aten.add.Scalar %1669, %float1.000000e-05, %int1 : !torch.vtensor<[2,1024,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1673 = torch.aten.rsqrt %1672 : !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,1],f16>
    %1674 = torch.aten.sub.Tensor %1661, %1671, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1675 = torch.aten.mul.Tensor %1674, %1673 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1676 = torch.aten.mul.Tensor %1675, %553 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1677 = torch.aten.add.Tensor %1676, %552, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1678 = torch.aten.transpose.int %551, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1679 = torch.aten.view %1677, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1680 = torch.aten.mm %1679, %1678 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1681 = torch.aten.view %1680, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1682 = torch.aten.view %1681, %1404 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1683 = torch.aten.permute %1682, %847 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1684 = torch.aten.clone %1683, %int0 : !torch.vtensor<[2,10,1024,64],f16>, !torch.int -> !torch.vtensor<[2,10,1024,64],f16>
    %1685 = torch.aten.view %1684, %1408 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[20,1024,64],f16>
    %1686 = torch.aten.transpose.int %550, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1687 = torch.aten.view %1677, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1688 = torch.aten.mm %1687, %1686 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1689 = torch.aten.view %1688, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1690 = torch.aten.transpose.int %549, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1691 = torch.aten.view %1677, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1692 = torch.aten.mm %1691, %1690 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1693 = torch.aten.view %1692, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1694 = torch.aten.view %1689, %1404 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1695 = torch.aten.permute %1694, %847 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1696 = torch.aten.clone %1695, %int0 : !torch.vtensor<[2,10,1024,64],f16>, !torch.int -> !torch.vtensor<[2,10,1024,64],f16>
    %1697 = torch.aten.view %1696, %1408 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[20,1024,64],f16>
    %1698 = torch.aten.view %1693, %1404 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1699 = torch.aten.permute %1698, %847 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1700 = torch.aten.clone %1699, %int0 : !torch.vtensor<[2,10,1024,64],f16>, !torch.int -> !torch.vtensor<[2,10,1024,64],f16>
    %1701 = torch.aten.view %1700, %1408 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[20,1024,64],f16>
    %1702 = torch.aten.empty.memory_format %1426, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,1024],f32>
    %1703 = torch.aten.transpose.int %1697, %int-1, %int-2 : !torch.vtensor<[20,1024,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[20,64,1024],f16>
    %1704 = torch.aten.bmm %1685, %1703 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,64,1024],f16> -> !torch.vtensor<[20,1024,1024],f16>
    %1705 = torch.aten.mul.Scalar %1704, %float1.250000e-01 : !torch.vtensor<[20,1024,1024],f16>, !torch.float -> !torch.vtensor<[20,1024,1024],f16>
    %1706 = torch.aten.to.dtype %1702, %int5, %false, %false, %none : !torch.vtensor<[20,1024,1024],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,1024],f16>
    %1707 = torch.aten.add.Tensor %1705, %1702, %int0 : !torch.vtensor<[20,1024,1024],f16>, !torch.vtensor<[20,1024,1024],f32>, !torch.int -> !torch.vtensor<[20,1024,1024],f16>
    %values_10, %indices_11 = torch.aten.max.dim %1707, %int-1, %true : !torch.vtensor<[20,1024,1024],f16>, !torch.int, !torch.bool -> !torch.vtensor<[20,1024,1],f16>, !torch.vtensor<[20,1024,1],si64>
    %1708 = torch.aten.sub.Tensor %1707, %values_10, %float1.000000e00 : !torch.vtensor<[20,1024,1024],f16>, !torch.vtensor<[20,1024,1],f16>, !torch.float -> !torch.vtensor<[20,1024,1024],f16>
    %1709 = torch.aten.exp %1708 : !torch.vtensor<[20,1024,1024],f16> -> !torch.vtensor<[20,1024,1024],f16>
    %1710 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %1711 = torch.aten.sum.dim_IntList %1709, %1710, %true, %none : !torch.vtensor<[20,1024,1024],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,1],f16>
    %1712 = torch.aten.div.Tensor %1709, %1711 : !torch.vtensor<[20,1024,1024],f16>, !torch.vtensor<[20,1024,1],f16> -> !torch.vtensor<[20,1024,1024],f16>
    %1713 = torch.aten.bmm %1712, %1701 : !torch.vtensor<[20,1024,1024],f16>, !torch.vtensor<[20,1024,64],f16> -> !torch.vtensor<[20,1024,64],f16>
    %1714 = torch.aten.view %1713, %1439 : !torch.vtensor<[20,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1715 = torch.aten.permute %1714, %847 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1716 = torch.aten.clone %1715, %int0 : !torch.vtensor<[2,1024,10,64],f16>, !torch.int -> !torch.vtensor<[2,1024,10,64],f16>
    %1717 = torch.aten.view %1716, %1372 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1718 = torch.aten.transpose.int %548, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1719 = torch.aten.view %1717, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1720 = torch.aten.mm %1719, %1718 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1721 = torch.aten.mul.Scalar %547, %int1 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[640],f16>
    %1722 = torch.aten.add.Tensor %1721, %1720, %int1 : !torch.vtensor<[640],f16>, !torch.vtensor<[2048,640],f16>, !torch.int -> !torch.vtensor<[2048,640],f16>
    %1723 = torch.aten.view %1722, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1724 = torch.aten.add.Tensor %1723, %1661, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1725 = torch.aten.to.dtype %1724, %int7, %false, %false, %none : !torch.vtensor<[2,1024,640],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,640],f64>
    %1726 = torch.aten.sum.dim_IntList %1725, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1727 = torch.aten.div.Scalar %1726, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1728 = torch.aten.sub.Tensor %1725, %1727, %float1.000000e00 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,1],f64>, !torch.float -> !torch.vtensor<[2,1024,640],f64>
    %1729 = torch.aten.mul.Tensor %1728, %1728 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,640],f64> -> !torch.vtensor<[2,1024,640],f64>
    %1730 = torch.aten.sum.dim_IntList %1729, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1731 = torch.aten.div.Scalar %1730, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1732 = torch.aten.to.dtype %1731, %int5, %false, %false, %none : !torch.vtensor<[2,1024,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1733 = torch.aten.sum.dim_IntList %1724, %688, %true, %none : !torch.vtensor<[2,1024,640],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1734 = torch.aten.div.Scalar %1733, %int640 : !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1735 = torch.aten.add.Scalar %1732, %float1.000000e-05, %int1 : !torch.vtensor<[2,1024,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1736 = torch.aten.rsqrt %1735 : !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,1],f16>
    %1737 = torch.aten.sub.Tensor %1724, %1734, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1738 = torch.aten.mul.Tensor %1737, %1736 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1739 = torch.aten.mul.Tensor %1738, %546 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1740 = torch.aten.add.Tensor %1739, %545, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1741 = torch.aten.transpose.int %544, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1742 = torch.aten.view %1740, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1743 = torch.aten.mm %1742, %1741 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1744 = torch.aten.view %1743, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1745 = torch.aten.view %1744, %1404 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1746 = torch.aten.permute %1745, %847 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1747 = torch.aten.clone %1746, %int0 : !torch.vtensor<[2,10,1024,64],f16>, !torch.int -> !torch.vtensor<[2,10,1024,64],f16>
    %1748 = torch.aten.view %1747, %1408 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[20,1024,64],f16>
    %1749 = torch.aten.transpose.int %543, %int0, %int1 : !torch.vtensor<[640,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,640],f16>
    %1750 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %1751 = torch.aten.mm %1750, %1749 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,640],f16> -> !torch.vtensor<[128,640],f16>
    %1752 = torch.aten.view %1751, %1478 : !torch.vtensor<[128,640],f16>, !torch.list<int> -> !torch.vtensor<[2,64,640],f16>
    %1753 = torch.aten.transpose.int %542, %int0, %int1 : !torch.vtensor<[640,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,640],f16>
    %1754 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %1755 = torch.aten.mm %1754, %1753 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,640],f16> -> !torch.vtensor<[128,640],f16>
    %1756 = torch.aten.view %1755, %1478 : !torch.vtensor<[128,640],f16>, !torch.list<int> -> !torch.vtensor<[2,64,640],f16>
    %1757 = torch.aten.view %1752, %1484 : !torch.vtensor<[2,64,640],f16>, !torch.list<int> -> !torch.vtensor<[2,64,10,64],f16>
    %1758 = torch.aten.permute %1757, %847 : !torch.vtensor<[2,64,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,64,64],f16>
    %1759 = torch.aten.clone %1758, %int0 : !torch.vtensor<[2,10,64,64],f16>, !torch.int -> !torch.vtensor<[2,10,64,64],f16>
    %1760 = torch.aten.view %1759, %1488 : !torch.vtensor<[2,10,64,64],f16>, !torch.list<int> -> !torch.vtensor<[20,64,64],f16>
    %1761 = torch.aten.view %1756, %1484 : !torch.vtensor<[2,64,640],f16>, !torch.list<int> -> !torch.vtensor<[2,64,10,64],f16>
    %1762 = torch.aten.permute %1761, %847 : !torch.vtensor<[2,64,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,64,64],f16>
    %1763 = torch.aten.clone %1762, %int0 : !torch.vtensor<[2,10,64,64],f16>, !torch.int -> !torch.vtensor<[2,10,64,64],f16>
    %1764 = torch.aten.view %1763, %1488 : !torch.vtensor<[2,10,64,64],f16>, !torch.list<int> -> !torch.vtensor<[20,64,64],f16>
    %1765 = torch.aten.empty.memory_format %1408, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,64],f32>
    %1766 = torch.aten.transpose.int %1760, %int-1, %int-2 : !torch.vtensor<[20,64,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[20,64,64],f16>
    %1767 = torch.aten.bmm %1748, %1766 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,64,64],f16> -> !torch.vtensor<[20,1024,64],f16>
    %1768 = torch.aten.mul.Scalar %1767, %float1.250000e-01 : !torch.vtensor<[20,1024,64],f16>, !torch.float -> !torch.vtensor<[20,1024,64],f16>
    %1769 = torch.aten.to.dtype %1765, %int5, %false, %false, %none : !torch.vtensor<[20,1024,64],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,64],f16>
    %1770 = torch.aten.add.Tensor %1768, %1765, %int0 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,1024,64],f32>, !torch.int -> !torch.vtensor<[20,1024,64],f16>
    %values_12, %indices_13 = torch.aten.max.dim %1770, %int-1, %true : !torch.vtensor<[20,1024,64],f16>, !torch.int, !torch.bool -> !torch.vtensor<[20,1024,1],f16>, !torch.vtensor<[20,1024,1],si64>
    %1771 = torch.aten.sub.Tensor %1770, %values_12, %float1.000000e00 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,1024,1],f16>, !torch.float -> !torch.vtensor<[20,1024,64],f16>
    %1772 = torch.aten.exp %1771 : !torch.vtensor<[20,1024,64],f16> -> !torch.vtensor<[20,1024,64],f16>
    %1773 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %1774 = torch.aten.sum.dim_IntList %1772, %1773, %true, %none : !torch.vtensor<[20,1024,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[20,1024,1],f16>
    %1775 = torch.aten.div.Tensor %1772, %1774 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,1024,1],f16> -> !torch.vtensor<[20,1024,64],f16>
    %1776 = torch.aten.bmm %1775, %1764 : !torch.vtensor<[20,1024,64],f16>, !torch.vtensor<[20,64,64],f16> -> !torch.vtensor<[20,1024,64],f16>
    %1777 = torch.aten.view %1776, %1439 : !torch.vtensor<[20,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[2,10,1024,64],f16>
    %1778 = torch.aten.permute %1777, %847 : !torch.vtensor<[2,10,1024,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,10,64],f16>
    %1779 = torch.aten.clone %1778, %int0 : !torch.vtensor<[2,1024,10,64],f16>, !torch.int -> !torch.vtensor<[2,1024,10,64],f16>
    %1780 = torch.aten.view %1779, %1372 : !torch.vtensor<[2,1024,10,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1781 = torch.aten.transpose.int %541, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1782 = torch.aten.view %1780, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1783 = torch.aten.mm %1782, %1781 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1784 = torch.aten.mul.Scalar %540, %int1 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[640],f16>
    %1785 = torch.aten.add.Tensor %1784, %1783, %int1 : !torch.vtensor<[640],f16>, !torch.vtensor<[2048,640],f16>, !torch.int -> !torch.vtensor<[2048,640],f16>
    %1786 = torch.aten.view %1785, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1787 = torch.aten.add.Tensor %1786, %1724, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1788 = torch.aten.to.dtype %1787, %int7, %false, %false, %none : !torch.vtensor<[2,1024,640],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,640],f64>
    %1789 = torch.aten.sum.dim_IntList %1788, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1790 = torch.aten.div.Scalar %1789, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1791 = torch.aten.sub.Tensor %1788, %1790, %float1.000000e00 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,1],f64>, !torch.float -> !torch.vtensor<[2,1024,640],f64>
    %1792 = torch.aten.mul.Tensor %1791, %1791 : !torch.vtensor<[2,1024,640],f64>, !torch.vtensor<[2,1024,640],f64> -> !torch.vtensor<[2,1024,640],f64>
    %1793 = torch.aten.sum.dim_IntList %1792, %688, %true, %none : !torch.vtensor<[2,1024,640],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f64>
    %1794 = torch.aten.div.Scalar %1793, %int640 : !torch.vtensor<[2,1024,1],f64>, !torch.int -> !torch.vtensor<[2,1024,1],f64>
    %1795 = torch.aten.to.dtype %1794, %int5, %false, %false, %none : !torch.vtensor<[2,1024,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1796 = torch.aten.sum.dim_IntList %1787, %688, %true, %none : !torch.vtensor<[2,1024,640],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,1024,1],f16>
    %1797 = torch.aten.div.Scalar %1796, %int640 : !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1798 = torch.aten.add.Scalar %1795, %float1.000000e-05, %int1 : !torch.vtensor<[2,1024,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,1024,1],f16>
    %1799 = torch.aten.rsqrt %1798 : !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,1],f16>
    %1800 = torch.aten.sub.Tensor %1787, %1797, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1801 = torch.aten.mul.Tensor %1800, %1799 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,1],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1802 = torch.aten.mul.Tensor %1801, %539 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16> -> !torch.vtensor<[2,1024,640],f16>
    %1803 = torch.aten.add.Tensor %1802, %538, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1804 = torch.aten.transpose.int %537, %int0, %int1 : !torch.vtensor<[5120,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,5120],f16>
    %1805 = torch.aten.view %1803, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1806 = torch.aten.mm %1805, %1804 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,5120],f16> -> !torch.vtensor<[2048,5120],f16>
    %1807 = torch.aten.mul.Scalar %536, %int1 : !torch.vtensor<[5120],f16>, !torch.int -> !torch.vtensor<[5120],f16>
    %1808 = torch.aten.add.Tensor %1807, %1806, %int1 : !torch.vtensor<[5120],f16>, !torch.vtensor<[2048,5120],f16>, !torch.int -> !torch.vtensor<[2048,5120],f16>
    %1809 = torch.aten.view %1808, %1538 : !torch.vtensor<[2048,5120],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,5120],f16>
    %1810 = torch.aten.slice.Tensor %1809, %int-1, %int0, %int2560, %int1 : !torch.vtensor<[2,1024,5120],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,1024,2560],f16>
    %1811 = torch.aten.slice.Tensor %1809, %int-1, %int2560, %int5120, %int1 : !torch.vtensor<[2,1024,5120],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,1024,2560],f16>
    %1812 = torch.aten.gelu %1811, %str : !torch.vtensor<[2,1024,2560],f16>, !torch.str -> !torch.vtensor<[2,1024,2560],f16>
    %1813 = torch.aten.mul.Tensor %1810, %1812 : !torch.vtensor<[2,1024,2560],f16>, !torch.vtensor<[2,1024,2560],f16> -> !torch.vtensor<[2,1024,2560],f16>
    %1814 = torch.aten.transpose.int %535, %int0, %int1 : !torch.vtensor<[640,2560],f16>, !torch.int, !torch.int -> !torch.vtensor<[2560,640],f16>
    %1815 = torch.aten.view %1813, %1545 : !torch.vtensor<[2,1024,2560],f16>, !torch.list<int> -> !torch.vtensor<[2048,2560],f16>
    %1816 = torch.aten.mm %1815, %1814 : !torch.vtensor<[2048,2560],f16>, !torch.vtensor<[2560,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1817 = torch.aten.mul.Scalar %534, %int1 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[640],f16>
    %1818 = torch.aten.add.Tensor %1817, %1816, %int1 : !torch.vtensor<[640],f16>, !torch.vtensor<[2048,640],f16>, !torch.int -> !torch.vtensor<[2048,640],f16>
    %1819 = torch.aten.view %1818, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1820 = torch.aten.add.Tensor %1819, %1787, %int1 : !torch.vtensor<[2,1024,640],f16>, !torch.vtensor<[2,1024,640],f16>, !torch.int -> !torch.vtensor<[2,1024,640],f16>
    %1821 = torch.aten.transpose.int %533, %int0, %int1 : !torch.vtensor<[640,640],f16>, !torch.int, !torch.int -> !torch.vtensor<[640,640],f16>
    %1822 = torch.aten.view %1820, %1400 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2048,640],f16>
    %1823 = torch.aten.mm %1822, %1821 : !torch.vtensor<[2048,640],f16>, !torch.vtensor<[640,640],f16> -> !torch.vtensor<[2048,640],f16>
    %1824 = torch.aten.mul.Scalar %532, %int1 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[640],f16>
    %1825 = torch.aten.add.Tensor %1824, %1823, %int1 : !torch.vtensor<[640],f16>, !torch.vtensor<[2048,640],f16>, !torch.int -> !torch.vtensor<[2048,640],f16>
    %1826 = torch.aten.view %1825, %1372 : !torch.vtensor<[2048,640],f16>, !torch.list<int> -> !torch.vtensor<[2,1024,640],f16>
    %1827 = torch.aten.view %1826, %1558 : !torch.vtensor<[2,1024,640],f16>, !torch.list<int> -> !torch.vtensor<[2,32,32,640],f16>
    %1828 = torch.aten.permute %1827, %1003 : !torch.vtensor<[2,32,32,640],f16>, !torch.list<int> -> !torch.vtensor<[2,640,32,32],f16>
    %1829 = torch.aten.clone %1828, %int0 : !torch.vtensor<[2,640,32,32],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1830 = torch.aten.add.Tensor %1829, %1627, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[2,640,32,32],f16>, !torch.int -> !torch.vtensor<[2,640,32,32],f16>
    %1831 = torch.aten.convolution %1830, %531, %530, %1275, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,640,32,32],f16>, !torch.vtensor<[640,640,3,3],f16>, !torch.vtensor<[640],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,640,16,16],f16>
    %1832 = torch.prim.ListConstruct %int2, %int32, %int20, %int256 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1833 = torch.aten.view %1831, %1832 : !torch.vtensor<[2,640,16,16],f16>, !torch.list<int> -> !torch.vtensor<[2,32,20,256],f16>
    %1834 = torch.aten.to.dtype %1833, %int7, %false, %false, %none : !torch.vtensor<[2,32,20,256],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,20,256],f64>
    %1835 = torch.aten.sum.dim_IntList %1834, %721, %true, %none : !torch.vtensor<[2,32,20,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1836 = torch.aten.div.Scalar %1835, %int5120 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1837 = torch.aten.sub.Tensor %1834, %1836, %float1.000000e00 : !torch.vtensor<[2,32,20,256],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,20,256],f64>
    %1838 = torch.aten.mul.Tensor %1837, %1837 : !torch.vtensor<[2,32,20,256],f64>, !torch.vtensor<[2,32,20,256],f64> -> !torch.vtensor<[2,32,20,256],f64>
    %1839 = torch.aten.sum.dim_IntList %1838, %721, %true, %none : !torch.vtensor<[2,32,20,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1840 = torch.aten.div.Scalar %1839, %int5120 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1841 = torch.aten.to.dtype %1840, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1842 = torch.aten.sum.dim_IntList %1833, %721, %true, %none : !torch.vtensor<[2,32,20,256],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1843 = torch.aten.div.Scalar %1842, %int5120 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1844 = torch.aten.add.Scalar %1841, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1845 = torch.aten.rsqrt %1844 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1846 = torch.aten.sub.Tensor %1833, %1843, %int1 : !torch.vtensor<[2,32,20,256],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,20,256],f16>
    %1847 = torch.aten.mul.Tensor %1846, %1845 : !torch.vtensor<[2,32,20,256],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,20,256],f16>
    %1848 = torch.prim.ListConstruct %int2, %int640, %int16, %int16 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1849 = torch.aten.view %1847, %1848 : !torch.vtensor<[2,32,20,256],f16>, !torch.list<int> -> !torch.vtensor<[2,640,16,16],f16>
    %1850 = torch.aten.unsqueeze %529, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1851 = torch.aten.unsqueeze %1850, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1852 = torch.aten.unsqueeze %1851, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1853 = torch.aten.unsqueeze %528, %int0 : !torch.vtensor<[640],f16>, !torch.int -> !torch.vtensor<[1,640],f16>
    %1854 = torch.aten.unsqueeze %1853, %int2 : !torch.vtensor<[1,640],f16>, !torch.int -> !torch.vtensor<[1,640,1],f16>
    %1855 = torch.aten.unsqueeze %1854, %int3 : !torch.vtensor<[1,640,1],f16>, !torch.int -> !torch.vtensor<[1,640,1,1],f16>
    %1856 = torch.aten.mul.Tensor %1849, %1855 : !torch.vtensor<[2,640,16,16],f16>, !torch.vtensor<[1,640,1,1],f16> -> !torch.vtensor<[2,640,16,16],f16>
    %1857 = torch.aten.add.Tensor %1856, %1852, %int1 : !torch.vtensor<[2,640,16,16],f16>, !torch.vtensor<[1,640,1,1],f16>, !torch.int -> !torch.vtensor<[2,640,16,16],f16>
    %1858 = torch.aten.sigmoid %1857 : !torch.vtensor<[2,640,16,16],f16> -> !torch.vtensor<[2,640,16,16],f16>
    %1859 = torch.aten.mul.Tensor %1858, %1857 : !torch.vtensor<[2,640,16,16],f16>, !torch.vtensor<[2,640,16,16],f16> -> !torch.vtensor<[2,640,16,16],f16>
    %1860 = torch.aten.convolution %1859, %527, %526, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,640,16,16],f16>, !torch.vtensor<[1280,640,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %1861 = torch.aten.sigmoid %715 : !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %1862 = torch.aten.mul.Tensor %1861, %715 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %1863 = torch.aten.transpose.int %525, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %1864 = torch.aten.mm %1862, %1863 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %1865 = torch.aten.mul.Scalar %524, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %1866 = torch.aten.add.Tensor %1865, %1864, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280],f16>
    %1867 = torch.aten.unsqueeze %1866, %int2 : !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280,1],f16>
    %1868 = torch.aten.unsqueeze %1867, %int3 : !torch.vtensor<[2,1280,1],f16>, !torch.int -> !torch.vtensor<[2,1280,1,1],f16>
    %1869 = torch.aten.add.Tensor %1860, %1868, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[2,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %1870 = torch.prim.ListConstruct %int2, %int32, %int40, %int256 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1871 = torch.aten.view %1869, %1870 : !torch.vtensor<[2,1280,16,16],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,256],f16>
    %1872 = torch.aten.to.dtype %1871, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,256],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,256],f64>
    %1873 = torch.aten.sum.dim_IntList %1872, %721, %true, %none : !torch.vtensor<[2,32,40,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1874 = torch.aten.div.Scalar %1873, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1875 = torch.aten.sub.Tensor %1872, %1874, %float1.000000e00 : !torch.vtensor<[2,32,40,256],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,256],f64>
    %1876 = torch.aten.mul.Tensor %1875, %1875 : !torch.vtensor<[2,32,40,256],f64>, !torch.vtensor<[2,32,40,256],f64> -> !torch.vtensor<[2,32,40,256],f64>
    %1877 = torch.aten.sum.dim_IntList %1876, %721, %true, %none : !torch.vtensor<[2,32,40,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1878 = torch.aten.div.Scalar %1877, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1879 = torch.aten.to.dtype %1878, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1880 = torch.aten.sum.dim_IntList %1871, %721, %true, %none : !torch.vtensor<[2,32,40,256],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1881 = torch.aten.div.Scalar %1880, %int10240 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1882 = torch.aten.add.Scalar %1879, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1883 = torch.aten.rsqrt %1882 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1884 = torch.aten.sub.Tensor %1871, %1881, %int1 : !torch.vtensor<[2,32,40,256],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,256],f16>
    %1885 = torch.aten.mul.Tensor %1884, %1883 : !torch.vtensor<[2,32,40,256],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,256],f16>
    %1886 = torch.prim.ListConstruct %int2, %int1280, %int16, %int16 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1887 = torch.aten.view %1885, %1886 : !torch.vtensor<[2,32,40,256],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,16,16],f16>
    %1888 = torch.aten.unsqueeze %523, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %1889 = torch.aten.unsqueeze %1888, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %1890 = torch.aten.unsqueeze %1889, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %1891 = torch.aten.unsqueeze %522, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %1892 = torch.aten.unsqueeze %1891, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %1893 = torch.aten.unsqueeze %1892, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %1894 = torch.aten.mul.Tensor %1887, %1893 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %1895 = torch.aten.add.Tensor %1894, %1890, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %1896 = torch.aten.sigmoid %1895 : !torch.vtensor<[2,1280,16,16],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %1897 = torch.aten.mul.Tensor %1896, %1895 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[2,1280,16,16],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %1898 = torch.aten.convolution %1897, %521, %520, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1280,1280,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %1899 = torch.aten.convolution %1831, %519, %518, %716, %717, %716, %false, %717, %int1 : !torch.vtensor<[2,640,16,16],f16>, !torch.vtensor<[1280,640,1,1],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %1900 = torch.aten.add.Tensor %1899, %1898, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[2,1280,16,16],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %1901 = torch.aten.div.Scalar %1900, %float1.000000e00 : !torch.vtensor<[2,1280,16,16],f16>, !torch.float -> !torch.vtensor<[2,1280,16,16],f16>
    %1902 = torch.aten.view %1901, %1870 : !torch.vtensor<[2,1280,16,16],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,256],f16>
    %1903 = torch.aten.to.dtype %1902, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,256],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,256],f64>
    %1904 = torch.aten.sum.dim_IntList %1903, %721, %true, %none : !torch.vtensor<[2,32,40,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1905 = torch.aten.div.Scalar %1904, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1906 = torch.aten.sub.Tensor %1903, %1905, %float1.000000e00 : !torch.vtensor<[2,32,40,256],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,256],f64>
    %1907 = torch.aten.mul.Tensor %1906, %1906 : !torch.vtensor<[2,32,40,256],f64>, !torch.vtensor<[2,32,40,256],f64> -> !torch.vtensor<[2,32,40,256],f64>
    %1908 = torch.aten.sum.dim_IntList %1907, %721, %true, %none : !torch.vtensor<[2,32,40,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %1909 = torch.aten.div.Scalar %1908, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %1910 = torch.aten.to.dtype %1909, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1911 = torch.aten.sum.dim_IntList %1902, %721, %true, %none : !torch.vtensor<[2,32,40,256],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %1912 = torch.aten.div.Scalar %1911, %int10240 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1913 = torch.aten.add.Scalar %1910, %float9.999990e-07, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %1914 = torch.aten.rsqrt %1913 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %1915 = torch.aten.sub.Tensor %1902, %1912, %int1 : !torch.vtensor<[2,32,40,256],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,256],f16>
    %1916 = torch.aten.mul.Tensor %1915, %1914 : !torch.vtensor<[2,32,40,256],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,256],f16>
    %1917 = torch.aten.view %1916, %1886 : !torch.vtensor<[2,32,40,256],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,16,16],f16>
    %1918 = torch.aten.unsqueeze %517, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %1919 = torch.aten.unsqueeze %1918, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %1920 = torch.aten.unsqueeze %1919, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %1921 = torch.aten.unsqueeze %516, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %1922 = torch.aten.unsqueeze %1921, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %1923 = torch.aten.unsqueeze %1922, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %1924 = torch.aten.mul.Tensor %1917, %1923 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %1925 = torch.aten.add.Tensor %1924, %1920, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %1926 = torch.aten.permute %1925, %811 : !torch.vtensor<[2,1280,16,16],f16>, !torch.list<int> -> !torch.vtensor<[2,16,16,1280],f16>
    %1927 = torch.prim.ListConstruct %int2, %int256, %int1280 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1928 = torch.aten.view %1926, %1927 : !torch.vtensor<[2,16,16,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %1929 = torch.aten.transpose.int %515, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %1930 = torch.aten.broadcast_to %1928, %1927 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %1931 = torch.aten.view %1930, %1927 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %1932 = torch.prim.ListConstruct %int2, %int1280, %int1280 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1933 = torch.aten.broadcast_to %1929, %1932 : !torch.vtensor<[1280,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,1280],f16>
    %1934 = torch.aten.view %1933, %1932 : !torch.vtensor<[2,1280,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,1280],f16>
    %1935 = torch.aten.bmm %1931, %1934 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,1280,1280],f16> -> !torch.vtensor<[2,256,1280],f16>
    %1936 = torch.aten.view %1935, %1927 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %1937 = torch.aten.add.Tensor %1936, %514, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %1938 = torch.aten.to.dtype %1937, %int7, %false, %false, %none : !torch.vtensor<[2,256,1280],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1280],f64>
    %1939 = torch.aten.sum.dim_IntList %1938, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %1940 = torch.aten.div.Scalar %1939, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %1941 = torch.aten.sub.Tensor %1938, %1940, %float1.000000e00 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1],f64>, !torch.float -> !torch.vtensor<[2,256,1280],f64>
    %1942 = torch.aten.mul.Tensor %1941, %1941 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1280],f64> -> !torch.vtensor<[2,256,1280],f64>
    %1943 = torch.aten.sum.dim_IntList %1942, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %1944 = torch.aten.div.Scalar %1943, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %1945 = torch.aten.to.dtype %1944, %int5, %false, %false, %none : !torch.vtensor<[2,256,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %1946 = torch.aten.sum.dim_IntList %1937, %688, %true, %none : !torch.vtensor<[2,256,1280],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %1947 = torch.aten.div.Scalar %1946, %int1280 : !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %1948 = torch.aten.add.Scalar %1945, %float1.000000e-05, %int1 : !torch.vtensor<[2,256,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %1949 = torch.aten.rsqrt %1948 : !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1],f16>
    %1950 = torch.aten.sub.Tensor %1937, %1947, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %1951 = torch.aten.mul.Tensor %1950, %1949 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1280],f16>
    %1952 = torch.aten.mul.Tensor %1951, %513 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16> -> !torch.vtensor<[2,256,1280],f16>
    %1953 = torch.aten.add.Tensor %1952, %512, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %1954 = torch.aten.transpose.int %511, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %1955 = torch.prim.ListConstruct %int512, %int1280 : (!torch.int, !torch.int) -> !torch.list<int>
    %1956 = torch.aten.view %1953, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %1957 = torch.aten.mm %1956, %1954 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %1958 = torch.aten.view %1957, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %1959 = torch.prim.ListConstruct %int2, %int256, %int20, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1960 = torch.aten.view %1958, %1959 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %1961 = torch.aten.permute %1960, %847 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %1962 = torch.aten.clone %1961, %int0 : !torch.vtensor<[2,20,256,64],f16>, !torch.int -> !torch.vtensor<[2,20,256,64],f16>
    %1963 = torch.prim.ListConstruct %int40, %int256, %int64 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1964 = torch.aten.view %1962, %1963 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[40,256,64],f16>
    %1965 = torch.aten.transpose.int %510, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %1966 = torch.aten.view %1953, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %1967 = torch.aten.mm %1966, %1965 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %1968 = torch.aten.view %1967, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %1969 = torch.aten.transpose.int %509, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %1970 = torch.aten.view %1953, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %1971 = torch.aten.mm %1970, %1969 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %1972 = torch.aten.view %1971, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %1973 = torch.aten.view %1968, %1959 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %1974 = torch.aten.permute %1973, %847 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %1975 = torch.aten.clone %1974, %int0 : !torch.vtensor<[2,20,256,64],f16>, !torch.int -> !torch.vtensor<[2,20,256,64],f16>
    %1976 = torch.aten.view %1975, %1963 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[40,256,64],f16>
    %1977 = torch.aten.view %1972, %1959 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %1978 = torch.aten.permute %1977, %847 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %1979 = torch.aten.clone %1978, %int0 : !torch.vtensor<[2,20,256,64],f16>, !torch.int -> !torch.vtensor<[2,20,256,64],f16>
    %1980 = torch.aten.view %1979, %1963 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[40,256,64],f16>
    %1981 = torch.prim.ListConstruct %int40, %int256, %int256 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1982 = torch.aten.empty.memory_format %1981, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[40,256,256],f32>
    %1983 = torch.aten.transpose.int %1976, %int-1, %int-2 : !torch.vtensor<[40,256,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[40,64,256],f16>
    %1984 = torch.aten.bmm %1964, %1983 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,64,256],f16> -> !torch.vtensor<[40,256,256],f16>
    %1985 = torch.aten.mul.Scalar %1984, %float1.250000e-01 : !torch.vtensor<[40,256,256],f16>, !torch.float -> !torch.vtensor<[40,256,256],f16>
    %1986 = torch.aten.to.dtype %1982, %int5, %false, %false, %none : !torch.vtensor<[40,256,256],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[40,256,256],f16>
    %1987 = torch.aten.add.Tensor %1985, %1982, %int0 : !torch.vtensor<[40,256,256],f16>, !torch.vtensor<[40,256,256],f32>, !torch.int -> !torch.vtensor<[40,256,256],f16>
    %values_14, %indices_15 = torch.aten.max.dim %1987, %int-1, %true : !torch.vtensor<[40,256,256],f16>, !torch.int, !torch.bool -> !torch.vtensor<[40,256,1],f16>, !torch.vtensor<[40,256,1],si64>
    %1988 = torch.aten.sub.Tensor %1987, %values_14, %float1.000000e00 : !torch.vtensor<[40,256,256],f16>, !torch.vtensor<[40,256,1],f16>, !torch.float -> !torch.vtensor<[40,256,256],f16>
    %1989 = torch.aten.exp %1988 : !torch.vtensor<[40,256,256],f16> -> !torch.vtensor<[40,256,256],f16>
    %1990 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %1991 = torch.aten.sum.dim_IntList %1989, %1990, %true, %none : !torch.vtensor<[40,256,256],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[40,256,1],f16>
    %1992 = torch.aten.div.Tensor %1989, %1991 : !torch.vtensor<[40,256,256],f16>, !torch.vtensor<[40,256,1],f16> -> !torch.vtensor<[40,256,256],f16>
    %1993 = torch.aten.bmm %1992, %1980 : !torch.vtensor<[40,256,256],f16>, !torch.vtensor<[40,256,64],f16> -> !torch.vtensor<[40,256,64],f16>
    %1994 = torch.prim.ListConstruct %int2, %int20, %int256, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %1995 = torch.aten.view %1993, %1994 : !torch.vtensor<[40,256,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %1996 = torch.aten.permute %1995, %847 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %1997 = torch.aten.clone %1996, %int0 : !torch.vtensor<[2,256,20,64],f16>, !torch.int -> !torch.vtensor<[2,256,20,64],f16>
    %1998 = torch.aten.view %1997, %1927 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %1999 = torch.aten.transpose.int %508, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2000 = torch.aten.view %1998, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2001 = torch.aten.mm %2000, %1999 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2002 = torch.aten.mul.Scalar %507, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2003 = torch.aten.add.Tensor %2002, %2001, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[512,1280],f16>, !torch.int -> !torch.vtensor<[512,1280],f16>
    %2004 = torch.aten.view %2003, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2005 = torch.aten.add.Tensor %2004, %1937, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2006 = torch.aten.to.dtype %2005, %int7, %false, %false, %none : !torch.vtensor<[2,256,1280],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1280],f64>
    %2007 = torch.aten.sum.dim_IntList %2006, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %2008 = torch.aten.div.Scalar %2007, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %2009 = torch.aten.sub.Tensor %2006, %2008, %float1.000000e00 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1],f64>, !torch.float -> !torch.vtensor<[2,256,1280],f64>
    %2010 = torch.aten.mul.Tensor %2009, %2009 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1280],f64> -> !torch.vtensor<[2,256,1280],f64>
    %2011 = torch.aten.sum.dim_IntList %2010, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %2012 = torch.aten.div.Scalar %2011, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %2013 = torch.aten.to.dtype %2012, %int5, %false, %false, %none : !torch.vtensor<[2,256,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %2014 = torch.aten.sum.dim_IntList %2005, %688, %true, %none : !torch.vtensor<[2,256,1280],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %2015 = torch.aten.div.Scalar %2014, %int1280 : !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %2016 = torch.aten.add.Scalar %2013, %float1.000000e-05, %int1 : !torch.vtensor<[2,256,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %2017 = torch.aten.rsqrt %2016 : !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1],f16>
    %2018 = torch.aten.sub.Tensor %2005, %2015, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2019 = torch.aten.mul.Tensor %2018, %2017 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2020 = torch.aten.mul.Tensor %2019, %506 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2021 = torch.aten.add.Tensor %2020, %505, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2022 = torch.aten.transpose.int %504, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2023 = torch.aten.view %2021, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2024 = torch.aten.mm %2023, %2022 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2025 = torch.aten.view %2024, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2026 = torch.aten.view %2025, %1959 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %2027 = torch.aten.permute %2026, %847 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %2028 = torch.aten.clone %2027, %int0 : !torch.vtensor<[2,20,256,64],f16>, !torch.int -> !torch.vtensor<[2,20,256,64],f16>
    %2029 = torch.aten.view %2028, %1963 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[40,256,64],f16>
    %2030 = torch.aten.transpose.int %503, %int0, %int1 : !torch.vtensor<[1280,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,1280],f16>
    %2031 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %2032 = torch.aten.mm %2031, %2030 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2033 = torch.prim.ListConstruct %int2, %int64, %int1280 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %2034 = torch.aten.view %2032, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2035 = torch.aten.transpose.int %502, %int0, %int1 : !torch.vtensor<[1280,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,1280],f16>
    %2036 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %2037 = torch.aten.mm %2036, %2035 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2038 = torch.aten.view %2037, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2039 = torch.prim.ListConstruct %int2, %int64, %int20, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %2040 = torch.aten.view %2034, %2039 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2041 = torch.aten.permute %2040, %847 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2042 = torch.aten.clone %2041, %int0 : !torch.vtensor<[2,20,64,64],f16>, !torch.int -> !torch.vtensor<[2,20,64,64],f16>
    %2043 = torch.prim.ListConstruct %int40, %int64, %int64 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %2044 = torch.aten.view %2042, %2043 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[40,64,64],f16>
    %2045 = torch.aten.view %2038, %2039 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2046 = torch.aten.permute %2045, %847 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2047 = torch.aten.clone %2046, %int0 : !torch.vtensor<[2,20,64,64],f16>, !torch.int -> !torch.vtensor<[2,20,64,64],f16>
    %2048 = torch.aten.view %2047, %2043 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[40,64,64],f16>
    %2049 = torch.aten.empty.memory_format %1963, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[40,256,64],f32>
    %2050 = torch.aten.transpose.int %2044, %int-1, %int-2 : !torch.vtensor<[40,64,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[40,64,64],f16>
    %2051 = torch.aten.bmm %2029, %2050 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,64,64],f16> -> !torch.vtensor<[40,256,64],f16>
    %2052 = torch.aten.mul.Scalar %2051, %float1.250000e-01 : !torch.vtensor<[40,256,64],f16>, !torch.float -> !torch.vtensor<[40,256,64],f16>
    %2053 = torch.aten.to.dtype %2049, %int5, %false, %false, %none : !torch.vtensor<[40,256,64],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[40,256,64],f16>
    %2054 = torch.aten.add.Tensor %2052, %2049, %int0 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,256,64],f32>, !torch.int -> !torch.vtensor<[40,256,64],f16>
    %values_16, %indices_17 = torch.aten.max.dim %2054, %int-1, %true : !torch.vtensor<[40,256,64],f16>, !torch.int, !torch.bool -> !torch.vtensor<[40,256,1],f16>, !torch.vtensor<[40,256,1],si64>
    %2055 = torch.aten.sub.Tensor %2054, %values_16, %float1.000000e00 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,256,1],f16>, !torch.float -> !torch.vtensor<[40,256,64],f16>
    %2056 = torch.aten.exp %2055 : !torch.vtensor<[40,256,64],f16> -> !torch.vtensor<[40,256,64],f16>
    %2057 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %2058 = torch.aten.sum.dim_IntList %2056, %2057, %true, %none : !torch.vtensor<[40,256,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[40,256,1],f16>
    %2059 = torch.aten.div.Tensor %2056, %2058 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,256,1],f16> -> !torch.vtensor<[40,256,64],f16>
    %2060 = torch.aten.bmm %2059, %2048 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,64,64],f16> -> !torch.vtensor<[40,256,64],f16>
    %2061 = torch.aten.view %2060, %1994 : !torch.vtensor<[40,256,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %2062 = torch.aten.permute %2061, %847 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %2063 = torch.aten.clone %2062, %int0 : !torch.vtensor<[2,256,20,64],f16>, !torch.int -> !torch.vtensor<[2,256,20,64],f16>
    %2064 = torch.aten.view %2063, %1927 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2065 = torch.aten.transpose.int %501, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2066 = torch.aten.view %2064, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2067 = torch.aten.mm %2066, %2065 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2068 = torch.aten.mul.Scalar %500, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2069 = torch.aten.add.Tensor %2068, %2067, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[512,1280],f16>, !torch.int -> !torch.vtensor<[512,1280],f16>
    %2070 = torch.aten.view %2069, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2071 = torch.aten.add.Tensor %2070, %2005, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2072 = torch.aten.to.dtype %2071, %int7, %false, %false, %none : !torch.vtensor<[2,256,1280],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1280],f64>
    %2073 = torch.aten.sum.dim_IntList %2072, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %2074 = torch.aten.div.Scalar %2073, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %2075 = torch.aten.sub.Tensor %2072, %2074, %float1.000000e00 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1],f64>, !torch.float -> !torch.vtensor<[2,256,1280],f64>
    %2076 = torch.aten.mul.Tensor %2075, %2075 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1280],f64> -> !torch.vtensor<[2,256,1280],f64>
    %2077 = torch.aten.sum.dim_IntList %2076, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %2078 = torch.aten.div.Scalar %2077, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %2079 = torch.aten.to.dtype %2078, %int5, %false, %false, %none : !torch.vtensor<[2,256,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %2080 = torch.aten.sum.dim_IntList %2071, %688, %true, %none : !torch.vtensor<[2,256,1280],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %2081 = torch.aten.div.Scalar %2080, %int1280 : !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %2082 = torch.aten.add.Scalar %2079, %float1.000000e-05, %int1 : !torch.vtensor<[2,256,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %2083 = torch.aten.rsqrt %2082 : !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1],f16>
    %2084 = torch.aten.sub.Tensor %2071, %2081, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2085 = torch.aten.mul.Tensor %2084, %2083 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2086 = torch.aten.mul.Tensor %2085, %499 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2087 = torch.aten.add.Tensor %2086, %498, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2088 = torch.aten.transpose.int %497, %int0, %int1 : !torch.vtensor<[10240,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,10240],f16>
    %2089 = torch.aten.view %2087, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2090 = torch.aten.mm %2089, %2088 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,10240],f16> -> !torch.vtensor<[512,10240],f16>
    %2091 = torch.aten.mul.Scalar %496, %int1 : !torch.vtensor<[10240],f16>, !torch.int -> !torch.vtensor<[10240],f16>
    %2092 = torch.aten.add.Tensor %2091, %2090, %int1 : !torch.vtensor<[10240],f16>, !torch.vtensor<[512,10240],f16>, !torch.int -> !torch.vtensor<[512,10240],f16>
    %2093 = torch.prim.ListConstruct %int2, %int256, %int10240 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %2094 = torch.aten.view %2092, %2093 : !torch.vtensor<[512,10240],f16>, !torch.list<int> -> !torch.vtensor<[2,256,10240],f16>
    %2095 = torch.aten.slice.Tensor %2094, %int-1, %int0, %int5120, %int1 : !torch.vtensor<[2,256,10240],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,256,5120],f16>
    %2096 = torch.aten.slice.Tensor %2094, %int-1, %int5120, %int10240, %int1 : !torch.vtensor<[2,256,10240],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,256,5120],f16>
    %2097 = torch.aten.gelu %2096, %str : !torch.vtensor<[2,256,5120],f16>, !torch.str -> !torch.vtensor<[2,256,5120],f16>
    %2098 = torch.aten.mul.Tensor %2095, %2097 : !torch.vtensor<[2,256,5120],f16>, !torch.vtensor<[2,256,5120],f16> -> !torch.vtensor<[2,256,5120],f16>
    %2099 = torch.aten.transpose.int %495, %int0, %int1 : !torch.vtensor<[1280,5120],f16>, !torch.int, !torch.int -> !torch.vtensor<[5120,1280],f16>
    %2100 = torch.prim.ListConstruct %int512, %int5120 : (!torch.int, !torch.int) -> !torch.list<int>
    %2101 = torch.aten.view %2098, %2100 : !torch.vtensor<[2,256,5120],f16>, !torch.list<int> -> !torch.vtensor<[512,5120],f16>
    %2102 = torch.aten.mm %2101, %2099 : !torch.vtensor<[512,5120],f16>, !torch.vtensor<[5120,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2103 = torch.aten.mul.Scalar %494, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2104 = torch.aten.add.Tensor %2103, %2102, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[512,1280],f16>, !torch.int -> !torch.vtensor<[512,1280],f16>
    %2105 = torch.aten.view %2104, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2106 = torch.aten.add.Tensor %2105, %2071, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2107 = torch.aten.transpose.int %493, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2108 = torch.aten.view %2106, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2109 = torch.aten.mm %2108, %2107 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2110 = torch.aten.mul.Scalar %492, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2111 = torch.aten.add.Tensor %2110, %2109, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[512,1280],f16>, !torch.int -> !torch.vtensor<[512,1280],f16>
    %2112 = torch.aten.view %2111, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2113 = torch.prim.ListConstruct %int2, %int16, %int16, %int1280 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %2114 = torch.aten.view %2112, %2113 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,16,16,1280],f16>
    %2115 = torch.aten.permute %2114, %1003 : !torch.vtensor<[2,16,16,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,16,16],f16>
    %2116 = torch.aten.clone %2115, %int0 : !torch.vtensor<[2,1280,16,16],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2117 = torch.aten.add.Tensor %2116, %1901, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[2,1280,16,16],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2118 = torch.aten.view %2117, %1870 : !torch.vtensor<[2,1280,16,16],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,256],f16>
    %2119 = torch.aten.to.dtype %2118, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,256],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,256],f64>
    %2120 = torch.aten.sum.dim_IntList %2119, %721, %true, %none : !torch.vtensor<[2,32,40,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2121 = torch.aten.div.Scalar %2120, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2122 = torch.aten.sub.Tensor %2119, %2121, %float1.000000e00 : !torch.vtensor<[2,32,40,256],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,256],f64>
    %2123 = torch.aten.mul.Tensor %2122, %2122 : !torch.vtensor<[2,32,40,256],f64>, !torch.vtensor<[2,32,40,256],f64> -> !torch.vtensor<[2,32,40,256],f64>
    %2124 = torch.aten.sum.dim_IntList %2123, %721, %true, %none : !torch.vtensor<[2,32,40,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2125 = torch.aten.div.Scalar %2124, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2126 = torch.aten.to.dtype %2125, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2127 = torch.aten.sum.dim_IntList %2118, %721, %true, %none : !torch.vtensor<[2,32,40,256],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2128 = torch.aten.div.Scalar %2127, %int10240 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2129 = torch.aten.add.Scalar %2126, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2130 = torch.aten.rsqrt %2129 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %2131 = torch.aten.sub.Tensor %2118, %2128, %int1 : !torch.vtensor<[2,32,40,256],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,256],f16>
    %2132 = torch.aten.mul.Tensor %2131, %2130 : !torch.vtensor<[2,32,40,256],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,256],f16>
    %2133 = torch.aten.view %2132, %1886 : !torch.vtensor<[2,32,40,256],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,16,16],f16>
    %2134 = torch.aten.unsqueeze %491, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2135 = torch.aten.unsqueeze %2134, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2136 = torch.aten.unsqueeze %2135, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2137 = torch.aten.unsqueeze %490, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2138 = torch.aten.unsqueeze %2137, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2139 = torch.aten.unsqueeze %2138, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2140 = torch.aten.mul.Tensor %2133, %2139 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %2141 = torch.aten.add.Tensor %2140, %2136, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2142 = torch.aten.sigmoid %2141 : !torch.vtensor<[2,1280,16,16],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %2143 = torch.aten.mul.Tensor %2142, %2141 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[2,1280,16,16],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %2144 = torch.aten.convolution %2143, %489, %488, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1280,1280,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2145 = torch.aten.sigmoid %715 : !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2146 = torch.aten.mul.Tensor %2145, %715 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2147 = torch.aten.transpose.int %487, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2148 = torch.aten.mm %2146, %2147 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2149 = torch.aten.mul.Scalar %486, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2150 = torch.aten.add.Tensor %2149, %2148, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280],f16>
    %2151 = torch.aten.unsqueeze %2150, %int2 : !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280,1],f16>
    %2152 = torch.aten.unsqueeze %2151, %int3 : !torch.vtensor<[2,1280,1],f16>, !torch.int -> !torch.vtensor<[2,1280,1,1],f16>
    %2153 = torch.aten.add.Tensor %2144, %2152, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[2,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2154 = torch.aten.view %2153, %1870 : !torch.vtensor<[2,1280,16,16],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,256],f16>
    %2155 = torch.aten.to.dtype %2154, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,256],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,256],f64>
    %2156 = torch.aten.sum.dim_IntList %2155, %721, %true, %none : !torch.vtensor<[2,32,40,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2157 = torch.aten.div.Scalar %2156, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2158 = torch.aten.sub.Tensor %2155, %2157, %float1.000000e00 : !torch.vtensor<[2,32,40,256],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,256],f64>
    %2159 = torch.aten.mul.Tensor %2158, %2158 : !torch.vtensor<[2,32,40,256],f64>, !torch.vtensor<[2,32,40,256],f64> -> !torch.vtensor<[2,32,40,256],f64>
    %2160 = torch.aten.sum.dim_IntList %2159, %721, %true, %none : !torch.vtensor<[2,32,40,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2161 = torch.aten.div.Scalar %2160, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2162 = torch.aten.to.dtype %2161, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2163 = torch.aten.sum.dim_IntList %2154, %721, %true, %none : !torch.vtensor<[2,32,40,256],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2164 = torch.aten.div.Scalar %2163, %int10240 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2165 = torch.aten.add.Scalar %2162, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2166 = torch.aten.rsqrt %2165 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %2167 = torch.aten.sub.Tensor %2154, %2164, %int1 : !torch.vtensor<[2,32,40,256],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,256],f16>
    %2168 = torch.aten.mul.Tensor %2167, %2166 : !torch.vtensor<[2,32,40,256],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,256],f16>
    %2169 = torch.aten.view %2168, %1886 : !torch.vtensor<[2,32,40,256],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,16,16],f16>
    %2170 = torch.aten.unsqueeze %485, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2171 = torch.aten.unsqueeze %2170, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2172 = torch.aten.unsqueeze %2171, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2173 = torch.aten.unsqueeze %484, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2174 = torch.aten.unsqueeze %2173, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2175 = torch.aten.unsqueeze %2174, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2176 = torch.aten.mul.Tensor %2169, %2175 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %2177 = torch.aten.add.Tensor %2176, %2172, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2178 = torch.aten.sigmoid %2177 : !torch.vtensor<[2,1280,16,16],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %2179 = torch.aten.mul.Tensor %2178, %2177 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[2,1280,16,16],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %2180 = torch.aten.convolution %2179, %483, %482, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1280,1280,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2181 = torch.aten.add.Tensor %2117, %2180, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[2,1280,16,16],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2182 = torch.aten.div.Scalar %2181, %float1.000000e00 : !torch.vtensor<[2,1280,16,16],f16>, !torch.float -> !torch.vtensor<[2,1280,16,16],f16>
    %2183 = torch.aten.view %2182, %1870 : !torch.vtensor<[2,1280,16,16],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,256],f16>
    %2184 = torch.aten.to.dtype %2183, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,256],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,256],f64>
    %2185 = torch.aten.sum.dim_IntList %2184, %721, %true, %none : !torch.vtensor<[2,32,40,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2186 = torch.aten.div.Scalar %2185, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2187 = torch.aten.sub.Tensor %2184, %2186, %float1.000000e00 : !torch.vtensor<[2,32,40,256],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,256],f64>
    %2188 = torch.aten.mul.Tensor %2187, %2187 : !torch.vtensor<[2,32,40,256],f64>, !torch.vtensor<[2,32,40,256],f64> -> !torch.vtensor<[2,32,40,256],f64>
    %2189 = torch.aten.sum.dim_IntList %2188, %721, %true, %none : !torch.vtensor<[2,32,40,256],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2190 = torch.aten.div.Scalar %2189, %int10240 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2191 = torch.aten.to.dtype %2190, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2192 = torch.aten.sum.dim_IntList %2183, %721, %true, %none : !torch.vtensor<[2,32,40,256],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2193 = torch.aten.div.Scalar %2192, %int10240 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2194 = torch.aten.add.Scalar %2191, %float9.999990e-07, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2195 = torch.aten.rsqrt %2194 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %2196 = torch.aten.sub.Tensor %2183, %2193, %int1 : !torch.vtensor<[2,32,40,256],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,256],f16>
    %2197 = torch.aten.mul.Tensor %2196, %2195 : !torch.vtensor<[2,32,40,256],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,256],f16>
    %2198 = torch.aten.view %2197, %1886 : !torch.vtensor<[2,32,40,256],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,16,16],f16>
    %2199 = torch.aten.unsqueeze %481, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2200 = torch.aten.unsqueeze %2199, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2201 = torch.aten.unsqueeze %2200, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2202 = torch.aten.unsqueeze %480, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2203 = torch.aten.unsqueeze %2202, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2204 = torch.aten.unsqueeze %2203, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2205 = torch.aten.mul.Tensor %2198, %2204 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,16,16],f16>
    %2206 = torch.aten.add.Tensor %2205, %2201, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2207 = torch.aten.permute %2206, %811 : !torch.vtensor<[2,1280,16,16],f16>, !torch.list<int> -> !torch.vtensor<[2,16,16,1280],f16>
    %2208 = torch.aten.view %2207, %1927 : !torch.vtensor<[2,16,16,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2209 = torch.aten.transpose.int %479, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2210 = torch.aten.broadcast_to %2208, %1927 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2211 = torch.aten.view %2210, %1927 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2212 = torch.aten.broadcast_to %2209, %1932 : !torch.vtensor<[1280,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,1280],f16>
    %2213 = torch.aten.view %2212, %1932 : !torch.vtensor<[2,1280,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,1280],f16>
    %2214 = torch.aten.bmm %2211, %2213 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,1280,1280],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2215 = torch.aten.view %2214, %1927 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2216 = torch.aten.add.Tensor %2215, %478, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2217 = torch.aten.to.dtype %2216, %int7, %false, %false, %none : !torch.vtensor<[2,256,1280],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1280],f64>
    %2218 = torch.aten.sum.dim_IntList %2217, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %2219 = torch.aten.div.Scalar %2218, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %2220 = torch.aten.sub.Tensor %2217, %2219, %float1.000000e00 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1],f64>, !torch.float -> !torch.vtensor<[2,256,1280],f64>
    %2221 = torch.aten.mul.Tensor %2220, %2220 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1280],f64> -> !torch.vtensor<[2,256,1280],f64>
    %2222 = torch.aten.sum.dim_IntList %2221, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %2223 = torch.aten.div.Scalar %2222, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %2224 = torch.aten.to.dtype %2223, %int5, %false, %false, %none : !torch.vtensor<[2,256,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %2225 = torch.aten.sum.dim_IntList %2216, %688, %true, %none : !torch.vtensor<[2,256,1280],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %2226 = torch.aten.div.Scalar %2225, %int1280 : !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %2227 = torch.aten.add.Scalar %2224, %float1.000000e-05, %int1 : !torch.vtensor<[2,256,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %2228 = torch.aten.rsqrt %2227 : !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1],f16>
    %2229 = torch.aten.sub.Tensor %2216, %2226, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2230 = torch.aten.mul.Tensor %2229, %2228 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2231 = torch.aten.mul.Tensor %2230, %477 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2232 = torch.aten.add.Tensor %2231, %476, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2233 = torch.aten.transpose.int %475, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2234 = torch.aten.view %2232, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2235 = torch.aten.mm %2234, %2233 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2236 = torch.aten.view %2235, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2237 = torch.aten.view %2236, %1959 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %2238 = torch.aten.permute %2237, %847 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %2239 = torch.aten.clone %2238, %int0 : !torch.vtensor<[2,20,256,64],f16>, !torch.int -> !torch.vtensor<[2,20,256,64],f16>
    %2240 = torch.aten.view %2239, %1963 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[40,256,64],f16>
    %2241 = torch.aten.transpose.int %474, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2242 = torch.aten.view %2232, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2243 = torch.aten.mm %2242, %2241 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2244 = torch.aten.view %2243, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2245 = torch.aten.transpose.int %473, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2246 = torch.aten.view %2232, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2247 = torch.aten.mm %2246, %2245 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2248 = torch.aten.view %2247, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2249 = torch.aten.view %2244, %1959 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %2250 = torch.aten.permute %2249, %847 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %2251 = torch.aten.clone %2250, %int0 : !torch.vtensor<[2,20,256,64],f16>, !torch.int -> !torch.vtensor<[2,20,256,64],f16>
    %2252 = torch.aten.view %2251, %1963 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[40,256,64],f16>
    %2253 = torch.aten.view %2248, %1959 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %2254 = torch.aten.permute %2253, %847 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %2255 = torch.aten.clone %2254, %int0 : !torch.vtensor<[2,20,256,64],f16>, !torch.int -> !torch.vtensor<[2,20,256,64],f16>
    %2256 = torch.aten.view %2255, %1963 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[40,256,64],f16>
    %2257 = torch.aten.empty.memory_format %1981, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[40,256,256],f32>
    %2258 = torch.aten.transpose.int %2252, %int-1, %int-2 : !torch.vtensor<[40,256,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[40,64,256],f16>
    %2259 = torch.aten.bmm %2240, %2258 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,64,256],f16> -> !torch.vtensor<[40,256,256],f16>
    %2260 = torch.aten.mul.Scalar %2259, %float1.250000e-01 : !torch.vtensor<[40,256,256],f16>, !torch.float -> !torch.vtensor<[40,256,256],f16>
    %2261 = torch.aten.to.dtype %2257, %int5, %false, %false, %none : !torch.vtensor<[40,256,256],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[40,256,256],f16>
    %2262 = torch.aten.add.Tensor %2260, %2257, %int0 : !torch.vtensor<[40,256,256],f16>, !torch.vtensor<[40,256,256],f32>, !torch.int -> !torch.vtensor<[40,256,256],f16>
    %values_18, %indices_19 = torch.aten.max.dim %2262, %int-1, %true : !torch.vtensor<[40,256,256],f16>, !torch.int, !torch.bool -> !torch.vtensor<[40,256,1],f16>, !torch.vtensor<[40,256,1],si64>
    %2263 = torch.aten.sub.Tensor %2262, %values_18, %float1.000000e00 : !torch.vtensor<[40,256,256],f16>, !torch.vtensor<[40,256,1],f16>, !torch.float -> !torch.vtensor<[40,256,256],f16>
    %2264 = torch.aten.exp %2263 : !torch.vtensor<[40,256,256],f16> -> !torch.vtensor<[40,256,256],f16>
    %2265 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %2266 = torch.aten.sum.dim_IntList %2264, %2265, %true, %none : !torch.vtensor<[40,256,256],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[40,256,1],f16>
    %2267 = torch.aten.div.Tensor %2264, %2266 : !torch.vtensor<[40,256,256],f16>, !torch.vtensor<[40,256,1],f16> -> !torch.vtensor<[40,256,256],f16>
    %2268 = torch.aten.bmm %2267, %2256 : !torch.vtensor<[40,256,256],f16>, !torch.vtensor<[40,256,64],f16> -> !torch.vtensor<[40,256,64],f16>
    %2269 = torch.aten.view %2268, %1994 : !torch.vtensor<[40,256,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %2270 = torch.aten.permute %2269, %847 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %2271 = torch.aten.clone %2270, %int0 : !torch.vtensor<[2,256,20,64],f16>, !torch.int -> !torch.vtensor<[2,256,20,64],f16>
    %2272 = torch.aten.view %2271, %1927 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2273 = torch.aten.transpose.int %472, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2274 = torch.aten.view %2272, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2275 = torch.aten.mm %2274, %2273 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2276 = torch.aten.mul.Scalar %471, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2277 = torch.aten.add.Tensor %2276, %2275, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[512,1280],f16>, !torch.int -> !torch.vtensor<[512,1280],f16>
    %2278 = torch.aten.view %2277, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2279 = torch.aten.add.Tensor %2278, %2216, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2280 = torch.aten.to.dtype %2279, %int7, %false, %false, %none : !torch.vtensor<[2,256,1280],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1280],f64>
    %2281 = torch.aten.sum.dim_IntList %2280, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %2282 = torch.aten.div.Scalar %2281, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %2283 = torch.aten.sub.Tensor %2280, %2282, %float1.000000e00 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1],f64>, !torch.float -> !torch.vtensor<[2,256,1280],f64>
    %2284 = torch.aten.mul.Tensor %2283, %2283 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1280],f64> -> !torch.vtensor<[2,256,1280],f64>
    %2285 = torch.aten.sum.dim_IntList %2284, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %2286 = torch.aten.div.Scalar %2285, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %2287 = torch.aten.to.dtype %2286, %int5, %false, %false, %none : !torch.vtensor<[2,256,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %2288 = torch.aten.sum.dim_IntList %2279, %688, %true, %none : !torch.vtensor<[2,256,1280],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %2289 = torch.aten.div.Scalar %2288, %int1280 : !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %2290 = torch.aten.add.Scalar %2287, %float1.000000e-05, %int1 : !torch.vtensor<[2,256,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %2291 = torch.aten.rsqrt %2290 : !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1],f16>
    %2292 = torch.aten.sub.Tensor %2279, %2289, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2293 = torch.aten.mul.Tensor %2292, %2291 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2294 = torch.aten.mul.Tensor %2293, %470 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2295 = torch.aten.add.Tensor %2294, %469, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2296 = torch.aten.transpose.int %468, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2297 = torch.aten.view %2295, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2298 = torch.aten.mm %2297, %2296 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2299 = torch.aten.view %2298, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2300 = torch.aten.view %2299, %1959 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %2301 = torch.aten.permute %2300, %847 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %2302 = torch.aten.clone %2301, %int0 : !torch.vtensor<[2,20,256,64],f16>, !torch.int -> !torch.vtensor<[2,20,256,64],f16>
    %2303 = torch.aten.view %2302, %1963 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[40,256,64],f16>
    %2304 = torch.aten.transpose.int %467, %int0, %int1 : !torch.vtensor<[1280,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,1280],f16>
    %2305 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %2306 = torch.aten.mm %2305, %2304 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2307 = torch.aten.view %2306, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2308 = torch.aten.transpose.int %466, %int0, %int1 : !torch.vtensor<[1280,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,1280],f16>
    %2309 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %2310 = torch.aten.mm %2309, %2308 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2311 = torch.aten.view %2310, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2312 = torch.aten.view %2307, %2039 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2313 = torch.aten.permute %2312, %847 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2314 = torch.aten.clone %2313, %int0 : !torch.vtensor<[2,20,64,64],f16>, !torch.int -> !torch.vtensor<[2,20,64,64],f16>
    %2315 = torch.aten.view %2314, %2043 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[40,64,64],f16>
    %2316 = torch.aten.view %2311, %2039 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2317 = torch.aten.permute %2316, %847 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2318 = torch.aten.clone %2317, %int0 : !torch.vtensor<[2,20,64,64],f16>, !torch.int -> !torch.vtensor<[2,20,64,64],f16>
    %2319 = torch.aten.view %2318, %2043 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[40,64,64],f16>
    %2320 = torch.aten.empty.memory_format %1963, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[40,256,64],f32>
    %2321 = torch.aten.transpose.int %2315, %int-1, %int-2 : !torch.vtensor<[40,64,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[40,64,64],f16>
    %2322 = torch.aten.bmm %2303, %2321 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,64,64],f16> -> !torch.vtensor<[40,256,64],f16>
    %2323 = torch.aten.mul.Scalar %2322, %float1.250000e-01 : !torch.vtensor<[40,256,64],f16>, !torch.float -> !torch.vtensor<[40,256,64],f16>
    %2324 = torch.aten.to.dtype %2320, %int5, %false, %false, %none : !torch.vtensor<[40,256,64],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[40,256,64],f16>
    %2325 = torch.aten.add.Tensor %2323, %2320, %int0 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,256,64],f32>, !torch.int -> !torch.vtensor<[40,256,64],f16>
    %values_20, %indices_21 = torch.aten.max.dim %2325, %int-1, %true : !torch.vtensor<[40,256,64],f16>, !torch.int, !torch.bool -> !torch.vtensor<[40,256,1],f16>, !torch.vtensor<[40,256,1],si64>
    %2326 = torch.aten.sub.Tensor %2325, %values_20, %float1.000000e00 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,256,1],f16>, !torch.float -> !torch.vtensor<[40,256,64],f16>
    %2327 = torch.aten.exp %2326 : !torch.vtensor<[40,256,64],f16> -> !torch.vtensor<[40,256,64],f16>
    %2328 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %2329 = torch.aten.sum.dim_IntList %2327, %2328, %true, %none : !torch.vtensor<[40,256,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[40,256,1],f16>
    %2330 = torch.aten.div.Tensor %2327, %2329 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,256,1],f16> -> !torch.vtensor<[40,256,64],f16>
    %2331 = torch.aten.bmm %2330, %2319 : !torch.vtensor<[40,256,64],f16>, !torch.vtensor<[40,64,64],f16> -> !torch.vtensor<[40,256,64],f16>
    %2332 = torch.aten.view %2331, %1994 : !torch.vtensor<[40,256,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,256,64],f16>
    %2333 = torch.aten.permute %2332, %847 : !torch.vtensor<[2,20,256,64],f16>, !torch.list<int> -> !torch.vtensor<[2,256,20,64],f16>
    %2334 = torch.aten.clone %2333, %int0 : !torch.vtensor<[2,256,20,64],f16>, !torch.int -> !torch.vtensor<[2,256,20,64],f16>
    %2335 = torch.aten.view %2334, %1927 : !torch.vtensor<[2,256,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2336 = torch.aten.transpose.int %465, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2337 = torch.aten.view %2335, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2338 = torch.aten.mm %2337, %2336 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2339 = torch.aten.mul.Scalar %464, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2340 = torch.aten.add.Tensor %2339, %2338, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[512,1280],f16>, !torch.int -> !torch.vtensor<[512,1280],f16>
    %2341 = torch.aten.view %2340, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2342 = torch.aten.add.Tensor %2341, %2279, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2343 = torch.aten.to.dtype %2342, %int7, %false, %false, %none : !torch.vtensor<[2,256,1280],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1280],f64>
    %2344 = torch.aten.sum.dim_IntList %2343, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %2345 = torch.aten.div.Scalar %2344, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %2346 = torch.aten.sub.Tensor %2343, %2345, %float1.000000e00 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1],f64>, !torch.float -> !torch.vtensor<[2,256,1280],f64>
    %2347 = torch.aten.mul.Tensor %2346, %2346 : !torch.vtensor<[2,256,1280],f64>, !torch.vtensor<[2,256,1280],f64> -> !torch.vtensor<[2,256,1280],f64>
    %2348 = torch.aten.sum.dim_IntList %2347, %688, %true, %none : !torch.vtensor<[2,256,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f64>
    %2349 = torch.aten.div.Scalar %2348, %int1280 : !torch.vtensor<[2,256,1],f64>, !torch.int -> !torch.vtensor<[2,256,1],f64>
    %2350 = torch.aten.to.dtype %2349, %int5, %false, %false, %none : !torch.vtensor<[2,256,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %2351 = torch.aten.sum.dim_IntList %2342, %688, %true, %none : !torch.vtensor<[2,256,1280],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,256,1],f16>
    %2352 = torch.aten.div.Scalar %2351, %int1280 : !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %2353 = torch.aten.add.Scalar %2350, %float1.000000e-05, %int1 : !torch.vtensor<[2,256,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,256,1],f16>
    %2354 = torch.aten.rsqrt %2353 : !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1],f16>
    %2355 = torch.aten.sub.Tensor %2342, %2352, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2356 = torch.aten.mul.Tensor %2355, %2354 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2357 = torch.aten.mul.Tensor %2356, %463 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16> -> !torch.vtensor<[2,256,1280],f16>
    %2358 = torch.aten.add.Tensor %2357, %462, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2359 = torch.aten.transpose.int %461, %int0, %int1 : !torch.vtensor<[10240,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,10240],f16>
    %2360 = torch.aten.view %2358, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2361 = torch.aten.mm %2360, %2359 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,10240],f16> -> !torch.vtensor<[512,10240],f16>
    %2362 = torch.aten.mul.Scalar %460, %int1 : !torch.vtensor<[10240],f16>, !torch.int -> !torch.vtensor<[10240],f16>
    %2363 = torch.aten.add.Tensor %2362, %2361, %int1 : !torch.vtensor<[10240],f16>, !torch.vtensor<[512,10240],f16>, !torch.int -> !torch.vtensor<[512,10240],f16>
    %2364 = torch.aten.view %2363, %2093 : !torch.vtensor<[512,10240],f16>, !torch.list<int> -> !torch.vtensor<[2,256,10240],f16>
    %2365 = torch.aten.slice.Tensor %2364, %int-1, %int0, %int5120, %int1 : !torch.vtensor<[2,256,10240],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,256,5120],f16>
    %2366 = torch.aten.slice.Tensor %2364, %int-1, %int5120, %int10240, %int1 : !torch.vtensor<[2,256,10240],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,256,5120],f16>
    %2367 = torch.aten.gelu %2366, %str : !torch.vtensor<[2,256,5120],f16>, !torch.str -> !torch.vtensor<[2,256,5120],f16>
    %2368 = torch.aten.mul.Tensor %2365, %2367 : !torch.vtensor<[2,256,5120],f16>, !torch.vtensor<[2,256,5120],f16> -> !torch.vtensor<[2,256,5120],f16>
    %2369 = torch.aten.transpose.int %459, %int0, %int1 : !torch.vtensor<[1280,5120],f16>, !torch.int, !torch.int -> !torch.vtensor<[5120,1280],f16>
    %2370 = torch.aten.view %2368, %2100 : !torch.vtensor<[2,256,5120],f16>, !torch.list<int> -> !torch.vtensor<[512,5120],f16>
    %2371 = torch.aten.mm %2370, %2369 : !torch.vtensor<[512,5120],f16>, !torch.vtensor<[5120,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2372 = torch.aten.mul.Scalar %458, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2373 = torch.aten.add.Tensor %2372, %2371, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[512,1280],f16>, !torch.int -> !torch.vtensor<[512,1280],f16>
    %2374 = torch.aten.view %2373, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2375 = torch.aten.add.Tensor %2374, %2342, %int1 : !torch.vtensor<[2,256,1280],f16>, !torch.vtensor<[2,256,1280],f16>, !torch.int -> !torch.vtensor<[2,256,1280],f16>
    %2376 = torch.aten.transpose.int %457, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2377 = torch.aten.view %2375, %1955 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[512,1280],f16>
    %2378 = torch.aten.mm %2377, %2376 : !torch.vtensor<[512,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[512,1280],f16>
    %2379 = torch.aten.mul.Scalar %456, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2380 = torch.aten.add.Tensor %2379, %2378, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[512,1280],f16>, !torch.int -> !torch.vtensor<[512,1280],f16>
    %2381 = torch.aten.view %2380, %1927 : !torch.vtensor<[512,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,256,1280],f16>
    %2382 = torch.aten.view %2381, %2113 : !torch.vtensor<[2,256,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,16,16,1280],f16>
    %2383 = torch.aten.permute %2382, %1003 : !torch.vtensor<[2,16,16,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,16,16],f16>
    %2384 = torch.aten.clone %2383, %int0 : !torch.vtensor<[2,1280,16,16],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2385 = torch.aten.add.Tensor %2384, %2182, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[2,1280,16,16],f16>, !torch.int -> !torch.vtensor<[2,1280,16,16],f16>
    %2386 = torch.aten.convolution %2385, %455, %454, %1275, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,1280,16,16],f16>, !torch.vtensor<[1280,1280,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2387 = torch.prim.ListConstruct %int2, %int32, %int40, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %2388 = torch.aten.view %2386, %2387 : !torch.vtensor<[2,1280,8,8],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,64],f16>
    %2389 = torch.aten.to.dtype %2388, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,64],f64>
    %2390 = torch.aten.sum.dim_IntList %2389, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2391 = torch.aten.div.Scalar %2390, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2392 = torch.aten.sub.Tensor %2389, %2391, %float1.000000e00 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,64],f64>
    %2393 = torch.aten.mul.Tensor %2392, %2392 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,40,64],f64> -> !torch.vtensor<[2,32,40,64],f64>
    %2394 = torch.aten.sum.dim_IntList %2393, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2395 = torch.aten.div.Scalar %2394, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2396 = torch.aten.to.dtype %2395, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2397 = torch.aten.sum.dim_IntList %2388, %721, %true, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2398 = torch.aten.div.Scalar %2397, %int2560 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2399 = torch.aten.add.Scalar %2396, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2400 = torch.aten.rsqrt %2399 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %2401 = torch.aten.sub.Tensor %2388, %2398, %int1 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,64],f16>
    %2402 = torch.aten.mul.Tensor %2401, %2400 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,64],f16>
    %2403 = torch.prim.ListConstruct %int2, %int1280, %int8, %int8 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %2404 = torch.aten.view %2402, %2403 : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,8,8],f16>
    %2405 = torch.aten.unsqueeze %453, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2406 = torch.aten.unsqueeze %2405, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2407 = torch.aten.unsqueeze %2406, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2408 = torch.aten.unsqueeze %452, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2409 = torch.aten.unsqueeze %2408, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2410 = torch.aten.unsqueeze %2409, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2411 = torch.aten.mul.Tensor %2404, %2410 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2412 = torch.aten.add.Tensor %2411, %2407, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2413 = torch.aten.sigmoid %2412 : !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2414 = torch.aten.mul.Tensor %2413, %2412 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2415 = torch.aten.convolution %2414, %451, %450, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1280,1280,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2416 = torch.aten.sigmoid %715 : !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2417 = torch.aten.mul.Tensor %2416, %715 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2418 = torch.aten.transpose.int %449, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2419 = torch.aten.mm %2417, %2418 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2420 = torch.aten.mul.Scalar %448, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2421 = torch.aten.add.Tensor %2420, %2419, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280],f16>
    %2422 = torch.aten.unsqueeze %2421, %int2 : !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280,1],f16>
    %2423 = torch.aten.unsqueeze %2422, %int3 : !torch.vtensor<[2,1280,1],f16>, !torch.int -> !torch.vtensor<[2,1280,1,1],f16>
    %2424 = torch.aten.add.Tensor %2415, %2423, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2425 = torch.aten.view %2424, %2387 : !torch.vtensor<[2,1280,8,8],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,64],f16>
    %2426 = torch.aten.to.dtype %2425, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,64],f64>
    %2427 = torch.aten.sum.dim_IntList %2426, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2428 = torch.aten.div.Scalar %2427, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2429 = torch.aten.sub.Tensor %2426, %2428, %float1.000000e00 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,64],f64>
    %2430 = torch.aten.mul.Tensor %2429, %2429 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,40,64],f64> -> !torch.vtensor<[2,32,40,64],f64>
    %2431 = torch.aten.sum.dim_IntList %2430, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2432 = torch.aten.div.Scalar %2431, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2433 = torch.aten.to.dtype %2432, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2434 = torch.aten.sum.dim_IntList %2425, %721, %true, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2435 = torch.aten.div.Scalar %2434, %int2560 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2436 = torch.aten.add.Scalar %2433, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2437 = torch.aten.rsqrt %2436 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %2438 = torch.aten.sub.Tensor %2425, %2435, %int1 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,64],f16>
    %2439 = torch.aten.mul.Tensor %2438, %2437 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,64],f16>
    %2440 = torch.aten.view %2439, %2403 : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,8,8],f16>
    %2441 = torch.aten.unsqueeze %447, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2442 = torch.aten.unsqueeze %2441, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2443 = torch.aten.unsqueeze %2442, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2444 = torch.aten.unsqueeze %446, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2445 = torch.aten.unsqueeze %2444, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2446 = torch.aten.unsqueeze %2445, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2447 = torch.aten.mul.Tensor %2440, %2446 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2448 = torch.aten.add.Tensor %2447, %2443, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2449 = torch.aten.sigmoid %2448 : !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2450 = torch.aten.mul.Tensor %2449, %2448 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2451 = torch.aten.convolution %2450, %445, %444, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1280,1280,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2452 = torch.aten.add.Tensor %2386, %2451, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,8,8],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2453 = torch.aten.div.Scalar %2452, %float1.000000e00 : !torch.vtensor<[2,1280,8,8],f16>, !torch.float -> !torch.vtensor<[2,1280,8,8],f16>
    %2454 = torch.aten.view %2453, %2387 : !torch.vtensor<[2,1280,8,8],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,64],f16>
    %2455 = torch.aten.to.dtype %2454, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,64],f64>
    %2456 = torch.aten.sum.dim_IntList %2455, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2457 = torch.aten.div.Scalar %2456, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2458 = torch.aten.sub.Tensor %2455, %2457, %float1.000000e00 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,64],f64>
    %2459 = torch.aten.mul.Tensor %2458, %2458 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,40,64],f64> -> !torch.vtensor<[2,32,40,64],f64>
    %2460 = torch.aten.sum.dim_IntList %2459, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2461 = torch.aten.div.Scalar %2460, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2462 = torch.aten.to.dtype %2461, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2463 = torch.aten.sum.dim_IntList %2454, %721, %true, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2464 = torch.aten.div.Scalar %2463, %int2560 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2465 = torch.aten.add.Scalar %2462, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2466 = torch.aten.rsqrt %2465 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %2467 = torch.aten.sub.Tensor %2454, %2464, %int1 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,64],f16>
    %2468 = torch.aten.mul.Tensor %2467, %2466 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,64],f16>
    %2469 = torch.aten.view %2468, %2403 : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,8,8],f16>
    %2470 = torch.aten.unsqueeze %443, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2471 = torch.aten.unsqueeze %2470, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2472 = torch.aten.unsqueeze %2471, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2473 = torch.aten.unsqueeze %442, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2474 = torch.aten.unsqueeze %2473, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2475 = torch.aten.unsqueeze %2474, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2476 = torch.aten.mul.Tensor %2469, %2475 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2477 = torch.aten.add.Tensor %2476, %2472, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2478 = torch.aten.sigmoid %2477 : !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2479 = torch.aten.mul.Tensor %2478, %2477 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2480 = torch.aten.convolution %2479, %441, %440, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1280,1280,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2481 = torch.aten.sigmoid %715 : !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2482 = torch.aten.mul.Tensor %2481, %715 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2483 = torch.aten.transpose.int %439, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2484 = torch.aten.mm %2482, %2483 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2485 = torch.aten.mul.Scalar %438, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2486 = torch.aten.add.Tensor %2485, %2484, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280],f16>
    %2487 = torch.aten.unsqueeze %2486, %int2 : !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280,1],f16>
    %2488 = torch.aten.unsqueeze %2487, %int3 : !torch.vtensor<[2,1280,1],f16>, !torch.int -> !torch.vtensor<[2,1280,1,1],f16>
    %2489 = torch.aten.add.Tensor %2480, %2488, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2490 = torch.aten.view %2489, %2387 : !torch.vtensor<[2,1280,8,8],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,64],f16>
    %2491 = torch.aten.to.dtype %2490, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,64],f64>
    %2492 = torch.aten.sum.dim_IntList %2491, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2493 = torch.aten.div.Scalar %2492, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2494 = torch.aten.sub.Tensor %2491, %2493, %float1.000000e00 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,64],f64>
    %2495 = torch.aten.mul.Tensor %2494, %2494 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,40,64],f64> -> !torch.vtensor<[2,32,40,64],f64>
    %2496 = torch.aten.sum.dim_IntList %2495, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2497 = torch.aten.div.Scalar %2496, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2498 = torch.aten.to.dtype %2497, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2499 = torch.aten.sum.dim_IntList %2490, %721, %true, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2500 = torch.aten.div.Scalar %2499, %int2560 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2501 = torch.aten.add.Scalar %2498, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2502 = torch.aten.rsqrt %2501 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %2503 = torch.aten.sub.Tensor %2490, %2500, %int1 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,64],f16>
    %2504 = torch.aten.mul.Tensor %2503, %2502 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,64],f16>
    %2505 = torch.aten.view %2504, %2403 : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,8,8],f16>
    %2506 = torch.aten.unsqueeze %437, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2507 = torch.aten.unsqueeze %2506, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2508 = torch.aten.unsqueeze %2507, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2509 = torch.aten.unsqueeze %436, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2510 = torch.aten.unsqueeze %2509, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2511 = torch.aten.unsqueeze %2510, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2512 = torch.aten.mul.Tensor %2505, %2511 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2513 = torch.aten.add.Tensor %2512, %2508, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2514 = torch.aten.sigmoid %2513 : !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2515 = torch.aten.mul.Tensor %2514, %2513 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2516 = torch.aten.convolution %2515, %435, %434, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1280,1280,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2517 = torch.aten.add.Tensor %2453, %2516, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,8,8],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2518 = torch.aten.div.Scalar %2517, %float1.000000e00 : !torch.vtensor<[2,1280,8,8],f16>, !torch.float -> !torch.vtensor<[2,1280,8,8],f16>
    %2519 = torch.aten.view %2518, %2387 : !torch.vtensor<[2,1280,8,8],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,64],f16>
    %2520 = torch.aten.to.dtype %2519, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,64],f64>
    %2521 = torch.aten.sum.dim_IntList %2520, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2522 = torch.aten.div.Scalar %2521, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2523 = torch.aten.sub.Tensor %2520, %2522, %float1.000000e00 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,64],f64>
    %2524 = torch.aten.mul.Tensor %2523, %2523 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,40,64],f64> -> !torch.vtensor<[2,32,40,64],f64>
    %2525 = torch.aten.sum.dim_IntList %2524, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2526 = torch.aten.div.Scalar %2525, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2527 = torch.aten.to.dtype %2526, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2528 = torch.aten.sum.dim_IntList %2519, %721, %true, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2529 = torch.aten.div.Scalar %2528, %int2560 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2530 = torch.aten.add.Scalar %2527, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2531 = torch.aten.rsqrt %2530 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %2532 = torch.aten.sub.Tensor %2519, %2529, %int1 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,64],f16>
    %2533 = torch.aten.mul.Tensor %2532, %2531 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,64],f16>
    %2534 = torch.aten.view %2533, %2403 : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,8,8],f16>
    %2535 = torch.aten.unsqueeze %433, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2536 = torch.aten.unsqueeze %2535, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2537 = torch.aten.unsqueeze %2536, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2538 = torch.aten.unsqueeze %432, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2539 = torch.aten.unsqueeze %2538, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2540 = torch.aten.unsqueeze %2539, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2541 = torch.aten.mul.Tensor %2534, %2540 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2542 = torch.aten.add.Tensor %2541, %2537, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2543 = torch.aten.sigmoid %2542 : !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2544 = torch.aten.mul.Tensor %2543, %2542 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2545 = torch.aten.convolution %2544, %431, %430, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1280,1280,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2546 = torch.aten.sigmoid %715 : !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2547 = torch.aten.mul.Tensor %2546, %715 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[2,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2548 = torch.aten.transpose.int %429, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2549 = torch.aten.mm %2547, %2548 : !torch.vtensor<[2,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[2,1280],f16>
    %2550 = torch.aten.mul.Scalar %428, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2551 = torch.aten.add.Tensor %2550, %2549, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280],f16>
    %2552 = torch.aten.unsqueeze %2551, %int2 : !torch.vtensor<[2,1280],f16>, !torch.int -> !torch.vtensor<[2,1280,1],f16>
    %2553 = torch.aten.unsqueeze %2552, %int3 : !torch.vtensor<[2,1280,1],f16>, !torch.int -> !torch.vtensor<[2,1280,1,1],f16>
    %2554 = torch.aten.add.Tensor %2545, %2553, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2555 = torch.aten.view %2554, %2387 : !torch.vtensor<[2,1280,8,8],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,64],f16>
    %2556 = torch.aten.to.dtype %2555, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,64],f64>
    %2557 = torch.aten.sum.dim_IntList %2556, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2558 = torch.aten.div.Scalar %2557, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2559 = torch.aten.sub.Tensor %2556, %2558, %float1.000000e00 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,64],f64>
    %2560 = torch.aten.mul.Tensor %2559, %2559 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,40,64],f64> -> !torch.vtensor<[2,32,40,64],f64>
    %2561 = torch.aten.sum.dim_IntList %2560, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2562 = torch.aten.div.Scalar %2561, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2563 = torch.aten.to.dtype %2562, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2564 = torch.aten.sum.dim_IntList %2555, %721, %true, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2565 = torch.aten.div.Scalar %2564, %int2560 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2566 = torch.aten.add.Scalar %2563, %float1.000000e-05, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2567 = torch.aten.rsqrt %2566 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %2568 = torch.aten.sub.Tensor %2555, %2565, %int1 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,64],f16>
    %2569 = torch.aten.mul.Tensor %2568, %2567 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,64],f16>
    %2570 = torch.aten.view %2569, %2403 : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,8,8],f16>
    %2571 = torch.aten.unsqueeze %427, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2572 = torch.aten.unsqueeze %2571, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2573 = torch.aten.unsqueeze %2572, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2574 = torch.aten.unsqueeze %426, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2575 = torch.aten.unsqueeze %2574, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2576 = torch.aten.unsqueeze %2575, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2577 = torch.aten.mul.Tensor %2570, %2576 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2578 = torch.aten.add.Tensor %2577, %2573, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2579 = torch.aten.sigmoid %2578 : !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2580 = torch.aten.mul.Tensor %2579, %2578 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,8,8],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2581 = torch.aten.convolution %2580, %425, %424, %716, %716, %716, %false, %717, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1280,1280,3,3],f16>, !torch.vtensor<[1280],f16>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2582 = torch.aten.add.Tensor %2518, %2581, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,8,8],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2583 = torch.aten.div.Scalar %2582, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2584 = torch.aten.view %2583, %2387 : !torch.vtensor<[2,1280,8,8],f16>, !torch.list<int> -> !torch.vtensor<[2,32,40,64],f16>
    %2585 = torch.aten.to.dtype %2584, %int7, %false, %false, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,40,64],f64>
    %2586 = torch.aten.sum.dim_IntList %2585, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2587 = torch.aten.div.Scalar %2586, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2588 = torch.aten.sub.Tensor %2585, %2587, %float1.000000e00 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,1,1],f64>, !torch.float -> !torch.vtensor<[2,32,40,64],f64>
    %2589 = torch.aten.mul.Tensor %2588, %2588 : !torch.vtensor<[2,32,40,64],f64>, !torch.vtensor<[2,32,40,64],f64> -> !torch.vtensor<[2,32,40,64],f64>
    %2590 = torch.aten.sum.dim_IntList %2589, %721, %true, %none : !torch.vtensor<[2,32,40,64],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f64>
    %2591 = torch.aten.div.Scalar %2590, %int2560 : !torch.vtensor<[2,32,1,1],f64>, !torch.int -> !torch.vtensor<[2,32,1,1],f64>
    %2592 = torch.aten.to.dtype %2591, %int5, %false, %false, %none : !torch.vtensor<[2,32,1,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2593 = torch.aten.sum.dim_IntList %2584, %721, %true, %none : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,32,1,1],f16>
    %2594 = torch.aten.div.Scalar %2593, %int2560 : !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2595 = torch.aten.add.Scalar %2592, %float9.999990e-07, %int1 : !torch.vtensor<[2,32,1,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,32,1,1],f16>
    %2596 = torch.aten.rsqrt %2595 : !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,1,1],f16>
    %2597 = torch.aten.sub.Tensor %2584, %2594, %int1 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16>, !torch.int -> !torch.vtensor<[2,32,40,64],f16>
    %2598 = torch.aten.mul.Tensor %2597, %2596 : !torch.vtensor<[2,32,40,64],f16>, !torch.vtensor<[2,32,1,1],f16> -> !torch.vtensor<[2,32,40,64],f16>
    %2599 = torch.aten.view %2598, %2403 : !torch.vtensor<[2,32,40,64],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,8,8],f16>
    %2600 = torch.aten.unsqueeze %423, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2601 = torch.aten.unsqueeze %2600, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2602 = torch.aten.unsqueeze %2601, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2603 = torch.aten.unsqueeze %422, %int0 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1,1280],f16>
    %2604 = torch.aten.unsqueeze %2603, %int2 : !torch.vtensor<[1,1280],f16>, !torch.int -> !torch.vtensor<[1,1280,1],f16>
    %2605 = torch.aten.unsqueeze %2604, %int3 : !torch.vtensor<[1,1280,1],f16>, !torch.int -> !torch.vtensor<[1,1280,1,1],f16>
    %2606 = torch.aten.mul.Tensor %2599, %2605 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16> -> !torch.vtensor<[2,1280,8,8],f16>
    %2607 = torch.aten.add.Tensor %2606, %2602, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[1,1280,1,1],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2608 = torch.aten.permute %2607, %811 : !torch.vtensor<[2,1280,8,8],f16>, !torch.list<int> -> !torch.vtensor<[2,8,8,1280],f16>
    %2609 = torch.aten.view %2608, %2033 : !torch.vtensor<[2,8,8,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2610 = torch.aten.transpose.int %421, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2611 = torch.aten.broadcast_to %2609, %2033 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2612 = torch.aten.view %2611, %2033 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2613 = torch.aten.broadcast_to %2610, %1932 : !torch.vtensor<[1280,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,1280],f16>
    %2614 = torch.aten.view %2613, %1932 : !torch.vtensor<[2,1280,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,1280],f16>
    %2615 = torch.aten.bmm %2612, %2614 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[2,1280,1280],f16> -> !torch.vtensor<[2,64,1280],f16>
    %2616 = torch.aten.view %2615, %2033 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2617 = torch.aten.add.Tensor %2616, %420, %int1 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,64,1280],f16>
    %2618 = torch.aten.to.dtype %2617, %int7, %false, %false, %none : !torch.vtensor<[2,64,1280],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1280],f64>
    %2619 = torch.aten.sum.dim_IntList %2618, %688, %true, %none : !torch.vtensor<[2,64,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f64>
    %2620 = torch.aten.div.Scalar %2619, %int1280 : !torch.vtensor<[2,64,1],f64>, !torch.int -> !torch.vtensor<[2,64,1],f64>
    %2621 = torch.aten.sub.Tensor %2618, %2620, %float1.000000e00 : !torch.vtensor<[2,64,1280],f64>, !torch.vtensor<[2,64,1],f64>, !torch.float -> !torch.vtensor<[2,64,1280],f64>
    %2622 = torch.aten.mul.Tensor %2621, %2621 : !torch.vtensor<[2,64,1280],f64>, !torch.vtensor<[2,64,1280],f64> -> !torch.vtensor<[2,64,1280],f64>
    %2623 = torch.aten.sum.dim_IntList %2622, %688, %true, %none : !torch.vtensor<[2,64,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f64>
    %2624 = torch.aten.div.Scalar %2623, %int1280 : !torch.vtensor<[2,64,1],f64>, !torch.int -> !torch.vtensor<[2,64,1],f64>
    %2625 = torch.aten.to.dtype %2624, %int5, %false, %false, %none : !torch.vtensor<[2,64,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f16>
    %2626 = torch.aten.sum.dim_IntList %2617, %688, %true, %none : !torch.vtensor<[2,64,1280],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f16>
    %2627 = torch.aten.div.Scalar %2626, %int1280 : !torch.vtensor<[2,64,1],f16>, !torch.int -> !torch.vtensor<[2,64,1],f16>
    %2628 = torch.aten.add.Scalar %2625, %float1.000000e-05, %int1 : !torch.vtensor<[2,64,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,64,1],f16>
    %2629 = torch.aten.rsqrt %2628 : !torch.vtensor<[2,64,1],f16> -> !torch.vtensor<[2,64,1],f16>
    %2630 = torch.aten.sub.Tensor %2617, %2627, %int1 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[2,64,1],f16>, !torch.int -> !torch.vtensor<[2,64,1280],f16>
    %2631 = torch.aten.mul.Tensor %2630, %2629 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[2,64,1],f16> -> !torch.vtensor<[2,64,1280],f16>
    %2632 = torch.aten.mul.Tensor %2631, %419 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[1280],f16> -> !torch.vtensor<[2,64,1280],f16>
    %2633 = torch.aten.add.Tensor %2632, %418, %int1 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,64,1280],f16>
    %2634 = torch.aten.transpose.int %417, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2635 = torch.prim.ListConstruct %int128, %int1280 : (!torch.int, !torch.int) -> !torch.list<int>
    %2636 = torch.aten.view %2633, %2635 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[128,1280],f16>
    %2637 = torch.aten.mm %2636, %2634 : !torch.vtensor<[128,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2638 = torch.aten.view %2637, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2639 = torch.aten.view %2638, %2039 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2640 = torch.aten.permute %2639, %847 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2641 = torch.aten.clone %2640, %int0 : !torch.vtensor<[2,20,64,64],f16>, !torch.int -> !torch.vtensor<[2,20,64,64],f16>
    %2642 = torch.aten.view %2641, %2043 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[40,64,64],f16>
    %2643 = torch.aten.transpose.int %416, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2644 = torch.aten.view %2633, %2635 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[128,1280],f16>
    %2645 = torch.aten.mm %2644, %2643 : !torch.vtensor<[128,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2646 = torch.aten.view %2645, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2647 = torch.aten.transpose.int %415, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2648 = torch.aten.view %2633, %2635 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[128,1280],f16>
    %2649 = torch.aten.mm %2648, %2647 : !torch.vtensor<[128,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2650 = torch.aten.view %2649, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2651 = torch.aten.view %2646, %2039 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2652 = torch.aten.permute %2651, %847 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2653 = torch.aten.clone %2652, %int0 : !torch.vtensor<[2,20,64,64],f16>, !torch.int -> !torch.vtensor<[2,20,64,64],f16>
    %2654 = torch.aten.view %2653, %2043 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[40,64,64],f16>
    %2655 = torch.aten.view %2650, %2039 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2656 = torch.aten.permute %2655, %847 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2657 = torch.aten.clone %2656, %int0 : !torch.vtensor<[2,20,64,64],f16>, !torch.int -> !torch.vtensor<[2,20,64,64],f16>
    %2658 = torch.aten.view %2657, %2043 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[40,64,64],f16>
    %2659 = torch.aten.empty.memory_format %2043, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[40,64,64],f32>
    %2660 = torch.aten.transpose.int %2654, %int-1, %int-2 : !torch.vtensor<[40,64,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[40,64,64],f16>
    %2661 = torch.aten.bmm %2642, %2660 : !torch.vtensor<[40,64,64],f16>, !torch.vtensor<[40,64,64],f16> -> !torch.vtensor<[40,64,64],f16>
    %2662 = torch.aten.mul.Scalar %2661, %float1.250000e-01 : !torch.vtensor<[40,64,64],f16>, !torch.float -> !torch.vtensor<[40,64,64],f16>
    %2663 = torch.aten.to.dtype %2659, %int5, %false, %false, %none : !torch.vtensor<[40,64,64],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[40,64,64],f16>
    %2664 = torch.aten.add.Tensor %2662, %2659, %int0 : !torch.vtensor<[40,64,64],f16>, !torch.vtensor<[40,64,64],f32>, !torch.int -> !torch.vtensor<[40,64,64],f16>
    %values_22, %indices_23 = torch.aten.max.dim %2664, %int-1, %true : !torch.vtensor<[40,64,64],f16>, !torch.int, !torch.bool -> !torch.vtensor<[40,64,1],f16>, !torch.vtensor<[40,64,1],si64>
    %2665 = torch.aten.sub.Tensor %2664, %values_22, %float1.000000e00 : !torch.vtensor<[40,64,64],f16>, !torch.vtensor<[40,64,1],f16>, !torch.float -> !torch.vtensor<[40,64,64],f16>
    %2666 = torch.aten.exp %2665 : !torch.vtensor<[40,64,64],f16> -> !torch.vtensor<[40,64,64],f16>
    %2667 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %2668 = torch.aten.sum.dim_IntList %2666, %2667, %true, %none : !torch.vtensor<[40,64,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[40,64,1],f16>
    %2669 = torch.aten.div.Tensor %2666, %2668 : !torch.vtensor<[40,64,64],f16>, !torch.vtensor<[40,64,1],f16> -> !torch.vtensor<[40,64,64],f16>
    %2670 = torch.aten.bmm %2669, %2658 : !torch.vtensor<[40,64,64],f16>, !torch.vtensor<[40,64,64],f16> -> !torch.vtensor<[40,64,64],f16>
    %2671 = torch.prim.ListConstruct %int2, %int20, %int64, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %2672 = torch.aten.view %2670, %2671 : !torch.vtensor<[40,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2673 = torch.aten.permute %2672, %847 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2674 = torch.aten.clone %2673, %int0 : !torch.vtensor<[2,64,20,64],f16>, !torch.int -> !torch.vtensor<[2,64,20,64],f16>
    %2675 = torch.aten.view %2674, %2033 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2676 = torch.aten.transpose.int %414, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2677 = torch.aten.view %2675, %2635 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[128,1280],f16>
    %2678 = torch.aten.mm %2677, %2676 : !torch.vtensor<[128,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2679 = torch.aten.mul.Scalar %413, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2680 = torch.aten.add.Tensor %2679, %2678, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[128,1280],f16>, !torch.int -> !torch.vtensor<[128,1280],f16>
    %2681 = torch.aten.view %2680, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2682 = torch.aten.add.Tensor %2681, %2617, %int1 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[2,64,1280],f16>, !torch.int -> !torch.vtensor<[2,64,1280],f16>
    %2683 = torch.aten.to.dtype %2682, %int7, %false, %false, %none : !torch.vtensor<[2,64,1280],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1280],f64>
    %2684 = torch.aten.sum.dim_IntList %2683, %688, %true, %none : !torch.vtensor<[2,64,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f64>
    %2685 = torch.aten.div.Scalar %2684, %int1280 : !torch.vtensor<[2,64,1],f64>, !torch.int -> !torch.vtensor<[2,64,1],f64>
    %2686 = torch.aten.sub.Tensor %2683, %2685, %float1.000000e00 : !torch.vtensor<[2,64,1280],f64>, !torch.vtensor<[2,64,1],f64>, !torch.float -> !torch.vtensor<[2,64,1280],f64>
    %2687 = torch.aten.mul.Tensor %2686, %2686 : !torch.vtensor<[2,64,1280],f64>, !torch.vtensor<[2,64,1280],f64> -> !torch.vtensor<[2,64,1280],f64>
    %2688 = torch.aten.sum.dim_IntList %2687, %688, %true, %none : !torch.vtensor<[2,64,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f64>
    %2689 = torch.aten.div.Scalar %2688, %int1280 : !torch.vtensor<[2,64,1],f64>, !torch.int -> !torch.vtensor<[2,64,1],f64>
    %2690 = torch.aten.to.dtype %2689, %int5, %false, %false, %none : !torch.vtensor<[2,64,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f16>
    %2691 = torch.aten.sum.dim_IntList %2682, %688, %true, %none : !torch.vtensor<[2,64,1280],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f16>
    %2692 = torch.aten.div.Scalar %2691, %int1280 : !torch.vtensor<[2,64,1],f16>, !torch.int -> !torch.vtensor<[2,64,1],f16>
    %2693 = torch.aten.add.Scalar %2690, %float1.000000e-05, %int1 : !torch.vtensor<[2,64,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,64,1],f16>
    %2694 = torch.aten.rsqrt %2693 : !torch.vtensor<[2,64,1],f16> -> !torch.vtensor<[2,64,1],f16>
    %2695 = torch.aten.sub.Tensor %2682, %2692, %int1 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[2,64,1],f16>, !torch.int -> !torch.vtensor<[2,64,1280],f16>
    %2696 = torch.aten.mul.Tensor %2695, %2694 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[2,64,1],f16> -> !torch.vtensor<[2,64,1280],f16>
    %2697 = torch.aten.mul.Tensor %2696, %412 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[1280],f16> -> !torch.vtensor<[2,64,1280],f16>
    %2698 = torch.aten.add.Tensor %2697, %411, %int1 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,64,1280],f16>
    %2699 = torch.aten.transpose.int %410, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2700 = torch.aten.view %2698, %2635 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[128,1280],f16>
    %2701 = torch.aten.mm %2700, %2699 : !torch.vtensor<[128,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2702 = torch.aten.view %2701, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2703 = torch.aten.view %2702, %2039 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2704 = torch.aten.permute %2703, %847 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2705 = torch.aten.clone %2704, %int0 : !torch.vtensor<[2,20,64,64],f16>, !torch.int -> !torch.vtensor<[2,20,64,64],f16>
    %2706 = torch.aten.view %2705, %2043 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[40,64,64],f16>
    %2707 = torch.aten.transpose.int %409, %int0, %int1 : !torch.vtensor<[1280,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,1280],f16>
    %2708 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %2709 = torch.aten.mm %2708, %2707 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2710 = torch.aten.view %2709, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2711 = torch.aten.transpose.int %408, %int0, %int1 : !torch.vtensor<[1280,1024],f16>, !torch.int, !torch.int -> !torch.vtensor<[1024,1280],f16>
    %2712 = torch.aten.view %arg2, %918 : !torch.vtensor<[2,64,1024],f16>, !torch.list<int> -> !torch.vtensor<[128,1024],f16>
    %2713 = torch.aten.mm %2712, %2711 : !torch.vtensor<[128,1024],f16>, !torch.vtensor<[1024,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2714 = torch.aten.view %2713, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2715 = torch.aten.view %2710, %2039 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2716 = torch.aten.permute %2715, %847 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2717 = torch.aten.clone %2716, %int0 : !torch.vtensor<[2,20,64,64],f16>, !torch.int -> !torch.vtensor<[2,20,64,64],f16>
    %2718 = torch.aten.view %2717, %2043 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[40,64,64],f16>
    %2719 = torch.aten.view %2714, %2039 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2720 = torch.aten.permute %2719, %847 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2721 = torch.aten.clone %2720, %int0 : !torch.vtensor<[2,20,64,64],f16>, !torch.int -> !torch.vtensor<[2,20,64,64],f16>
    %2722 = torch.aten.view %2721, %2043 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[40,64,64],f16>
    %2723 = torch.aten.empty.memory_format %2043, %int6, %none, %cpu, %false, %none : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool, !torch.none -> !torch.vtensor<[40,64,64],f32>
    %2724 = torch.aten.transpose.int %2718, %int-1, %int-2 : !torch.vtensor<[40,64,64],f16>, !torch.int, !torch.int -> !torch.vtensor<[40,64,64],f16>
    %2725 = torch.aten.bmm %2706, %2724 : !torch.vtensor<[40,64,64],f16>, !torch.vtensor<[40,64,64],f16> -> !torch.vtensor<[40,64,64],f16>
    %2726 = torch.aten.mul.Scalar %2725, %float1.250000e-01 : !torch.vtensor<[40,64,64],f16>, !torch.float -> !torch.vtensor<[40,64,64],f16>
    %2727 = torch.aten.to.dtype %2723, %int5, %false, %false, %none : !torch.vtensor<[40,64,64],f32>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[40,64,64],f16>
    %2728 = torch.aten.add.Tensor %2726, %2723, %int0 : !torch.vtensor<[40,64,64],f16>, !torch.vtensor<[40,64,64],f32>, !torch.int -> !torch.vtensor<[40,64,64],f16>
    %values_24, %indices_25 = torch.aten.max.dim %2728, %int-1, %true : !torch.vtensor<[40,64,64],f16>, !torch.int, !torch.bool -> !torch.vtensor<[40,64,1],f16>, !torch.vtensor<[40,64,1],si64>
    %2729 = torch.aten.sub.Tensor %2728, %values_24, %float1.000000e00 : !torch.vtensor<[40,64,64],f16>, !torch.vtensor<[40,64,1],f16>, !torch.float -> !torch.vtensor<[40,64,64],f16>
    %2730 = torch.aten.exp %2729 : !torch.vtensor<[40,64,64],f16> -> !torch.vtensor<[40,64,64],f16>
    %2731 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
    %2732 = torch.aten.sum.dim_IntList %2730, %2731, %true, %none : !torch.vtensor<[40,64,64],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[40,64,1],f16>
    %2733 = torch.aten.div.Tensor %2730, %2732 : !torch.vtensor<[40,64,64],f16>, !torch.vtensor<[40,64,1],f16> -> !torch.vtensor<[40,64,64],f16>
    %2734 = torch.aten.bmm %2733, %2722 : !torch.vtensor<[40,64,64],f16>, !torch.vtensor<[40,64,64],f16> -> !torch.vtensor<[40,64,64],f16>
    %2735 = torch.aten.view %2734, %2671 : !torch.vtensor<[40,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,20,64,64],f16>
    %2736 = torch.aten.permute %2735, %847 : !torch.vtensor<[2,20,64,64],f16>, !torch.list<int> -> !torch.vtensor<[2,64,20,64],f16>
    %2737 = torch.aten.clone %2736, %int0 : !torch.vtensor<[2,64,20,64],f16>, !torch.int -> !torch.vtensor<[2,64,20,64],f16>
    %2738 = torch.aten.view %2737, %2033 : !torch.vtensor<[2,64,20,64],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2739 = torch.aten.transpose.int %407, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2740 = torch.aten.view %2738, %2635 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[128,1280],f16>
    %2741 = torch.aten.mm %2740, %2739 : !torch.vtensor<[128,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2742 = torch.aten.mul.Scalar %406, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2743 = torch.aten.add.Tensor %2742, %2741, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[128,1280],f16>, !torch.int -> !torch.vtensor<[128,1280],f16>
    %2744 = torch.aten.view %2743, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2745 = torch.aten.add.Tensor %2744, %2682, %int1 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[2,64,1280],f16>, !torch.int -> !torch.vtensor<[2,64,1280],f16>
    %2746 = torch.aten.to.dtype %2745, %int7, %false, %false, %none : !torch.vtensor<[2,64,1280],f16>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1280],f64>
    %2747 = torch.aten.sum.dim_IntList %2746, %688, %true, %none : !torch.vtensor<[2,64,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f64>
    %2748 = torch.aten.div.Scalar %2747, %int1280 : !torch.vtensor<[2,64,1],f64>, !torch.int -> !torch.vtensor<[2,64,1],f64>
    %2749 = torch.aten.sub.Tensor %2746, %2748, %float1.000000e00 : !torch.vtensor<[2,64,1280],f64>, !torch.vtensor<[2,64,1],f64>, !torch.float -> !torch.vtensor<[2,64,1280],f64>
    %2750 = torch.aten.mul.Tensor %2749, %2749 : !torch.vtensor<[2,64,1280],f64>, !torch.vtensor<[2,64,1280],f64> -> !torch.vtensor<[2,64,1280],f64>
    %2751 = torch.aten.sum.dim_IntList %2750, %688, %true, %none : !torch.vtensor<[2,64,1280],f64>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f64>
    %2752 = torch.aten.div.Scalar %2751, %int1280 : !torch.vtensor<[2,64,1],f64>, !torch.int -> !torch.vtensor<[2,64,1],f64>
    %2753 = torch.aten.to.dtype %2752, %int5, %false, %false, %none : !torch.vtensor<[2,64,1],f64>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f16>
    %2754 = torch.aten.sum.dim_IntList %2745, %688, %true, %none : !torch.vtensor<[2,64,1280],f16>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[2,64,1],f16>
    %2755 = torch.aten.div.Scalar %2754, %int1280 : !torch.vtensor<[2,64,1],f16>, !torch.int -> !torch.vtensor<[2,64,1],f16>
    %2756 = torch.aten.add.Scalar %2753, %float1.000000e-05, %int1 : !torch.vtensor<[2,64,1],f16>, !torch.float, !torch.int -> !torch.vtensor<[2,64,1],f16>
    %2757 = torch.aten.rsqrt %2756 : !torch.vtensor<[2,64,1],f16> -> !torch.vtensor<[2,64,1],f16>
    %2758 = torch.aten.sub.Tensor %2745, %2755, %int1 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[2,64,1],f16>, !torch.int -> !torch.vtensor<[2,64,1280],f16>
    %2759 = torch.aten.mul.Tensor %2758, %2757 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[2,64,1],f16> -> !torch.vtensor<[2,64,1280],f16>
    %2760 = torch.aten.mul.Tensor %2759, %405 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[1280],f16> -> !torch.vtensor<[2,64,1280],f16>
    %2761 = torch.aten.add.Tensor %2760, %404, %int1 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[2,64,1280],f16>
    %2762 = torch.aten.transpose.int %403, %int0, %int1 : !torch.vtensor<[10240,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,10240],f16>
    %2763 = torch.aten.view %2761, %2635 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[128,1280],f16>
    %2764 = torch.aten.mm %2763, %2762 : !torch.vtensor<[128,1280],f16>, !torch.vtensor<[1280,10240],f16> -> !torch.vtensor<[128,10240],f16>
    %2765 = torch.aten.mul.Scalar %402, %int1 : !torch.vtensor<[10240],f16>, !torch.int -> !torch.vtensor<[10240],f16>
    %2766 = torch.aten.add.Tensor %2765, %2764, %int1 : !torch.vtensor<[10240],f16>, !torch.vtensor<[128,10240],f16>, !torch.int -> !torch.vtensor<[128,10240],f16>
    %2767 = torch.prim.ListConstruct %int2, %int64, %int10240 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %2768 = torch.aten.view %2766, %2767 : !torch.vtensor<[128,10240],f16>, !torch.list<int> -> !torch.vtensor<[2,64,10240],f16>
    %2769 = torch.aten.slice.Tensor %2768, %int-1, %int0, %int5120, %int1 : !torch.vtensor<[2,64,10240],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,64,5120],f16>
    %2770 = torch.aten.slice.Tensor %2768, %int-1, %int5120, %int10240, %int1 : !torch.vtensor<[2,64,10240],f16>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[2,64,5120],f16>
    %2771 = torch.aten.gelu %2770, %str : !torch.vtensor<[2,64,5120],f16>, !torch.str -> !torch.vtensor<[2,64,5120],f16>
    %2772 = torch.aten.mul.Tensor %2769, %2771 : !torch.vtensor<[2,64,5120],f16>, !torch.vtensor<[2,64,5120],f16> -> !torch.vtensor<[2,64,5120],f16>
    %2773 = torch.aten.transpose.int %401, %int0, %int1 : !torch.vtensor<[1280,5120],f16>, !torch.int, !torch.int -> !torch.vtensor<[5120,1280],f16>
    %2774 = torch.prim.ListConstruct %int128, %int5120 : (!torch.int, !torch.int) -> !torch.list<int>
    %2775 = torch.aten.view %2772, %2774 : !torch.vtensor<[2,64,5120],f16>, !torch.list<int> -> !torch.vtensor<[128,5120],f16>
    %2776 = torch.aten.mm %2775, %2773 : !torch.vtensor<[128,5120],f16>, !torch.vtensor<[5120,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2777 = torch.aten.mul.Scalar %400, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2778 = torch.aten.add.Tensor %2777, %2776, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[128,1280],f16>, !torch.int -> !torch.vtensor<[128,1280],f16>
    %2779 = torch.aten.view %2778, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2780 = torch.aten.add.Tensor %2779, %2745, %int1 : !torch.vtensor<[2,64,1280],f16>, !torch.vtensor<[2,64,1280],f16>, !torch.int -> !torch.vtensor<[2,64,1280],f16>
    %2781 = torch.aten.transpose.int %399, %int0, %int1 : !torch.vtensor<[1280,1280],f16>, !torch.int, !torch.int -> !torch.vtensor<[1280,1280],f16>
    %2782 = torch.aten.view %2780, %2635 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[128,1280],f16>
    %2783 = torch.aten.mm %2782, %2781 : !torch.vtensor<[128,1280],f16>, !torch.vtensor<[1280,1280],f16> -> !torch.vtensor<[128,1280],f16>
    %2784 = torch.aten.mul.Scalar %398, %int1 : !torch.vtensor<[1280],f16>, !torch.int -> !torch.vtensor<[1280],f16>
    %2785 = torch.aten.add.Tensor %2784, %2783, %int1 : !torch.vtensor<[1280],f16>, !torch.vtensor<[128,1280],f16>, !torch.int -> !torch.vtensor<[128,1280],f16>
    %2786 = torch.aten.view %2785, %2033 : !torch.vtensor<[128,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,64,1280],f16>
    %2787 = torch.prim.ListConstruct %int2, %int8, %int8, %int1280 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %2788 = torch.aten.view %2786, %2787 : !torch.vtensor<[2,64,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,8,8,1280],f16>
    %2789 = torch.aten.permute %2788, %1003 : !torch.vtensor<[2,8,8,1280],f16>, !torch.list<int> -> !torch.vtensor<[2,1280,8,8],f16>
    %2790 = torch.aten.clone %2789, %int0 : !torch.vtensor<[2,1280,8,8],f16>, !torch.int -> !torch.vtensor<[2,1280,8,8],f16>
    %2791 = torch.aten.add.Tensor %2790, %2583, %int1 : !torch.vtensor<[2,1280,8,8],f16>, !torch.vtensor<[2,1280,8,8],f16>, !torch.int -> !torch.vtenso