distilgpt2_torchtotosa_debug.mlir

func.func @forward(%arg0: !torch.vtensor<[1,128],si64>) -> !torch.vtensor<[1,2],f32> {
  %0 = builtin.unrealized_conversion_cast %arg0 : !torch.vtensor<[1,128],si64> to tensor<1x128xi64>
  %int1 = torch.constant.int 1
  %1 = builtin.unrealized_conversion_cast %int1 : !torch.int to i64
  %int0 = torch.constant.int 0
  %2 = builtin.unrealized_conversion_cast %int0 : !torch.int to i64
  %int-1 = torch.constant.int -1
  %3 = builtin.unrealized_conversion_cast %int-1 : !torch.int to i64
  %true = torch.constant.bool true
  %4 = builtin.unrealized_conversion_cast %true : !torch.bool to i1
  %none = torch.constant.none
  %false = torch.constant.bool false
  %5 = builtin.unrealized_conversion_cast %false : !torch.bool to i1
  %int128 = torch.constant.int 128
  %6 = builtin.unrealized_conversion_cast %int128 : !torch.int to i64
  %7 = "tosa.const"() {value = dense_resource<__elided__> : tensor<2x768xf32>} : () -> tensor<2x768xf32>
  %8 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2x768xf32>) : !torch.vtensor<[2,768],f32>
  %9 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %10 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %11 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %12 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %13 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072x768xf32>} : () -> tensor<3072x768xf32>
  %14 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072x768xf32>) : !torch.vtensor<[3072,768],f32>
  %15 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %16 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %17 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x3072xf32>} : () -> tensor<768x3072xf32>
  %18 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x3072xf32>) : !torch.vtensor<[768,3072],f32>
  %19 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072xf32>} : () -> tensor<3072xf32>
  %20 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072xf32>) : !torch.vtensor<[3072],f32>
  %21 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %22 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %23 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %24 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %25 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x768xf32>} : () -> tensor<768x768xf32>
  %26 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x768xf32>) : !torch.vtensor<[768,768],f32>
  %27 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %28 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %29 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x2304xf32>} : () -> tensor<768x2304xf32>
  %30 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x2304xf32>) : !torch.vtensor<[768,2304],f32>
  %31 = "tosa.const"() {value = dense_resource<__elided__> : tensor<2304xf32>} : () -> tensor<2304xf32>
  %32 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2304xf32>) : !torch.vtensor<[2304],f32>
  %33 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %34 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %35 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %36 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %37 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072x768xf32>} : () -> tensor<3072x768xf32>
  %38 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072x768xf32>) : !torch.vtensor<[3072,768],f32>
  %39 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %40 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %41 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x3072xf32>} : () -> tensor<768x3072xf32>
  %42 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x3072xf32>) : !torch.vtensor<[768,3072],f32>
  %43 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072xf32>} : () -> tensor<3072xf32>
  %44 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072xf32>) : !torch.vtensor<[3072],f32>
  %45 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %46 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %47 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %48 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %49 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x768xf32>} : () -> tensor<768x768xf32>
  %50 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x768xf32>) : !torch.vtensor<[768,768],f32>
  %51 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %52 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %53 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x2304xf32>} : () -> tensor<768x2304xf32>
  %54 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x2304xf32>) : !torch.vtensor<[768,2304],f32>
  %55 = "tosa.const"() {value = dense_resource<__elided__> : tensor<2304xf32>} : () -> tensor<2304xf32>
  %56 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2304xf32>) : !torch.vtensor<[2304],f32>
  %57 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %58 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %59 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %60 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %61 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072x768xf32>} : () -> tensor<3072x768xf32>
  %62 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072x768xf32>) : !torch.vtensor<[3072,768],f32>
  %63 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %64 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %65 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x3072xf32>} : () -> tensor<768x3072xf32>
  %66 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x3072xf32>) : !torch.vtensor<[768,3072],f32>
  %67 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072xf32>} : () -> tensor<3072xf32>
  %68 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072xf32>) : !torch.vtensor<[3072],f32>
  %69 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %70 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %71 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %72 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %73 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x768xf32>} : () -> tensor<768x768xf32>
  %74 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x768xf32>) : !torch.vtensor<[768,768],f32>
  %75 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %76 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %77 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x2304xf32>} : () -> tensor<768x2304xf32>
  %78 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x2304xf32>) : !torch.vtensor<[768,2304],f32>
  %79 = "tosa.const"() {value = dense_resource<__elided__> : tensor<2304xf32>} : () -> tensor<2304xf32>
  %80 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2304xf32>) : !torch.vtensor<[2304],f32>
  %81 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %82 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %83 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %84 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %85 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072x768xf32>} : () -> tensor<3072x768xf32>
  %86 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072x768xf32>) : !torch.vtensor<[3072,768],f32>
  %87 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %88 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %89 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x3072xf32>} : () -> tensor<768x3072xf32>
  %90 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x3072xf32>) : !torch.vtensor<[768,3072],f32>
  %91 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072xf32>} : () -> tensor<3072xf32>
  %92 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072xf32>) : !torch.vtensor<[3072],f32>
  %93 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %94 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %95 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %96 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %97 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x768xf32>} : () -> tensor<768x768xf32>
  %98 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x768xf32>) : !torch.vtensor<[768,768],f32>
  %99 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %100 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %101 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x2304xf32>} : () -> tensor<768x2304xf32>
  %102 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x2304xf32>) : !torch.vtensor<[768,2304],f32>
  %103 = "tosa.const"() {value = dense_resource<__elided__> : tensor<2304xf32>} : () -> tensor<2304xf32>
  %104 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2304xf32>) : !torch.vtensor<[2304],f32>
  %105 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %106 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %107 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %108 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %109 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072x768xf32>} : () -> tensor<3072x768xf32>
  %110 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072x768xf32>) : !torch.vtensor<[3072,768],f32>
  %111 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %112 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %113 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x3072xf32>} : () -> tensor<768x3072xf32>
  %114 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x3072xf32>) : !torch.vtensor<[768,3072],f32>
  %115 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072xf32>} : () -> tensor<3072xf32>
  %116 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072xf32>) : !torch.vtensor<[3072],f32>
  %117 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %118 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %119 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %120 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %121 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x768xf32>} : () -> tensor<768x768xf32>
  %122 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x768xf32>) : !torch.vtensor<[768,768],f32>
  %123 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %124 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %125 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x2304xf32>} : () -> tensor<768x2304xf32>
  %126 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x2304xf32>) : !torch.vtensor<[768,2304],f32>
  %127 = "tosa.const"() {value = dense_resource<__elided__> : tensor<2304xf32>} : () -> tensor<2304xf32>
  %128 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2304xf32>) : !torch.vtensor<[2304],f32>
  %129 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %130 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %131 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %132 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %133 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072x768xf32>} : () -> tensor<3072x768xf32>
  %134 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072x768xf32>) : !torch.vtensor<[3072,768],f32>
  %135 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %136 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %137 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x3072xf32>} : () -> tensor<768x3072xf32>
  %138 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x3072xf32>) : !torch.vtensor<[768,3072],f32>
  %139 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072xf32>} : () -> tensor<3072xf32>
  %140 = torch.vtensor.literal(dense_resource<__elided__> : tensor<3072xf32>) : !torch.vtensor<[3072],f32>
  %141 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %142 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %143 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %144 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %145 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x768xf32>} : () -> tensor<768x768xf32>
  %146 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x768xf32>) : !torch.vtensor<[768,768],f32>
  %147 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %148 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %149 = "tosa.const"() {value = dense<-3.40282347E+38> : tensor<f32>} : () -> tensor<f32>
  %150 = torch.vtensor.literal(dense<-3.40282347E+38> : tensor<f32>) : !torch.vtensor<[],f32>
  %151 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1x1x1024x1024xui8>} : () -> tensor<1x1x1024x1024xi8>
  %152 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1x1x1024x1024xui8>) : !torch.vtensor<[1,1,1024,1024],ui8>
  %153 = "tosa.const"() {value = dense<8.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %154 = torch.vtensor.literal(dense<8.000000e+00> : tensor<f32>) : !torch.vtensor<[],f32>
  %155 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x2304xf32>} : () -> tensor<768x2304xf32>
  %156 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768x2304xf32>) : !torch.vtensor<[768,2304],f32>
  %157 = "tosa.const"() {value = dense_resource<__elided__> : tensor<2304xf32>} : () -> tensor<2304xf32>
  %158 = torch.vtensor.literal(dense_resource<__elided__> : tensor<2304xf32>) : !torch.vtensor<[2304],f32>
  %159 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %160 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %161 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32>
  %162 = torch.vtensor.literal(dense_resource<__elided__> : tensor<768xf32>) : !torch.vtensor<[768],f32>
  %163 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1024x768xf32>} : () -> tensor<1024x768xf32>
  %164 = torch.vtensor.literal(dense_resource<__elided__> : tensor<1024x768xf32>) : !torch.vtensor<[1024,768],f32>
  %165 = "tosa.const"() {value = dense_resource<__elided__> : tensor<50257x768xf32>} : () -> tensor<50257x768xf32>
  %166 = torch.vtensor.literal(dense_resource<__elided__> : tensor<50257x768xf32>) : !torch.vtensor<[50257,768],f32>
  %int-2 = torch.constant.int -2
  %167 = builtin.unrealized_conversion_cast %int-2 : !torch.int to i64
  %int11 = torch.constant.int 11
  %168 = builtin.unrealized_conversion_cast %int11 : !torch.int to i64
  %int4 = torch.constant.int 4
  %169 = builtin.unrealized_conversion_cast %int4 : !torch.int to i64
  %int768 = torch.constant.int 768
  %float1.000000e-05 = torch.constant.float 1.000000e-05
  %170 = builtin.unrealized_conversion_cast %float1.000000e-05 : !torch.float to f64
  %int2 = torch.constant.int 2
  %int2304 = torch.constant.int 2304
  %int294912 = torch.constant.int 294912
  %int1536 = torch.constant.int 1536
  %int12 = torch.constant.int 12
  %int64 = torch.constant.int 64
  %int3 = torch.constant.int 3
  %int1024 = torch.constant.int 1024
  %int1048576 = torch.constant.int 1048576
  %int3072 = torch.constant.int 3072
  %float5.000000e-01 = torch.constant.float 5.000000e-01
  %171 = builtin.unrealized_conversion_cast %float5.000000e-01 : !torch.float to f64
  %float3.000000e00 = torch.constant.float 3.000000e+00
  %172 = builtin.unrealized_conversion_cast %float3.000000e00 : !torch.float to f64
  %float4.471500e-02 = torch.constant.float 4.471500e-02
  %173 = builtin.unrealized_conversion_cast %float4.471500e-02 : !torch.float to f64
  %float7.978850e-01 = torch.constant.float 0.79788456080286541
  %174 = builtin.unrealized_conversion_cast %float7.978850e-01 : !torch.float to f64
  %float1.000000e00 = torch.constant.float 1.000000e+00
  %175 = builtin.unrealized_conversion_cast %float1.000000e00 : !torch.float to f64
  %cpu = torch.constant.device "cpu"
  %176 = torch.prim.ListConstruct %int-1, %int128 : (!torch.int, !torch.int) -> !torch.list<int>
  %177 = "tosa.reshape"(%0) {new_shape = [-1, 128]} : (tensor<1x128xi64>) -> tensor<1x128xi64>
  %178 = torch.aten.view %arg0, %176 : !torch.vtensor<[1,128],si64>, !torch.list<int> -> !torch.vtensor<[1,128],si64>
  %179 = "tosa.const"() {value = dense<"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tensor<128xi64>} : () -> tensor<128xi64>
  %180 = "tosa.cast"(%179) : (tensor<128xi64>) -> tensor<128xi64>
  %181 = torch.aten.arange.start_step %int0, %int128, %int1, %int4, %none, %cpu, %false : !torch.int, !torch.int, !torch.int, !torch.int, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[128],si64>
  %182 = "tosa.reshape"(%180) {new_shape = [1, 128]} : (tensor<128xi64>) -> tensor<1x128xi64>
  %183 = torch.aten.unsqueeze %181, %int0 : !torch.vtensor<[128],si64>, !torch.int -> !torch.vtensor<[1,128],si64>
  %184 = "tosa.reshape"(%182) {new_shape = [-1, 128]} : (tensor<1x128xi64>) -> tensor<1x128xi64>
  %185 = torch.aten.view %183, %176 : !torch.vtensor<[1,128],si64>, !torch.list<int> -> !torch.vtensor<[1,128],si64>
  %186 = "tosa.reshape"(%165) {new_shape = [1, 50257, 768]} : (tensor<50257x768xf32>) -> tensor<1x50257x768xf32>
  %187 = "tosa.reshape"(%177) {new_shape = [1, 128]} : (tensor<1x128xi64>) -> tensor<1x128xi64>
  %188 = "tosa.cast"(%187) : (tensor<1x128xi64>) -> tensor<1x128xi32>
  %189 = "tosa.gather"(%186, %188) : (tensor<1x50257x768xf32>, tensor<1x128xi32>) -> tensor<1x128x768xf32>
  %190 = "tosa.reshape"(%189) {new_shape = [1, 128, 768]} : (tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %191 = torch.aten.embedding %166, %178, %int-1, %false, %false : !torch.vtensor<[50257,768],f32>, !torch.vtensor<[1,128],si64>, !torch.int, !torch.bool, !torch.bool -> !torch.vtensor<[1,128,768],f32>
  %192 = "tosa.reshape"(%163) {new_shape = [1, 1024, 768]} : (tensor<1024x768xf32>) -> tensor<1x1024x768xf32>
  %193 = "tosa.reshape"(%184) {new_shape = [1, 128]} : (tensor<1x128xi64>) -> tensor<1x128xi64>
  %194 = "tosa.cast"(%193) : (tensor<1x128xi64>) -> tensor<1x128xi32>
  %195 = "tosa.gather"(%192, %194) : (tensor<1x1024x768xf32>, tensor<1x128xi32>) -> tensor<1x128x768xf32>
  %196 = "tosa.reshape"(%195) {new_shape = [1, 128, 768]} : (tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %197 = torch.aten.embedding %164, %185, %int-1, %false, %false : !torch.vtensor<[1024,768],f32>, !torch.vtensor<[1,128],si64>, !torch.int, !torch.bool, !torch.bool -> !torch.vtensor<[1,128,768],f32>
  %198 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %199 = "tosa.mul"(%196, %198) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %200 = "tosa.add"(%190, %199) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %201 = torch.aten.add.Tensor %191, %197, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %202 = torch.prim.ListConstruct %int768 : (!torch.int) -> !torch.list<int>
  %203 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %204 = "tosa.reciprocal"(%203) : (tensor<1xf32>) -> tensor<1xf32>
  %205 = "tosa.reduce_sum"(%200) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %206 = "tosa.reshape"(%205) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %207 = "tosa.mul"(%206, %204) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %208 = "tosa.sub"(%200, %207) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %209 = "tosa.mul"(%208, %208) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %210 = "tosa.reduce_sum"(%209) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %211 = "tosa.reshape"(%210) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %212 = "tosa.mul"(%211, %204) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %213 = "tosa.reshape"(%161) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %214 = "tosa.reshape"(%159) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %215 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %216 = "tosa.sub"(%200, %207) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %217 = "tosa.add"(%212, %215) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %218 = "tosa.rsqrt"(%217) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %219 = "tosa.mul"(%216, %218) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %220 = "tosa.mul"(%219, %213) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %221 = "tosa.add"(%220, %214) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0, %result1, %result2 = torch.aten.native_layer_norm %201, %202, %162, %160, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %222 = torch.prim.ListConstruct %int-1, %int768 : (!torch.int, !torch.int) -> !torch.list<int>
  %223 = "tosa.reshape"(%221) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %224 = torch.aten.view %result0, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %225 = "tosa.reshape"(%223) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %226 = "tosa.reshape"(%155) {new_shape = [1, 768, 2304]} : (tensor<768x2304xf32>) -> tensor<1x768x2304xf32>
  %227 = "tosa.matmul"(%225, %226) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32>
  %228 = "tosa.reshape"(%227) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32>
  %cast = tensor.cast %228 : tensor<128x2304xf32> to tensor<128x2304xf32>
  %229 = torch.aten.mm %224, %156 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,2304],f32> -> !torch.vtensor<[128,2304],f32>
  %230 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %231 = "tosa.mul"(%157, %230) {shift = 0 : i32} : (tensor<2304xf32>, tensor<f32>) -> tensor<2304xf32>
  %232 = torch.aten.mul.Scalar %158, %int1 : !torch.vtensor<[2304],f32>, !torch.int -> !torch.vtensor<[2304],f32>
  %233 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %234 = "tosa.mul"(%cast, %233) {shift = 0 : i32} : (tensor<128x2304xf32>, tensor<f32>) -> tensor<128x2304xf32>
  %235 = "tosa.add"(%231, %234) : (tensor<2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32>
  %236 = torch.aten.add.Tensor %232, %229, %int1 : !torch.vtensor<[2304],f32>, !torch.vtensor<[128,2304],f32>, !torch.int -> !torch.vtensor<[128,2304],f32>
  %237 = torch.prim.ListConstruct %int1, %int128, %int2304 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %238 = "tosa.reshape"(%235) {new_shape = [1, 128, 2304]} : (tensor<128x2304xf32>) -> tensor<1x128x2304xf32>
  %239 = torch.aten.view %236, %237 : !torch.vtensor<[128,2304],f32>, !torch.list<int> -> !torch.vtensor<[1,128,2304],f32>
  %240 = torch.prim.ListConstruct %int1, %int128, %int768 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %241 = torch.prim.ListConstruct %int294912, %int2304, %int1 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %242 = torch.aten.as_strided %239, %240, %241, %int0 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %243 = builtin.unrealized_conversion_cast %242 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %244 = torch.aten.as_strided %239, %240, %241, %int768 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %245 = builtin.unrealized_conversion_cast %244 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %246 = torch.aten.as_strided %239, %240, %241, %int1536 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %247 = builtin.unrealized_conversion_cast %246 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %248 = torch.prim.ListConstruct %int1, %int128, %int12, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %249 = "tosa.reshape"(%243) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %250 = torch.aten.view %242, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %251 = torch.prim.ListConstruct %int0, %int2, %int1, %int3 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %252 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %253 = "tosa.transpose"(%249, %252) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %254 = torch.aten.permute %250, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %255 = "tosa.reshape"(%245) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %256 = torch.aten.view %244, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %257 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %258 = "tosa.transpose"(%255, %257) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %259 = torch.aten.permute %256, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %260 = "tosa.reshape"(%247) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %261 = torch.aten.view %246, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %262 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %263 = "tosa.transpose"(%260, %262) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %264 = torch.aten.permute %261, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %265 = "tosa.const"() {value = dense<[0, 1, 3, 2]> : tensor<4xi32>} : () -> tensor<4xi32>
  %266 = "tosa.transpose"(%258, %265) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32>
  %267 = torch.aten.transpose.int %259, %int-1, %int-2 : !torch.vtensor<[1,12,128,64],f32>, !torch.int, !torch.int -> !torch.vtensor<[1,12,64,128],f32>
  %268 = torch.prim.ListConstruct %int1, %int12, %int128, %int64 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %269 = torch.aten.broadcast_to %254, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %270 = torch.prim.ListConstruct %int12, %int128, %int64 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %271 = "tosa.reshape"(%253) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %272 = torch.aten.view %254, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %273 = torch.prim.ListConstruct %int1, %int12, %int64, %int128 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %274 = torch.aten.broadcast_to %267, %273 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,64,128],f32>
  %275 = torch.prim.ListConstruct %int12, %int64, %int128 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %276 = "tosa.reshape"(%266) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32>
  %277 = torch.aten.view %267, %275 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[12,64,128],f32>
  %278 = "tosa.matmul"(%271, %276) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32>
  %cast_0 = tensor.cast %278 : tensor<12x128x128xf32> to tensor<12x128x128xf32>
  %279 = torch.aten.bmm %272, %277 : !torch.vtensor<[12,128,64],f32>, !torch.vtensor<[12,64,128],f32> -> !torch.vtensor<[12,128,128],f32>
  %280 = torch.prim.ListConstruct %int1, %int12, %int128, %int128 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %281 = "tosa.reshape"(%cast_0) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %282 = torch.aten.view %279, %280 : !torch.vtensor<[12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %283 = "tosa.cast"(%153) : (tensor<f32>) -> tensor<f32>
  %284 = torch.aten.clone %154, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %285 = "tosa.reciprocal"(%283) : (tensor<f32>) -> tensor<f32>
  %286 = "tosa.mul"(%281, %285) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %287 = torch.aten.div.Tensor %282, %284 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %288 = torch.prim.ListConstruct %int1, %int1, %int1024, %int1024 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %289 = torch.prim.ListConstruct %int1048576, %int1048576, %int1024, %int1 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %290 = torch.aten.as_strided %152, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %291 = torch.aten.as_strided %290, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %292 = torch.prim.ListConstruct %int1, %int1, %int128, %int1024 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %293 = torch.aten.as_strided %291, %292, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,1024],ui8>
  %294 = torch.prim.ListConstruct %int1, %int1, %int128, %int128 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %295 = torch.aten.as_strided %293, %294, %289, %int0 : !torch.vtensor<[1,1,128,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,128],ui8>
  %296 = builtin.unrealized_conversion_cast %295 : !torch.vtensor<[1,1,128,128],ui8> to tensor<1x1x128x128xi8>
  %297 = "tosa.const"() {value = dense<0> : tensor<1x1x128x128xi8>} : () -> tensor<1x1x128x128xi8>
  %298 = "tosa.equal"(%296, %297) : (tensor<1x1x128x128xi8>, tensor<1x1x128x128xi8>) -> tensor<1x1x128x128xi1>
  %299 = "tosa.logical_not"(%298) : (tensor<1x1x128x128xi1>) -> tensor<1x1x128x128xi1>
  %300 = torch.aten.to.dtype %295, %int11, %false, %false, %none : !torch.vtensor<[1,1,128,128],ui8>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[1,1,128,128],i1>
  %301 = "tosa.cast"(%149) : (tensor<f32>) -> tensor<f32>
  %302 = torch.aten.clone %150, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %303 = "tosa.select"(%299, %286, %301) : (tensor<1x1x128x128xi1>, tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %304 = torch.aten.where.self %300, %287, %302 : !torch.vtensor<[1,1,128,128],i1>, !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %305 = torch.prim.ListConstruct %int-1 : (!torch.int) -> !torch.list<int>
  %306 = "tosa.reduce_max"(%303) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %307 = "tosa.argmax"(%303) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128xi64>
  %308 = "tosa.reshape"(%307) {new_shape = [1, 12, 128, 1]} : (tensor<1x12x128xi64>) -> tensor<1x12x128x1xi64>
  %values, %indices = torch.aten.max.dim %304, %int-1, %true : !torch.vtensor<[1,12,128,128],f32>, !torch.int, !torch.bool -> !torch.vtensor<[1,12,128,1],f32>, !torch.vtensor<[1,12,128,1],si64>
  %309 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %310 = "tosa.mul"(%306, %309) {shift = 0 : i32} : (tensor<1x12x128x1xf32>, tensor<f32>) -> tensor<1x12x128x1xf32>
  %311 = "tosa.sub"(%303, %310) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %312 = torch.aten.sub.Tensor %304, %values, %int1 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32>, !torch.int -> !torch.vtensor<[1,12,128,128],f32>
  %313 = "tosa.exp"(%311) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %314 = torch.aten.exp %312 : !torch.vtensor<[1,12,128,128],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %315 = "tosa.reduce_sum"(%313) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %316 = torch.aten.sum.dim_IntList %314, %305, %true, %none : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[1,12,128,1],f32>
  %317 = "tosa.reciprocal"(%315) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32>
  %318 = "tosa.mul"(%313, %317) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %319 = torch.aten.div.Tensor %314, %316 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %320 = torch.aten.broadcast_to %319, %280 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %321 = torch.prim.ListConstruct %int12, %int128, %int128 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %322 = "tosa.reshape"(%318) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32>
  %323 = torch.aten.view %319, %321 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[12,128,128],f32>
  %324 = torch.aten.broadcast_to %264, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %325 = "tosa.reshape"(%263) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %326 = torch.aten.view %264, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %327 = "tosa.matmul"(%322, %325) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32>
  %cast_1 = tensor.cast %327 : tensor<12x128x64xf32> to tensor<12x128x64xf32>
  %328 = torch.aten.bmm %323, %326 : !torch.vtensor<[12,128,128],f32>, !torch.vtensor<[12,128,64],f32> -> !torch.vtensor<[12,128,64],f32>
  %329 = "tosa.reshape"(%cast_1) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32>
  %330 = torch.aten.view %328, %268 : !torch.vtensor<[12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %331 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %332 = "tosa.transpose"(%329, %331) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32>
  %333 = torch.aten.permute %330, %251 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %334 = "tosa.cast"(%332) : (tensor<1x128x12x64xf32>) -> tensor<1x128x12x64xf32>
  %335 = torch.aten.clone %333, %int0 : !torch.vtensor<[1,128,12,64],f32>, !torch.int -> !torch.vtensor<[1,128,12,64],f32>
  %336 = "tosa.reshape"(%334) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32>
  %337 = torch.aten.view %335, %240 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %338 = "tosa.reshape"(%336) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %339 = torch.aten.view %337, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %340 = "tosa.reshape"(%338) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %341 = "tosa.reshape"(%145) {new_shape = [1, 768, 768]} : (tensor<768x768xf32>) -> tensor<1x768x768xf32>
  %342 = "tosa.matmul"(%340, %341) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32>
  %343 = "tosa.reshape"(%342) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_2 = tensor.cast %343 : tensor<128x768xf32> to tensor<128x768xf32>
  %344 = torch.aten.mm %339, %146 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,768],f32> -> !torch.vtensor<[128,768],f32>
  %345 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %346 = "tosa.mul"(%147, %345) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %347 = torch.aten.mul.Scalar %148, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %348 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %349 = "tosa.mul"(%cast_2, %348) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %350 = "tosa.add"(%346, %349) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %351 = torch.aten.add.Tensor %347, %344, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %352 = "tosa.reshape"(%350) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %353 = torch.aten.view %351, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %354 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %355 = "tosa.mul"(%200, %354) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %356 = "tosa.add"(%352, %355) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %357 = torch.aten.add.Tensor %353, %201, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %358 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %359 = "tosa.reciprocal"(%358) : (tensor<1xf32>) -> tensor<1xf32>
  %360 = "tosa.reduce_sum"(%356) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %361 = "tosa.reshape"(%360) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %362 = "tosa.mul"(%361, %359) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %363 = "tosa.sub"(%356, %362) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %364 = "tosa.mul"(%363, %363) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %365 = "tosa.reduce_sum"(%364) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %366 = "tosa.reshape"(%365) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %367 = "tosa.mul"(%366, %359) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %368 = "tosa.reshape"(%143) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %369 = "tosa.reshape"(%141) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %370 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %371 = "tosa.sub"(%356, %362) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %372 = "tosa.add"(%367, %370) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %373 = "tosa.rsqrt"(%372) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %374 = "tosa.mul"(%371, %373) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %375 = "tosa.mul"(%374, %368) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %376 = "tosa.add"(%375, %369) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_3, %result1_4, %result2_5 = torch.aten.native_layer_norm %357, %202, %144, %142, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %377 = "tosa.reshape"(%376) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %378 = torch.aten.view %result0_3, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %379 = "tosa.reshape"(%377) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %380 = "tosa.reshape"(%137) {new_shape = [1, 768, 3072]} : (tensor<768x3072xf32>) -> tensor<1x768x3072xf32>
  %381 = "tosa.matmul"(%379, %380) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32>
  %382 = "tosa.reshape"(%381) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %cast_6 = tensor.cast %382 : tensor<128x3072xf32> to tensor<128x3072xf32>
  %383 = torch.aten.mm %378, %138 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,3072],f32> -> !torch.vtensor<[128,3072],f32>
  %384 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %385 = "tosa.mul"(%139, %384) {shift = 0 : i32} : (tensor<3072xf32>, tensor<f32>) -> tensor<3072xf32>
  %386 = torch.aten.mul.Scalar %140, %int1 : !torch.vtensor<[3072],f32>, !torch.int -> !torch.vtensor<[3072],f32>
  %387 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %388 = "tosa.mul"(%cast_6, %387) {shift = 0 : i32} : (tensor<128x3072xf32>, tensor<f32>) -> tensor<128x3072xf32>
  %389 = "tosa.add"(%385, %388) : (tensor<3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32>
  %390 = torch.aten.add.Tensor %386, %383, %int1 : !torch.vtensor<[3072],f32>, !torch.vtensor<[128,3072],f32>, !torch.int -> !torch.vtensor<[128,3072],f32>
  %391 = torch.prim.ListConstruct %int1, %int128, %int3072 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %392 = "tosa.reshape"(%389) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %393 = torch.aten.view %390, %391 : !torch.vtensor<[128,3072],f32>, !torch.list<int> -> !torch.vtensor<[1,128,3072],f32>
  %394 = "tosa.const"() {value = dense<5.000000e-01> : tensor<f32>} : () -> tensor<f32>
  %395 = "tosa.mul"(%392, %394) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %396 = torch.aten.mul.Scalar %393, %float5.000000e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %397 = "tosa.const"() {value = dense<3.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %398 = "tosa.pow"(%392, %397) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %399 = torch.aten.pow.Tensor_Scalar %393, %float3.000000e00 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %400 = "tosa.const"() {value = dense<4.471500e-02> : tensor<f32>} : () -> tensor<f32>
  %401 = "tosa.mul"(%398, %400) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %402 = torch.aten.mul.Scalar %399, %float4.471500e-02 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %403 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %404 = "tosa.mul"(%401, %403) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %405 = "tosa.add"(%392, %404) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %406 = torch.aten.add.Tensor %393, %402, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32>, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %407 = "tosa.const"() {value = dense<0.797884583> : tensor<f32>} : () -> tensor<f32>
  %408 = "tosa.mul"(%405, %407) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %409 = torch.aten.mul.Scalar %406, %float7.978850e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %410 = "tosa.tanh"(%408) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %411 = torch.aten.tanh %409 : !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %412 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %413 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %414 = "tosa.mul"(%412, %413) {shift = 0 : i32} : (tensor<f32>, tensor<f32>) -> tensor<f32>
  %415 = "tosa.add"(%410, %414) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %416 = torch.aten.add.Scalar %411, %float1.000000e00, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.float, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %417 = "tosa.mul"(%395, %415) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %418 = torch.aten.mul.Tensor %396, %416 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %419 = torch.prim.ListConstruct %int-1, %int3072 : (!torch.int, !torch.int) -> !torch.list<int>
  %420 = "tosa.reshape"(%417) {new_shape = [-1, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %421 = torch.aten.view %418, %419 : !torch.vtensor<[1,128,3072],f32>, !torch.list<int> -> !torch.vtensor<[128,3072],f32>
  %422 = "tosa.reshape"(%420) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %423 = "tosa.reshape"(%133) {new_shape = [1, 3072, 768]} : (tensor<3072x768xf32>) -> tensor<1x3072x768xf32>
  %424 = "tosa.matmul"(%422, %423) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32>
  %425 = "tosa.reshape"(%424) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_7 = tensor.cast %425 : tensor<128x768xf32> to tensor<128x768xf32>
  %426 = torch.aten.mm %421, %134 : !torch.vtensor<[128,3072],f32>, !torch.vtensor<[3072,768],f32> -> !torch.vtensor<[128,768],f32>
  %427 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %428 = "tosa.mul"(%135, %427) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %429 = torch.aten.mul.Scalar %136, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %430 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %431 = "tosa.mul"(%cast_7, %430) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %432 = "tosa.add"(%428, %431) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %433 = torch.aten.add.Tensor %429, %426, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %434 = "tosa.reshape"(%432) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %435 = torch.aten.view %433, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %436 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %437 = "tosa.mul"(%434, %436) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %438 = "tosa.add"(%356, %437) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %439 = torch.aten.add.Tensor %357, %435, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %440 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %441 = "tosa.reciprocal"(%440) : (tensor<1xf32>) -> tensor<1xf32>
  %442 = "tosa.reduce_sum"(%438) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %443 = "tosa.reshape"(%442) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %444 = "tosa.mul"(%443, %441) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %445 = "tosa.sub"(%438, %444) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %446 = "tosa.mul"(%445, %445) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %447 = "tosa.reduce_sum"(%446) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %448 = "tosa.reshape"(%447) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %449 = "tosa.mul"(%448, %441) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %450 = "tosa.reshape"(%131) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %451 = "tosa.reshape"(%129) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %452 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %453 = "tosa.sub"(%438, %444) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %454 = "tosa.add"(%449, %452) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %455 = "tosa.rsqrt"(%454) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %456 = "tosa.mul"(%453, %455) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %457 = "tosa.mul"(%456, %450) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %458 = "tosa.add"(%457, %451) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_8, %result1_9, %result2_10 = torch.aten.native_layer_norm %439, %202, %132, %130, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %459 = "tosa.reshape"(%458) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %460 = torch.aten.view %result0_8, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %461 = "tosa.reshape"(%459) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %462 = "tosa.reshape"(%125) {new_shape = [1, 768, 2304]} : (tensor<768x2304xf32>) -> tensor<1x768x2304xf32>
  %463 = "tosa.matmul"(%461, %462) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32>
  %464 = "tosa.reshape"(%463) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32>
  %cast_11 = tensor.cast %464 : tensor<128x2304xf32> to tensor<128x2304xf32>
  %465 = torch.aten.mm %460, %126 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,2304],f32> -> !torch.vtensor<[128,2304],f32>
  %466 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %467 = "tosa.mul"(%127, %466) {shift = 0 : i32} : (tensor<2304xf32>, tensor<f32>) -> tensor<2304xf32>
  %468 = torch.aten.mul.Scalar %128, %int1 : !torch.vtensor<[2304],f32>, !torch.int -> !torch.vtensor<[2304],f32>
  %469 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %470 = "tosa.mul"(%cast_11, %469) {shift = 0 : i32} : (tensor<128x2304xf32>, tensor<f32>) -> tensor<128x2304xf32>
  %471 = "tosa.add"(%467, %470) : (tensor<2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32>
  %472 = torch.aten.add.Tensor %468, %465, %int1 : !torch.vtensor<[2304],f32>, !torch.vtensor<[128,2304],f32>, !torch.int -> !torch.vtensor<[128,2304],f32>
  %473 = "tosa.reshape"(%471) {new_shape = [1, 128, 2304]} : (tensor<128x2304xf32>) -> tensor<1x128x2304xf32>
  %474 = torch.aten.view %472, %237 : !torch.vtensor<[128,2304],f32>, !torch.list<int> -> !torch.vtensor<[1,128,2304],f32>
  %475 = torch.aten.as_strided %474, %240, %241, %int0 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %476 = builtin.unrealized_conversion_cast %475 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %477 = torch.aten.as_strided %474, %240, %241, %int768 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %478 = builtin.unrealized_conversion_cast %477 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %479 = torch.aten.as_strided %474, %240, %241, %int1536 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %480 = builtin.unrealized_conversion_cast %479 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %481 = "tosa.reshape"(%476) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %482 = torch.aten.view %475, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %483 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %484 = "tosa.transpose"(%481, %483) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %485 = torch.aten.permute %482, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %486 = "tosa.reshape"(%478) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %487 = torch.aten.view %477, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %488 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %489 = "tosa.transpose"(%486, %488) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %490 = torch.aten.permute %487, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %491 = "tosa.reshape"(%480) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %492 = torch.aten.view %479, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %493 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %494 = "tosa.transpose"(%491, %493) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %495 = torch.aten.permute %492, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %496 = "tosa.const"() {value = dense<[0, 1, 3, 2]> : tensor<4xi32>} : () -> tensor<4xi32>
  %497 = "tosa.transpose"(%489, %496) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32>
  %498 = torch.aten.transpose.int %490, %int-1, %int-2 : !torch.vtensor<[1,12,128,64],f32>, !torch.int, !torch.int -> !torch.vtensor<[1,12,64,128],f32>
  %499 = torch.aten.broadcast_to %485, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %500 = "tosa.reshape"(%484) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %501 = torch.aten.view %485, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %502 = torch.aten.broadcast_to %498, %273 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,64,128],f32>
  %503 = "tosa.reshape"(%497) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32>
  %504 = torch.aten.view %498, %275 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[12,64,128],f32>
  %505 = "tosa.matmul"(%500, %503) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32>
  %cast_12 = tensor.cast %505 : tensor<12x128x128xf32> to tensor<12x128x128xf32>
  %506 = torch.aten.bmm %501, %504 : !torch.vtensor<[12,128,64],f32>, !torch.vtensor<[12,64,128],f32> -> !torch.vtensor<[12,128,128],f32>
  %507 = "tosa.reshape"(%cast_12) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %508 = torch.aten.view %506, %280 : !torch.vtensor<[12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %509 = "tosa.cast"(%153) : (tensor<f32>) -> tensor<f32>
  %510 = torch.aten.clone %154, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %511 = "tosa.reciprocal"(%509) : (tensor<f32>) -> tensor<f32>
  %512 = "tosa.mul"(%507, %511) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %513 = torch.aten.div.Tensor %508, %510 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %514 = torch.aten.as_strided %152, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %515 = torch.aten.as_strided %514, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %516 = torch.aten.as_strided %515, %292, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,1024],ui8>
  %517 = torch.aten.as_strided %516, %294, %289, %int0 : !torch.vtensor<[1,1,128,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,128],ui8>
  %518 = builtin.unrealized_conversion_cast %517 : !torch.vtensor<[1,1,128,128],ui8> to tensor<1x1x128x128xi8>
  %519 = "tosa.const"() {value = dense<0> : tensor<1x1x128x128xi8>} : () -> tensor<1x1x128x128xi8>
  %520 = "tosa.equal"(%518, %519) : (tensor<1x1x128x128xi8>, tensor<1x1x128x128xi8>) -> tensor<1x1x128x128xi1>
  %521 = "tosa.logical_not"(%520) : (tensor<1x1x128x128xi1>) -> tensor<1x1x128x128xi1>
  %522 = torch.aten.to.dtype %517, %int11, %false, %false, %none : !torch.vtensor<[1,1,128,128],ui8>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[1,1,128,128],i1>
  %523 = "tosa.cast"(%149) : (tensor<f32>) -> tensor<f32>
  %524 = torch.aten.clone %150, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %525 = "tosa.select"(%521, %512, %523) : (tensor<1x1x128x128xi1>, tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %526 = torch.aten.where.self %522, %513, %524 : !torch.vtensor<[1,1,128,128],i1>, !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %527 = "tosa.reduce_max"(%525) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %528 = "tosa.argmax"(%525) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128xi64>
  %529 = "tosa.reshape"(%528) {new_shape = [1, 12, 128, 1]} : (tensor<1x12x128xi64>) -> tensor<1x12x128x1xi64>
  %values_13, %indices_14 = torch.aten.max.dim %526, %int-1, %true : !torch.vtensor<[1,12,128,128],f32>, !torch.int, !torch.bool -> !torch.vtensor<[1,12,128,1],f32>, !torch.vtensor<[1,12,128,1],si64>
  %530 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %531 = "tosa.mul"(%527, %530) {shift = 0 : i32} : (tensor<1x12x128x1xf32>, tensor<f32>) -> tensor<1x12x128x1xf32>
  %532 = "tosa.sub"(%525, %531) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %533 = torch.aten.sub.Tensor %526, %values_13, %int1 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32>, !torch.int -> !torch.vtensor<[1,12,128,128],f32>
  %534 = "tosa.exp"(%532) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %535 = torch.aten.exp %533 : !torch.vtensor<[1,12,128,128],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %536 = "tosa.reduce_sum"(%534) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %537 = torch.aten.sum.dim_IntList %535, %305, %true, %none : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[1,12,128,1],f32>
  %538 = "tosa.reciprocal"(%536) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32>
  %539 = "tosa.mul"(%534, %538) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %540 = torch.aten.div.Tensor %535, %537 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %541 = torch.aten.broadcast_to %540, %280 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %542 = "tosa.reshape"(%539) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32>
  %543 = torch.aten.view %540, %321 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[12,128,128],f32>
  %544 = torch.aten.broadcast_to %495, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %545 = "tosa.reshape"(%494) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %546 = torch.aten.view %495, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %547 = "tosa.matmul"(%542, %545) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32>
  %cast_15 = tensor.cast %547 : tensor<12x128x64xf32> to tensor<12x128x64xf32>
  %548 = torch.aten.bmm %543, %546 : !torch.vtensor<[12,128,128],f32>, !torch.vtensor<[12,128,64],f32> -> !torch.vtensor<[12,128,64],f32>
  %549 = "tosa.reshape"(%cast_15) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32>
  %550 = torch.aten.view %548, %268 : !torch.vtensor<[12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %551 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %552 = "tosa.transpose"(%549, %551) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32>
  %553 = torch.aten.permute %550, %251 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %554 = "tosa.cast"(%552) : (tensor<1x128x12x64xf32>) -> tensor<1x128x12x64xf32>
  %555 = torch.aten.clone %553, %int0 : !torch.vtensor<[1,128,12,64],f32>, !torch.int -> !torch.vtensor<[1,128,12,64],f32>
  %556 = "tosa.reshape"(%554) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32>
  %557 = torch.aten.view %555, %240 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %558 = "tosa.reshape"(%556) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %559 = torch.aten.view %557, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %560 = "tosa.reshape"(%558) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %561 = "tosa.reshape"(%121) {new_shape = [1, 768, 768]} : (tensor<768x768xf32>) -> tensor<1x768x768xf32>
  %562 = "tosa.matmul"(%560, %561) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32>
  %563 = "tosa.reshape"(%562) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_16 = tensor.cast %563 : tensor<128x768xf32> to tensor<128x768xf32>
  %564 = torch.aten.mm %559, %122 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,768],f32> -> !torch.vtensor<[128,768],f32>
  %565 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %566 = "tosa.mul"(%123, %565) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %567 = torch.aten.mul.Scalar %124, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %568 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %569 = "tosa.mul"(%cast_16, %568) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %570 = "tosa.add"(%566, %569) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %571 = torch.aten.add.Tensor %567, %564, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %572 = "tosa.reshape"(%570) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %573 = torch.aten.view %571, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %574 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %575 = "tosa.mul"(%438, %574) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %576 = "tosa.add"(%572, %575) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %577 = torch.aten.add.Tensor %573, %439, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %578 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %579 = "tosa.reciprocal"(%578) : (tensor<1xf32>) -> tensor<1xf32>
  %580 = "tosa.reduce_sum"(%576) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %581 = "tosa.reshape"(%580) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %582 = "tosa.mul"(%581, %579) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %583 = "tosa.sub"(%576, %582) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %584 = "tosa.mul"(%583, %583) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %585 = "tosa.reduce_sum"(%584) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %586 = "tosa.reshape"(%585) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %587 = "tosa.mul"(%586, %579) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %588 = "tosa.reshape"(%119) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %589 = "tosa.reshape"(%117) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %590 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %591 = "tosa.sub"(%576, %582) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %592 = "tosa.add"(%587, %590) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %593 = "tosa.rsqrt"(%592) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %594 = "tosa.mul"(%591, %593) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %595 = "tosa.mul"(%594, %588) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %596 = "tosa.add"(%595, %589) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_17, %result1_18, %result2_19 = torch.aten.native_layer_norm %577, %202, %120, %118, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %597 = "tosa.reshape"(%596) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %598 = torch.aten.view %result0_17, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %599 = "tosa.reshape"(%597) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %600 = "tosa.reshape"(%113) {new_shape = [1, 768, 3072]} : (tensor<768x3072xf32>) -> tensor<1x768x3072xf32>
  %601 = "tosa.matmul"(%599, %600) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32>
  %602 = "tosa.reshape"(%601) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %cast_20 = tensor.cast %602 : tensor<128x3072xf32> to tensor<128x3072xf32>
  %603 = torch.aten.mm %598, %114 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,3072],f32> -> !torch.vtensor<[128,3072],f32>
  %604 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %605 = "tosa.mul"(%115, %604) {shift = 0 : i32} : (tensor<3072xf32>, tensor<f32>) -> tensor<3072xf32>
  %606 = torch.aten.mul.Scalar %116, %int1 : !torch.vtensor<[3072],f32>, !torch.int -> !torch.vtensor<[3072],f32>
  %607 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %608 = "tosa.mul"(%cast_20, %607) {shift = 0 : i32} : (tensor<128x3072xf32>, tensor<f32>) -> tensor<128x3072xf32>
  %609 = "tosa.add"(%605, %608) : (tensor<3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32>
  %610 = torch.aten.add.Tensor %606, %603, %int1 : !torch.vtensor<[3072],f32>, !torch.vtensor<[128,3072],f32>, !torch.int -> !torch.vtensor<[128,3072],f32>
  %611 = "tosa.reshape"(%609) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %612 = torch.aten.view %610, %391 : !torch.vtensor<[128,3072],f32>, !torch.list<int> -> !torch.vtensor<[1,128,3072],f32>
  %613 = "tosa.const"() {value = dense<5.000000e-01> : tensor<f32>} : () -> tensor<f32>
  %614 = "tosa.mul"(%611, %613) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %615 = torch.aten.mul.Scalar %612, %float5.000000e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %616 = "tosa.const"() {value = dense<3.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %617 = "tosa.pow"(%611, %616) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %618 = torch.aten.pow.Tensor_Scalar %612, %float3.000000e00 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %619 = "tosa.const"() {value = dense<4.471500e-02> : tensor<f32>} : () -> tensor<f32>
  %620 = "tosa.mul"(%617, %619) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %621 = torch.aten.mul.Scalar %618, %float4.471500e-02 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %622 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %623 = "tosa.mul"(%620, %622) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %624 = "tosa.add"(%611, %623) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %625 = torch.aten.add.Tensor %612, %621, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32>, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %626 = "tosa.const"() {value = dense<0.797884583> : tensor<f32>} : () -> tensor<f32>
  %627 = "tosa.mul"(%624, %626) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %628 = torch.aten.mul.Scalar %625, %float7.978850e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %629 = "tosa.tanh"(%627) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %630 = torch.aten.tanh %628 : !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %631 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %632 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %633 = "tosa.mul"(%631, %632) {shift = 0 : i32} : (tensor<f32>, tensor<f32>) -> tensor<f32>
  %634 = "tosa.add"(%629, %633) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %635 = torch.aten.add.Scalar %630, %float1.000000e00, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.float, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %636 = "tosa.mul"(%614, %634) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %637 = torch.aten.mul.Tensor %615, %635 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %638 = "tosa.reshape"(%636) {new_shape = [-1, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %639 = torch.aten.view %637, %419 : !torch.vtensor<[1,128,3072],f32>, !torch.list<int> -> !torch.vtensor<[128,3072],f32>
  %640 = "tosa.reshape"(%638) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %641 = "tosa.reshape"(%109) {new_shape = [1, 3072, 768]} : (tensor<3072x768xf32>) -> tensor<1x3072x768xf32>
  %642 = "tosa.matmul"(%640, %641) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32>
  %643 = "tosa.reshape"(%642) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_21 = tensor.cast %643 : tensor<128x768xf32> to tensor<128x768xf32>
  %644 = torch.aten.mm %639, %110 : !torch.vtensor<[128,3072],f32>, !torch.vtensor<[3072,768],f32> -> !torch.vtensor<[128,768],f32>
  %645 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %646 = "tosa.mul"(%111, %645) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %647 = torch.aten.mul.Scalar %112, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %648 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %649 = "tosa.mul"(%cast_21, %648) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %650 = "tosa.add"(%646, %649) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %651 = torch.aten.add.Tensor %647, %644, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %652 = "tosa.reshape"(%650) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %653 = torch.aten.view %651, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %654 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %655 = "tosa.mul"(%652, %654) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %656 = "tosa.add"(%576, %655) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %657 = torch.aten.add.Tensor %577, %653, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %658 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %659 = "tosa.reciprocal"(%658) : (tensor<1xf32>) -> tensor<1xf32>
  %660 = "tosa.reduce_sum"(%656) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %661 = "tosa.reshape"(%660) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %662 = "tosa.mul"(%661, %659) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %663 = "tosa.sub"(%656, %662) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %664 = "tosa.mul"(%663, %663) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %665 = "tosa.reduce_sum"(%664) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %666 = "tosa.reshape"(%665) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %667 = "tosa.mul"(%666, %659) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %668 = "tosa.reshape"(%107) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %669 = "tosa.reshape"(%105) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %670 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %671 = "tosa.sub"(%656, %662) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %672 = "tosa.add"(%667, %670) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %673 = "tosa.rsqrt"(%672) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %674 = "tosa.mul"(%671, %673) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %675 = "tosa.mul"(%674, %668) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %676 = "tosa.add"(%675, %669) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_22, %result1_23, %result2_24 = torch.aten.native_layer_norm %657, %202, %108, %106, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %677 = "tosa.reshape"(%676) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %678 = torch.aten.view %result0_22, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %679 = "tosa.reshape"(%677) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %680 = "tosa.reshape"(%101) {new_shape = [1, 768, 2304]} : (tensor<768x2304xf32>) -> tensor<1x768x2304xf32>
  %681 = "tosa.matmul"(%679, %680) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32>
  %682 = "tosa.reshape"(%681) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32>
  %cast_25 = tensor.cast %682 : tensor<128x2304xf32> to tensor<128x2304xf32>
  %683 = torch.aten.mm %678, %102 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,2304],f32> -> !torch.vtensor<[128,2304],f32>
  %684 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %685 = "tosa.mul"(%103, %684) {shift = 0 : i32} : (tensor<2304xf32>, tensor<f32>) -> tensor<2304xf32>
  %686 = torch.aten.mul.Scalar %104, %int1 : !torch.vtensor<[2304],f32>, !torch.int -> !torch.vtensor<[2304],f32>
  %687 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %688 = "tosa.mul"(%cast_25, %687) {shift = 0 : i32} : (tensor<128x2304xf32>, tensor<f32>) -> tensor<128x2304xf32>
  %689 = "tosa.add"(%685, %688) : (tensor<2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32>
  %690 = torch.aten.add.Tensor %686, %683, %int1 : !torch.vtensor<[2304],f32>, !torch.vtensor<[128,2304],f32>, !torch.int -> !torch.vtensor<[128,2304],f32>
  %691 = "tosa.reshape"(%689) {new_shape = [1, 128, 2304]} : (tensor<128x2304xf32>) -> tensor<1x128x2304xf32>
  %692 = torch.aten.view %690, %237 : !torch.vtensor<[128,2304],f32>, !torch.list<int> -> !torch.vtensor<[1,128,2304],f32>
  %693 = torch.aten.as_strided %692, %240, %241, %int0 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %694 = builtin.unrealized_conversion_cast %693 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %695 = torch.aten.as_strided %692, %240, %241, %int768 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %696 = builtin.unrealized_conversion_cast %695 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %697 = torch.aten.as_strided %692, %240, %241, %int1536 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %698 = builtin.unrealized_conversion_cast %697 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %699 = "tosa.reshape"(%694) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %700 = torch.aten.view %693, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %701 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %702 = "tosa.transpose"(%699, %701) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %703 = torch.aten.permute %700, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %704 = "tosa.reshape"(%696) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %705 = torch.aten.view %695, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %706 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %707 = "tosa.transpose"(%704, %706) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %708 = torch.aten.permute %705, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %709 = "tosa.reshape"(%698) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %710 = torch.aten.view %697, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %711 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %712 = "tosa.transpose"(%709, %711) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %713 = torch.aten.permute %710, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %714 = "tosa.const"() {value = dense<[0, 1, 3, 2]> : tensor<4xi32>} : () -> tensor<4xi32>
  %715 = "tosa.transpose"(%707, %714) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32>
  %716 = torch.aten.transpose.int %708, %int-1, %int-2 : !torch.vtensor<[1,12,128,64],f32>, !torch.int, !torch.int -> !torch.vtensor<[1,12,64,128],f32>
  %717 = torch.aten.broadcast_to %703, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %718 = "tosa.reshape"(%702) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %719 = torch.aten.view %703, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %720 = torch.aten.broadcast_to %716, %273 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,64,128],f32>
  %721 = "tosa.reshape"(%715) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32>
  %722 = torch.aten.view %716, %275 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[12,64,128],f32>
  %723 = "tosa.matmul"(%718, %721) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32>
  %cast_26 = tensor.cast %723 : tensor<12x128x128xf32> to tensor<12x128x128xf32>
  %724 = torch.aten.bmm %719, %722 : !torch.vtensor<[12,128,64],f32>, !torch.vtensor<[12,64,128],f32> -> !torch.vtensor<[12,128,128],f32>
  %725 = "tosa.reshape"(%cast_26) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %726 = torch.aten.view %724, %280 : !torch.vtensor<[12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %727 = "tosa.cast"(%153) : (tensor<f32>) -> tensor<f32>
  %728 = torch.aten.clone %154, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %729 = "tosa.reciprocal"(%727) : (tensor<f32>) -> tensor<f32>
  %730 = "tosa.mul"(%725, %729) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %731 = torch.aten.div.Tensor %726, %728 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %732 = torch.aten.as_strided %152, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %733 = torch.aten.as_strided %732, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %734 = torch.aten.as_strided %733, %292, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,1024],ui8>
  %735 = torch.aten.as_strided %734, %294, %289, %int0 : !torch.vtensor<[1,1,128,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,128],ui8>
  %736 = builtin.unrealized_conversion_cast %735 : !torch.vtensor<[1,1,128,128],ui8> to tensor<1x1x128x128xi8>
  %737 = "tosa.const"() {value = dense<0> : tensor<1x1x128x128xi8>} : () -> tensor<1x1x128x128xi8>
  %738 = "tosa.equal"(%736, %737) : (tensor<1x1x128x128xi8>, tensor<1x1x128x128xi8>) -> tensor<1x1x128x128xi1>
  %739 = "tosa.logical_not"(%738) : (tensor<1x1x128x128xi1>) -> tensor<1x1x128x128xi1>
  %740 = torch.aten.to.dtype %735, %int11, %false, %false, %none : !torch.vtensor<[1,1,128,128],ui8>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[1,1,128,128],i1>
  %741 = "tosa.cast"(%149) : (tensor<f32>) -> tensor<f32>
  %742 = torch.aten.clone %150, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %743 = "tosa.select"(%739, %730, %741) : (tensor<1x1x128x128xi1>, tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %744 = torch.aten.where.self %740, %731, %742 : !torch.vtensor<[1,1,128,128],i1>, !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %745 = "tosa.reduce_max"(%743) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %746 = "tosa.argmax"(%743) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128xi64>
  %747 = "tosa.reshape"(%746) {new_shape = [1, 12, 128, 1]} : (tensor<1x12x128xi64>) -> tensor<1x12x128x1xi64>
  %values_27, %indices_28 = torch.aten.max.dim %744, %int-1, %true : !torch.vtensor<[1,12,128,128],f32>, !torch.int, !torch.bool -> !torch.vtensor<[1,12,128,1],f32>, !torch.vtensor<[1,12,128,1],si64>
  %748 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %749 = "tosa.mul"(%745, %748) {shift = 0 : i32} : (tensor<1x12x128x1xf32>, tensor<f32>) -> tensor<1x12x128x1xf32>
  %750 = "tosa.sub"(%743, %749) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %751 = torch.aten.sub.Tensor %744, %values_27, %int1 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32>, !torch.int -> !torch.vtensor<[1,12,128,128],f32>
  %752 = "tosa.exp"(%750) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %753 = torch.aten.exp %751 : !torch.vtensor<[1,12,128,128],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %754 = "tosa.reduce_sum"(%752) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %755 = torch.aten.sum.dim_IntList %753, %305, %true, %none : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[1,12,128,1],f32>
  %756 = "tosa.reciprocal"(%754) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32>
  %757 = "tosa.mul"(%752, %756) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %758 = torch.aten.div.Tensor %753, %755 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %759 = torch.aten.broadcast_to %758, %280 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %760 = "tosa.reshape"(%757) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32>
  %761 = torch.aten.view %758, %321 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[12,128,128],f32>
  %762 = torch.aten.broadcast_to %713, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %763 = "tosa.reshape"(%712) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %764 = torch.aten.view %713, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %765 = "tosa.matmul"(%760, %763) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32>
  %cast_29 = tensor.cast %765 : tensor<12x128x64xf32> to tensor<12x128x64xf32>
  %766 = torch.aten.bmm %761, %764 : !torch.vtensor<[12,128,128],f32>, !torch.vtensor<[12,128,64],f32> -> !torch.vtensor<[12,128,64],f32>
  %767 = "tosa.reshape"(%cast_29) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32>
  %768 = torch.aten.view %766, %268 : !torch.vtensor<[12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %769 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %770 = "tosa.transpose"(%767, %769) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32>
  %771 = torch.aten.permute %768, %251 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %772 = "tosa.cast"(%770) : (tensor<1x128x12x64xf32>) -> tensor<1x128x12x64xf32>
  %773 = torch.aten.clone %771, %int0 : !torch.vtensor<[1,128,12,64],f32>, !torch.int -> !torch.vtensor<[1,128,12,64],f32>
  %774 = "tosa.reshape"(%772) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32>
  %775 = torch.aten.view %773, %240 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %776 = "tosa.reshape"(%774) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %777 = torch.aten.view %775, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %778 = "tosa.reshape"(%776) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %779 = "tosa.reshape"(%97) {new_shape = [1, 768, 768]} : (tensor<768x768xf32>) -> tensor<1x768x768xf32>
  %780 = "tosa.matmul"(%778, %779) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32>
  %781 = "tosa.reshape"(%780) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_30 = tensor.cast %781 : tensor<128x768xf32> to tensor<128x768xf32>
  %782 = torch.aten.mm %777, %98 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,768],f32> -> !torch.vtensor<[128,768],f32>
  %783 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %784 = "tosa.mul"(%99, %783) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %785 = torch.aten.mul.Scalar %100, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %786 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %787 = "tosa.mul"(%cast_30, %786) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %788 = "tosa.add"(%784, %787) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %789 = torch.aten.add.Tensor %785, %782, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %790 = "tosa.reshape"(%788) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %791 = torch.aten.view %789, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %792 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %793 = "tosa.mul"(%656, %792) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %794 = "tosa.add"(%790, %793) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %795 = torch.aten.add.Tensor %791, %657, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %796 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %797 = "tosa.reciprocal"(%796) : (tensor<1xf32>) -> tensor<1xf32>
  %798 = "tosa.reduce_sum"(%794) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %799 = "tosa.reshape"(%798) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %800 = "tosa.mul"(%799, %797) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %801 = "tosa.sub"(%794, %800) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %802 = "tosa.mul"(%801, %801) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %803 = "tosa.reduce_sum"(%802) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %804 = "tosa.reshape"(%803) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %805 = "tosa.mul"(%804, %797) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %806 = "tosa.reshape"(%95) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %807 = "tosa.reshape"(%93) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %808 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %809 = "tosa.sub"(%794, %800) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %810 = "tosa.add"(%805, %808) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %811 = "tosa.rsqrt"(%810) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %812 = "tosa.mul"(%809, %811) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %813 = "tosa.mul"(%812, %806) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %814 = "tosa.add"(%813, %807) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_31, %result1_32, %result2_33 = torch.aten.native_layer_norm %795, %202, %96, %94, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %815 = "tosa.reshape"(%814) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %816 = torch.aten.view %result0_31, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %817 = "tosa.reshape"(%815) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %818 = "tosa.reshape"(%89) {new_shape = [1, 768, 3072]} : (tensor<768x3072xf32>) -> tensor<1x768x3072xf32>
  %819 = "tosa.matmul"(%817, %818) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32>
  %820 = "tosa.reshape"(%819) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %cast_34 = tensor.cast %820 : tensor<128x3072xf32> to tensor<128x3072xf32>
  %821 = torch.aten.mm %816, %90 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,3072],f32> -> !torch.vtensor<[128,3072],f32>
  %822 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %823 = "tosa.mul"(%91, %822) {shift = 0 : i32} : (tensor<3072xf32>, tensor<f32>) -> tensor<3072xf32>
  %824 = torch.aten.mul.Scalar %92, %int1 : !torch.vtensor<[3072],f32>, !torch.int -> !torch.vtensor<[3072],f32>
  %825 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %826 = "tosa.mul"(%cast_34, %825) {shift = 0 : i32} : (tensor<128x3072xf32>, tensor<f32>) -> tensor<128x3072xf32>
  %827 = "tosa.add"(%823, %826) : (tensor<3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32>
  %828 = torch.aten.add.Tensor %824, %821, %int1 : !torch.vtensor<[3072],f32>, !torch.vtensor<[128,3072],f32>, !torch.int -> !torch.vtensor<[128,3072],f32>
  %829 = "tosa.reshape"(%827) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %830 = torch.aten.view %828, %391 : !torch.vtensor<[128,3072],f32>, !torch.list<int> -> !torch.vtensor<[1,128,3072],f32>
  %831 = "tosa.const"() {value = dense<5.000000e-01> : tensor<f32>} : () -> tensor<f32>
  %832 = "tosa.mul"(%829, %831) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %833 = torch.aten.mul.Scalar %830, %float5.000000e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %834 = "tosa.const"() {value = dense<3.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %835 = "tosa.pow"(%829, %834) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %836 = torch.aten.pow.Tensor_Scalar %830, %float3.000000e00 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %837 = "tosa.const"() {value = dense<4.471500e-02> : tensor<f32>} : () -> tensor<f32>
  %838 = "tosa.mul"(%835, %837) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %839 = torch.aten.mul.Scalar %836, %float4.471500e-02 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %840 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %841 = "tosa.mul"(%838, %840) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %842 = "tosa.add"(%829, %841) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %843 = torch.aten.add.Tensor %830, %839, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32>, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %844 = "tosa.const"() {value = dense<0.797884583> : tensor<f32>} : () -> tensor<f32>
  %845 = "tosa.mul"(%842, %844) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %846 = torch.aten.mul.Scalar %843, %float7.978850e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %847 = "tosa.tanh"(%845) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %848 = torch.aten.tanh %846 : !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %849 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %850 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %851 = "tosa.mul"(%849, %850) {shift = 0 : i32} : (tensor<f32>, tensor<f32>) -> tensor<f32>
  %852 = "tosa.add"(%847, %851) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %853 = torch.aten.add.Scalar %848, %float1.000000e00, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.float, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %854 = "tosa.mul"(%832, %852) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %855 = torch.aten.mul.Tensor %833, %853 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %856 = "tosa.reshape"(%854) {new_shape = [-1, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %857 = torch.aten.view %855, %419 : !torch.vtensor<[1,128,3072],f32>, !torch.list<int> -> !torch.vtensor<[128,3072],f32>
  %858 = "tosa.reshape"(%856) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %859 = "tosa.reshape"(%85) {new_shape = [1, 3072, 768]} : (tensor<3072x768xf32>) -> tensor<1x3072x768xf32>
  %860 = "tosa.matmul"(%858, %859) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32>
  %861 = "tosa.reshape"(%860) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_35 = tensor.cast %861 : tensor<128x768xf32> to tensor<128x768xf32>
  %862 = torch.aten.mm %857, %86 : !torch.vtensor<[128,3072],f32>, !torch.vtensor<[3072,768],f32> -> !torch.vtensor<[128,768],f32>
  %863 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %864 = "tosa.mul"(%87, %863) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %865 = torch.aten.mul.Scalar %88, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %866 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %867 = "tosa.mul"(%cast_35, %866) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %868 = "tosa.add"(%864, %867) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %869 = torch.aten.add.Tensor %865, %862, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %870 = "tosa.reshape"(%868) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %871 = torch.aten.view %869, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %872 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %873 = "tosa.mul"(%870, %872) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %874 = "tosa.add"(%794, %873) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %875 = torch.aten.add.Tensor %795, %871, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %876 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %877 = "tosa.reciprocal"(%876) : (tensor<1xf32>) -> tensor<1xf32>
  %878 = "tosa.reduce_sum"(%874) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %879 = "tosa.reshape"(%878) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %880 = "tosa.mul"(%879, %877) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %881 = "tosa.sub"(%874, %880) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %882 = "tosa.mul"(%881, %881) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %883 = "tosa.reduce_sum"(%882) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %884 = "tosa.reshape"(%883) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %885 = "tosa.mul"(%884, %877) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %886 = "tosa.reshape"(%83) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %887 = "tosa.reshape"(%81) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %888 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %889 = "tosa.sub"(%874, %880) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %890 = "tosa.add"(%885, %888) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %891 = "tosa.rsqrt"(%890) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %892 = "tosa.mul"(%889, %891) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %893 = "tosa.mul"(%892, %886) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %894 = "tosa.add"(%893, %887) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_36, %result1_37, %result2_38 = torch.aten.native_layer_norm %875, %202, %84, %82, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %895 = "tosa.reshape"(%894) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %896 = torch.aten.view %result0_36, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %897 = "tosa.reshape"(%895) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %898 = "tosa.reshape"(%77) {new_shape = [1, 768, 2304]} : (tensor<768x2304xf32>) -> tensor<1x768x2304xf32>
  %899 = "tosa.matmul"(%897, %898) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32>
  %900 = "tosa.reshape"(%899) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32>
  %cast_39 = tensor.cast %900 : tensor<128x2304xf32> to tensor<128x2304xf32>
  %901 = torch.aten.mm %896, %78 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,2304],f32> -> !torch.vtensor<[128,2304],f32>
  %902 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %903 = "tosa.mul"(%79, %902) {shift = 0 : i32} : (tensor<2304xf32>, tensor<f32>) -> tensor<2304xf32>
  %904 = torch.aten.mul.Scalar %80, %int1 : !torch.vtensor<[2304],f32>, !torch.int -> !torch.vtensor<[2304],f32>
  %905 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %906 = "tosa.mul"(%cast_39, %905) {shift = 0 : i32} : (tensor<128x2304xf32>, tensor<f32>) -> tensor<128x2304xf32>
  %907 = "tosa.add"(%903, %906) : (tensor<2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32>
  %908 = torch.aten.add.Tensor %904, %901, %int1 : !torch.vtensor<[2304],f32>, !torch.vtensor<[128,2304],f32>, !torch.int -> !torch.vtensor<[128,2304],f32>
  %909 = "tosa.reshape"(%907) {new_shape = [1, 128, 2304]} : (tensor<128x2304xf32>) -> tensor<1x128x2304xf32>
  %910 = torch.aten.view %908, %237 : !torch.vtensor<[128,2304],f32>, !torch.list<int> -> !torch.vtensor<[1,128,2304],f32>
  %911 = torch.aten.as_strided %910, %240, %241, %int0 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %912 = builtin.unrealized_conversion_cast %911 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %913 = torch.aten.as_strided %910, %240, %241, %int768 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %914 = builtin.unrealized_conversion_cast %913 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %915 = torch.aten.as_strided %910, %240, %241, %int1536 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %916 = builtin.unrealized_conversion_cast %915 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %917 = "tosa.reshape"(%912) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %918 = torch.aten.view %911, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %919 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %920 = "tosa.transpose"(%917, %919) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %921 = torch.aten.permute %918, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %922 = "tosa.reshape"(%914) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %923 = torch.aten.view %913, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %924 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %925 = "tosa.transpose"(%922, %924) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %926 = torch.aten.permute %923, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %927 = "tosa.reshape"(%916) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %928 = torch.aten.view %915, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %929 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %930 = "tosa.transpose"(%927, %929) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %931 = torch.aten.permute %928, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %932 = "tosa.const"() {value = dense<[0, 1, 3, 2]> : tensor<4xi32>} : () -> tensor<4xi32>
  %933 = "tosa.transpose"(%925, %932) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32>
  %934 = torch.aten.transpose.int %926, %int-1, %int-2 : !torch.vtensor<[1,12,128,64],f32>, !torch.int, !torch.int -> !torch.vtensor<[1,12,64,128],f32>
  %935 = torch.aten.broadcast_to %921, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %936 = "tosa.reshape"(%920) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %937 = torch.aten.view %921, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %938 = torch.aten.broadcast_to %934, %273 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,64,128],f32>
  %939 = "tosa.reshape"(%933) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32>
  %940 = torch.aten.view %934, %275 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[12,64,128],f32>
  %941 = "tosa.matmul"(%936, %939) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32>
  %cast_40 = tensor.cast %941 : tensor<12x128x128xf32> to tensor<12x128x128xf32>
  %942 = torch.aten.bmm %937, %940 : !torch.vtensor<[12,128,64],f32>, !torch.vtensor<[12,64,128],f32> -> !torch.vtensor<[12,128,128],f32>
  %943 = "tosa.reshape"(%cast_40) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %944 = torch.aten.view %942, %280 : !torch.vtensor<[12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %945 = "tosa.cast"(%153) : (tensor<f32>) -> tensor<f32>
  %946 = torch.aten.clone %154, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %947 = "tosa.reciprocal"(%945) : (tensor<f32>) -> tensor<f32>
  %948 = "tosa.mul"(%943, %947) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %949 = torch.aten.div.Tensor %944, %946 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %950 = torch.aten.as_strided %152, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %951 = torch.aten.as_strided %950, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %952 = torch.aten.as_strided %951, %292, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,1024],ui8>
  %953 = torch.aten.as_strided %952, %294, %289, %int0 : !torch.vtensor<[1,1,128,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,128],ui8>
  %954 = builtin.unrealized_conversion_cast %953 : !torch.vtensor<[1,1,128,128],ui8> to tensor<1x1x128x128xi8>
  %955 = "tosa.const"() {value = dense<0> : tensor<1x1x128x128xi8>} : () -> tensor<1x1x128x128xi8>
  %956 = "tosa.equal"(%954, %955) : (tensor<1x1x128x128xi8>, tensor<1x1x128x128xi8>) -> tensor<1x1x128x128xi1>
  %957 = "tosa.logical_not"(%956) : (tensor<1x1x128x128xi1>) -> tensor<1x1x128x128xi1>
  %958 = torch.aten.to.dtype %953, %int11, %false, %false, %none : !torch.vtensor<[1,1,128,128],ui8>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[1,1,128,128],i1>
  %959 = "tosa.cast"(%149) : (tensor<f32>) -> tensor<f32>
  %960 = torch.aten.clone %150, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %961 = "tosa.select"(%957, %948, %959) : (tensor<1x1x128x128xi1>, tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %962 = torch.aten.where.self %958, %949, %960 : !torch.vtensor<[1,1,128,128],i1>, !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %963 = "tosa.reduce_max"(%961) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %964 = "tosa.argmax"(%961) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128xi64>
  %965 = "tosa.reshape"(%964) {new_shape = [1, 12, 128, 1]} : (tensor<1x12x128xi64>) -> tensor<1x12x128x1xi64>
  %values_41, %indices_42 = torch.aten.max.dim %962, %int-1, %true : !torch.vtensor<[1,12,128,128],f32>, !torch.int, !torch.bool -> !torch.vtensor<[1,12,128,1],f32>, !torch.vtensor<[1,12,128,1],si64>
  %966 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %967 = "tosa.mul"(%963, %966) {shift = 0 : i32} : (tensor<1x12x128x1xf32>, tensor<f32>) -> tensor<1x12x128x1xf32>
  %968 = "tosa.sub"(%961, %967) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %969 = torch.aten.sub.Tensor %962, %values_41, %int1 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32>, !torch.int -> !torch.vtensor<[1,12,128,128],f32>
  %970 = "tosa.exp"(%968) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %971 = torch.aten.exp %969 : !torch.vtensor<[1,12,128,128],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %972 = "tosa.reduce_sum"(%970) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %973 = torch.aten.sum.dim_IntList %971, %305, %true, %none : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[1,12,128,1],f32>
  %974 = "tosa.reciprocal"(%972) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32>
  %975 = "tosa.mul"(%970, %974) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %976 = torch.aten.div.Tensor %971, %973 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %977 = torch.aten.broadcast_to %976, %280 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %978 = "tosa.reshape"(%975) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32>
  %979 = torch.aten.view %976, %321 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[12,128,128],f32>
  %980 = torch.aten.broadcast_to %931, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %981 = "tosa.reshape"(%930) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %982 = torch.aten.view %931, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %983 = "tosa.matmul"(%978, %981) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32>
  %cast_43 = tensor.cast %983 : tensor<12x128x64xf32> to tensor<12x128x64xf32>
  %984 = torch.aten.bmm %979, %982 : !torch.vtensor<[12,128,128],f32>, !torch.vtensor<[12,128,64],f32> -> !torch.vtensor<[12,128,64],f32>
  %985 = "tosa.reshape"(%cast_43) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32>
  %986 = torch.aten.view %984, %268 : !torch.vtensor<[12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %987 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %988 = "tosa.transpose"(%985, %987) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32>
  %989 = torch.aten.permute %986, %251 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %990 = "tosa.cast"(%988) : (tensor<1x128x12x64xf32>) -> tensor<1x128x12x64xf32>
  %991 = torch.aten.clone %989, %int0 : !torch.vtensor<[1,128,12,64],f32>, !torch.int -> !torch.vtensor<[1,128,12,64],f32>
  %992 = "tosa.reshape"(%990) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32>
  %993 = torch.aten.view %991, %240 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %994 = "tosa.reshape"(%992) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %995 = torch.aten.view %993, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %996 = "tosa.reshape"(%994) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %997 = "tosa.reshape"(%73) {new_shape = [1, 768, 768]} : (tensor<768x768xf32>) -> tensor<1x768x768xf32>
  %998 = "tosa.matmul"(%996, %997) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32>
  %999 = "tosa.reshape"(%998) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_44 = tensor.cast %999 : tensor<128x768xf32> to tensor<128x768xf32>
  %1000 = torch.aten.mm %995, %74 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,768],f32> -> !torch.vtensor<[128,768],f32>
  %1001 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1002 = "tosa.mul"(%75, %1001) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %1003 = torch.aten.mul.Scalar %76, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %1004 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1005 = "tosa.mul"(%cast_44, %1004) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %1006 = "tosa.add"(%1002, %1005) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %1007 = torch.aten.add.Tensor %1003, %1000, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %1008 = "tosa.reshape"(%1006) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1009 = torch.aten.view %1007, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %1010 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1011 = "tosa.mul"(%874, %1010) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %1012 = "tosa.add"(%1008, %1011) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1013 = torch.aten.add.Tensor %1009, %875, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1014 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %1015 = "tosa.reciprocal"(%1014) : (tensor<1xf32>) -> tensor<1xf32>
  %1016 = "tosa.reduce_sum"(%1012) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1017 = "tosa.reshape"(%1016) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1018 = "tosa.mul"(%1017, %1015) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1019 = "tosa.sub"(%1012, %1018) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1020 = "tosa.mul"(%1019, %1019) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1021 = "tosa.reduce_sum"(%1020) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1022 = "tosa.reshape"(%1021) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1023 = "tosa.mul"(%1022, %1015) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1024 = "tosa.reshape"(%71) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1025 = "tosa.reshape"(%69) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1026 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %1027 = "tosa.sub"(%1012, %1018) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1028 = "tosa.add"(%1023, %1026) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %1029 = "tosa.rsqrt"(%1028) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1030 = "tosa.mul"(%1027, %1029) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1031 = "tosa.mul"(%1030, %1024) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %1032 = "tosa.add"(%1031, %1025) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_45, %result1_46, %result2_47 = torch.aten.native_layer_norm %1013, %202, %72, %70, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %1033 = "tosa.reshape"(%1032) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %1034 = torch.aten.view %result0_45, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %1035 = "tosa.reshape"(%1033) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1036 = "tosa.reshape"(%65) {new_shape = [1, 768, 3072]} : (tensor<768x3072xf32>) -> tensor<1x768x3072xf32>
  %1037 = "tosa.matmul"(%1035, %1036) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32>
  %1038 = "tosa.reshape"(%1037) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %cast_48 = tensor.cast %1038 : tensor<128x3072xf32> to tensor<128x3072xf32>
  %1039 = torch.aten.mm %1034, %66 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,3072],f32> -> !torch.vtensor<[128,3072],f32>
  %1040 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1041 = "tosa.mul"(%67, %1040) {shift = 0 : i32} : (tensor<3072xf32>, tensor<f32>) -> tensor<3072xf32>
  %1042 = torch.aten.mul.Scalar %68, %int1 : !torch.vtensor<[3072],f32>, !torch.int -> !torch.vtensor<[3072],f32>
  %1043 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1044 = "tosa.mul"(%cast_48, %1043) {shift = 0 : i32} : (tensor<128x3072xf32>, tensor<f32>) -> tensor<128x3072xf32>
  %1045 = "tosa.add"(%1041, %1044) : (tensor<3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32>
  %1046 = torch.aten.add.Tensor %1042, %1039, %int1 : !torch.vtensor<[3072],f32>, !torch.vtensor<[128,3072],f32>, !torch.int -> !torch.vtensor<[128,3072],f32>
  %1047 = "tosa.reshape"(%1045) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %1048 = torch.aten.view %1046, %391 : !torch.vtensor<[128,3072],f32>, !torch.list<int> -> !torch.vtensor<[1,128,3072],f32>
  %1049 = "tosa.const"() {value = dense<5.000000e-01> : tensor<f32>} : () -> tensor<f32>
  %1050 = "tosa.mul"(%1047, %1049) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1051 = torch.aten.mul.Scalar %1048, %float5.000000e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1052 = "tosa.const"() {value = dense<3.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1053 = "tosa.pow"(%1047, %1052) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1054 = torch.aten.pow.Tensor_Scalar %1048, %float3.000000e00 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1055 = "tosa.const"() {value = dense<4.471500e-02> : tensor<f32>} : () -> tensor<f32>
  %1056 = "tosa.mul"(%1053, %1055) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1057 = torch.aten.mul.Scalar %1054, %float4.471500e-02 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1058 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1059 = "tosa.mul"(%1056, %1058) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1060 = "tosa.add"(%1047, %1059) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %1061 = torch.aten.add.Tensor %1048, %1057, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32>, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %1062 = "tosa.const"() {value = dense<0.797884583> : tensor<f32>} : () -> tensor<f32>
  %1063 = "tosa.mul"(%1060, %1062) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1064 = torch.aten.mul.Scalar %1061, %float7.978850e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1065 = "tosa.tanh"(%1063) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %1066 = torch.aten.tanh %1064 : !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %1067 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1068 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1069 = "tosa.mul"(%1067, %1068) {shift = 0 : i32} : (tensor<f32>, tensor<f32>) -> tensor<f32>
  %1070 = "tosa.add"(%1065, %1069) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1071 = torch.aten.add.Scalar %1066, %float1.000000e00, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.float, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %1072 = "tosa.mul"(%1050, %1070) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %1073 = torch.aten.mul.Tensor %1051, %1071 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %1074 = "tosa.reshape"(%1072) {new_shape = [-1, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %1075 = torch.aten.view %1073, %419 : !torch.vtensor<[1,128,3072],f32>, !torch.list<int> -> !torch.vtensor<[128,3072],f32>
  %1076 = "tosa.reshape"(%1074) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %1077 = "tosa.reshape"(%61) {new_shape = [1, 3072, 768]} : (tensor<3072x768xf32>) -> tensor<1x3072x768xf32>
  %1078 = "tosa.matmul"(%1076, %1077) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32>
  %1079 = "tosa.reshape"(%1078) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_49 = tensor.cast %1079 : tensor<128x768xf32> to tensor<128x768xf32>
  %1080 = torch.aten.mm %1075, %62 : !torch.vtensor<[128,3072],f32>, !torch.vtensor<[3072,768],f32> -> !torch.vtensor<[128,768],f32>
  %1081 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1082 = "tosa.mul"(%63, %1081) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %1083 = torch.aten.mul.Scalar %64, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %1084 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1085 = "tosa.mul"(%cast_49, %1084) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %1086 = "tosa.add"(%1082, %1085) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %1087 = torch.aten.add.Tensor %1083, %1080, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %1088 = "tosa.reshape"(%1086) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1089 = torch.aten.view %1087, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %1090 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1091 = "tosa.mul"(%1088, %1090) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %1092 = "tosa.add"(%1012, %1091) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1093 = torch.aten.add.Tensor %1013, %1089, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1094 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %1095 = "tosa.reciprocal"(%1094) : (tensor<1xf32>) -> tensor<1xf32>
  %1096 = "tosa.reduce_sum"(%1092) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1097 = "tosa.reshape"(%1096) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1098 = "tosa.mul"(%1097, %1095) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1099 = "tosa.sub"(%1092, %1098) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1100 = "tosa.mul"(%1099, %1099) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1101 = "tosa.reduce_sum"(%1100) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1102 = "tosa.reshape"(%1101) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1103 = "tosa.mul"(%1102, %1095) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1104 = "tosa.reshape"(%59) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1105 = "tosa.reshape"(%57) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1106 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %1107 = "tosa.sub"(%1092, %1098) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1108 = "tosa.add"(%1103, %1106) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %1109 = "tosa.rsqrt"(%1108) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1110 = "tosa.mul"(%1107, %1109) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1111 = "tosa.mul"(%1110, %1104) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %1112 = "tosa.add"(%1111, %1105) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_50, %result1_51, %result2_52 = torch.aten.native_layer_norm %1093, %202, %60, %58, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %1113 = "tosa.reshape"(%1112) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %1114 = torch.aten.view %result0_50, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %1115 = "tosa.reshape"(%1113) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1116 = "tosa.reshape"(%53) {new_shape = [1, 768, 2304]} : (tensor<768x2304xf32>) -> tensor<1x768x2304xf32>
  %1117 = "tosa.matmul"(%1115, %1116) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32>
  %1118 = "tosa.reshape"(%1117) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32>
  %cast_53 = tensor.cast %1118 : tensor<128x2304xf32> to tensor<128x2304xf32>
  %1119 = torch.aten.mm %1114, %54 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,2304],f32> -> !torch.vtensor<[128,2304],f32>
  %1120 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1121 = "tosa.mul"(%55, %1120) {shift = 0 : i32} : (tensor<2304xf32>, tensor<f32>) -> tensor<2304xf32>
  %1122 = torch.aten.mul.Scalar %56, %int1 : !torch.vtensor<[2304],f32>, !torch.int -> !torch.vtensor<[2304],f32>
  %1123 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1124 = "tosa.mul"(%cast_53, %1123) {shift = 0 : i32} : (tensor<128x2304xf32>, tensor<f32>) -> tensor<128x2304xf32>
  %1125 = "tosa.add"(%1121, %1124) : (tensor<2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32>
  %1126 = torch.aten.add.Tensor %1122, %1119, %int1 : !torch.vtensor<[2304],f32>, !torch.vtensor<[128,2304],f32>, !torch.int -> !torch.vtensor<[128,2304],f32>
  %1127 = "tosa.reshape"(%1125) {new_shape = [1, 128, 2304]} : (tensor<128x2304xf32>) -> tensor<1x128x2304xf32>
  %1128 = torch.aten.view %1126, %237 : !torch.vtensor<[128,2304],f32>, !torch.list<int> -> !torch.vtensor<[1,128,2304],f32>
  %1129 = torch.aten.as_strided %1128, %240, %241, %int0 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1130 = builtin.unrealized_conversion_cast %1129 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %1131 = torch.aten.as_strided %1128, %240, %241, %int768 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1132 = builtin.unrealized_conversion_cast %1131 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %1133 = torch.aten.as_strided %1128, %240, %241, %int1536 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1134 = builtin.unrealized_conversion_cast %1133 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %1135 = "tosa.reshape"(%1130) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %1136 = torch.aten.view %1129, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %1137 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %1138 = "tosa.transpose"(%1135, %1137) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %1139 = torch.aten.permute %1136, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1140 = "tosa.reshape"(%1132) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %1141 = torch.aten.view %1131, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %1142 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %1143 = "tosa.transpose"(%1140, %1142) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %1144 = torch.aten.permute %1141, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1145 = "tosa.reshape"(%1134) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %1146 = torch.aten.view %1133, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %1147 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %1148 = "tosa.transpose"(%1145, %1147) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %1149 = torch.aten.permute %1146, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1150 = "tosa.const"() {value = dense<[0, 1, 3, 2]> : tensor<4xi32>} : () -> tensor<4xi32>
  %1151 = "tosa.transpose"(%1143, %1150) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32>
  %1152 = torch.aten.transpose.int %1144, %int-1, %int-2 : !torch.vtensor<[1,12,128,64],f32>, !torch.int, !torch.int -> !torch.vtensor<[1,12,64,128],f32>
  %1153 = torch.aten.broadcast_to %1139, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1154 = "tosa.reshape"(%1138) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %1155 = torch.aten.view %1139, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %1156 = torch.aten.broadcast_to %1152, %273 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,64,128],f32>
  %1157 = "tosa.reshape"(%1151) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32>
  %1158 = torch.aten.view %1152, %275 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[12,64,128],f32>
  %1159 = "tosa.matmul"(%1154, %1157) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32>
  %cast_54 = tensor.cast %1159 : tensor<12x128x128xf32> to tensor<12x128x128xf32>
  %1160 = torch.aten.bmm %1155, %1158 : !torch.vtensor<[12,128,64],f32>, !torch.vtensor<[12,64,128],f32> -> !torch.vtensor<[12,128,128],f32>
  %1161 = "tosa.reshape"(%cast_54) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %1162 = torch.aten.view %1160, %280 : !torch.vtensor<[12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %1163 = "tosa.cast"(%153) : (tensor<f32>) -> tensor<f32>
  %1164 = torch.aten.clone %154, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %1165 = "tosa.reciprocal"(%1163) : (tensor<f32>) -> tensor<f32>
  %1166 = "tosa.mul"(%1161, %1165) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %1167 = torch.aten.div.Tensor %1162, %1164 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %1168 = torch.aten.as_strided %152, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %1169 = torch.aten.as_strided %1168, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %1170 = torch.aten.as_strided %1169, %292, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,1024],ui8>
  %1171 = torch.aten.as_strided %1170, %294, %289, %int0 : !torch.vtensor<[1,1,128,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,128],ui8>
  %1172 = builtin.unrealized_conversion_cast %1171 : !torch.vtensor<[1,1,128,128],ui8> to tensor<1x1x128x128xi8>
  %1173 = "tosa.const"() {value = dense<0> : tensor<1x1x128x128xi8>} : () -> tensor<1x1x128x128xi8>
  %1174 = "tosa.equal"(%1172, %1173) : (tensor<1x1x128x128xi8>, tensor<1x1x128x128xi8>) -> tensor<1x1x128x128xi1>
  %1175 = "tosa.logical_not"(%1174) : (tensor<1x1x128x128xi1>) -> tensor<1x1x128x128xi1>
  %1176 = torch.aten.to.dtype %1171, %int11, %false, %false, %none : !torch.vtensor<[1,1,128,128],ui8>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[1,1,128,128],i1>
  %1177 = "tosa.cast"(%149) : (tensor<f32>) -> tensor<f32>
  %1178 = torch.aten.clone %150, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %1179 = "tosa.select"(%1175, %1166, %1177) : (tensor<1x1x128x128xi1>, tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %1180 = torch.aten.where.self %1176, %1167, %1178 : !torch.vtensor<[1,1,128,128],i1>, !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %1181 = "tosa.reduce_max"(%1179) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %1182 = "tosa.argmax"(%1179) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128xi64>
  %1183 = "tosa.reshape"(%1182) {new_shape = [1, 12, 128, 1]} : (tensor<1x12x128xi64>) -> tensor<1x12x128x1xi64>
  %values_55, %indices_56 = torch.aten.max.dim %1180, %int-1, %true : !torch.vtensor<[1,12,128,128],f32>, !torch.int, !torch.bool -> !torch.vtensor<[1,12,128,1],f32>, !torch.vtensor<[1,12,128,1],si64>
  %1184 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1185 = "tosa.mul"(%1181, %1184) {shift = 0 : i32} : (tensor<1x12x128x1xf32>, tensor<f32>) -> tensor<1x12x128x1xf32>
  %1186 = "tosa.sub"(%1179, %1185) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %1187 = torch.aten.sub.Tensor %1180, %values_55, %int1 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32>, !torch.int -> !torch.vtensor<[1,12,128,128],f32>
  %1188 = "tosa.exp"(%1186) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %1189 = torch.aten.exp %1187 : !torch.vtensor<[1,12,128,128],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %1190 = "tosa.reduce_sum"(%1188) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %1191 = torch.aten.sum.dim_IntList %1189, %305, %true, %none : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[1,12,128,1],f32>
  %1192 = "tosa.reciprocal"(%1190) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32>
  %1193 = "tosa.mul"(%1188, %1192) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %1194 = torch.aten.div.Tensor %1189, %1191 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %1195 = torch.aten.broadcast_to %1194, %280 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %1196 = "tosa.reshape"(%1193) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32>
  %1197 = torch.aten.view %1194, %321 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[12,128,128],f32>
  %1198 = torch.aten.broadcast_to %1149, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1199 = "tosa.reshape"(%1148) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %1200 = torch.aten.view %1149, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %1201 = "tosa.matmul"(%1196, %1199) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32>
  %cast_57 = tensor.cast %1201 : tensor<12x128x64xf32> to tensor<12x128x64xf32>
  %1202 = torch.aten.bmm %1197, %1200 : !torch.vtensor<[12,128,128],f32>, !torch.vtensor<[12,128,64],f32> -> !torch.vtensor<[12,128,64],f32>
  %1203 = "tosa.reshape"(%cast_57) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32>
  %1204 = torch.aten.view %1202, %268 : !torch.vtensor<[12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1205 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %1206 = "tosa.transpose"(%1203, %1205) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32>
  %1207 = torch.aten.permute %1204, %251 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %1208 = "tosa.cast"(%1206) : (tensor<1x128x12x64xf32>) -> tensor<1x128x12x64xf32>
  %1209 = torch.aten.clone %1207, %int0 : !torch.vtensor<[1,128,12,64],f32>, !torch.int -> !torch.vtensor<[1,128,12,64],f32>
  %1210 = "tosa.reshape"(%1208) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32>
  %1211 = torch.aten.view %1209, %240 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %1212 = "tosa.reshape"(%1210) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %1213 = torch.aten.view %1211, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %1214 = "tosa.reshape"(%1212) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1215 = "tosa.reshape"(%49) {new_shape = [1, 768, 768]} : (tensor<768x768xf32>) -> tensor<1x768x768xf32>
  %1216 = "tosa.matmul"(%1214, %1215) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32>
  %1217 = "tosa.reshape"(%1216) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_58 = tensor.cast %1217 : tensor<128x768xf32> to tensor<128x768xf32>
  %1218 = torch.aten.mm %1213, %50 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,768],f32> -> !torch.vtensor<[128,768],f32>
  %1219 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1220 = "tosa.mul"(%51, %1219) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %1221 = torch.aten.mul.Scalar %52, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %1222 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1223 = "tosa.mul"(%cast_58, %1222) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %1224 = "tosa.add"(%1220, %1223) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %1225 = torch.aten.add.Tensor %1221, %1218, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %1226 = "tosa.reshape"(%1224) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1227 = torch.aten.view %1225, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %1228 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1229 = "tosa.mul"(%1092, %1228) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %1230 = "tosa.add"(%1226, %1229) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1231 = torch.aten.add.Tensor %1227, %1093, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1232 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %1233 = "tosa.reciprocal"(%1232) : (tensor<1xf32>) -> tensor<1xf32>
  %1234 = "tosa.reduce_sum"(%1230) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1235 = "tosa.reshape"(%1234) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1236 = "tosa.mul"(%1235, %1233) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1237 = "tosa.sub"(%1230, %1236) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1238 = "tosa.mul"(%1237, %1237) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1239 = "tosa.reduce_sum"(%1238) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1240 = "tosa.reshape"(%1239) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1241 = "tosa.mul"(%1240, %1233) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1242 = "tosa.reshape"(%47) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1243 = "tosa.reshape"(%45) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1244 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %1245 = "tosa.sub"(%1230, %1236) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1246 = "tosa.add"(%1241, %1244) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %1247 = "tosa.rsqrt"(%1246) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1248 = "tosa.mul"(%1245, %1247) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1249 = "tosa.mul"(%1248, %1242) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %1250 = "tosa.add"(%1249, %1243) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_59, %result1_60, %result2_61 = torch.aten.native_layer_norm %1231, %202, %48, %46, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %1251 = "tosa.reshape"(%1250) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %1252 = torch.aten.view %result0_59, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %1253 = "tosa.reshape"(%1251) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1254 = "tosa.reshape"(%41) {new_shape = [1, 768, 3072]} : (tensor<768x3072xf32>) -> tensor<1x768x3072xf32>
  %1255 = "tosa.matmul"(%1253, %1254) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32>
  %1256 = "tosa.reshape"(%1255) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %cast_62 = tensor.cast %1256 : tensor<128x3072xf32> to tensor<128x3072xf32>
  %1257 = torch.aten.mm %1252, %42 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,3072],f32> -> !torch.vtensor<[128,3072],f32>
  %1258 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1259 = "tosa.mul"(%43, %1258) {shift = 0 : i32} : (tensor<3072xf32>, tensor<f32>) -> tensor<3072xf32>
  %1260 = torch.aten.mul.Scalar %44, %int1 : !torch.vtensor<[3072],f32>, !torch.int -> !torch.vtensor<[3072],f32>
  %1261 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1262 = "tosa.mul"(%cast_62, %1261) {shift = 0 : i32} : (tensor<128x3072xf32>, tensor<f32>) -> tensor<128x3072xf32>
  %1263 = "tosa.add"(%1259, %1262) : (tensor<3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32>
  %1264 = torch.aten.add.Tensor %1260, %1257, %int1 : !torch.vtensor<[3072],f32>, !torch.vtensor<[128,3072],f32>, !torch.int -> !torch.vtensor<[128,3072],f32>
  %1265 = "tosa.reshape"(%1263) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %1266 = torch.aten.view %1264, %391 : !torch.vtensor<[128,3072],f32>, !torch.list<int> -> !torch.vtensor<[1,128,3072],f32>
  %1267 = "tosa.const"() {value = dense<5.000000e-01> : tensor<f32>} : () -> tensor<f32>
  %1268 = "tosa.mul"(%1265, %1267) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1269 = torch.aten.mul.Scalar %1266, %float5.000000e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1270 = "tosa.const"() {value = dense<3.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1271 = "tosa.pow"(%1265, %1270) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1272 = torch.aten.pow.Tensor_Scalar %1266, %float3.000000e00 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1273 = "tosa.const"() {value = dense<4.471500e-02> : tensor<f32>} : () -> tensor<f32>
  %1274 = "tosa.mul"(%1271, %1273) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1275 = torch.aten.mul.Scalar %1272, %float4.471500e-02 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1276 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1277 = "tosa.mul"(%1274, %1276) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1278 = "tosa.add"(%1265, %1277) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %1279 = torch.aten.add.Tensor %1266, %1275, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32>, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %1280 = "tosa.const"() {value = dense<0.797884583> : tensor<f32>} : () -> tensor<f32>
  %1281 = "tosa.mul"(%1278, %1280) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1282 = torch.aten.mul.Scalar %1279, %float7.978850e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1283 = "tosa.tanh"(%1281) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %1284 = torch.aten.tanh %1282 : !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %1285 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1286 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1287 = "tosa.mul"(%1285, %1286) {shift = 0 : i32} : (tensor<f32>, tensor<f32>) -> tensor<f32>
  %1288 = "tosa.add"(%1283, %1287) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1289 = torch.aten.add.Scalar %1284, %float1.000000e00, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.float, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %1290 = "tosa.mul"(%1268, %1288) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %1291 = torch.aten.mul.Tensor %1269, %1289 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %1292 = "tosa.reshape"(%1290) {new_shape = [-1, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %1293 = torch.aten.view %1291, %419 : !torch.vtensor<[1,128,3072],f32>, !torch.list<int> -> !torch.vtensor<[128,3072],f32>
  %1294 = "tosa.reshape"(%1292) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %1295 = "tosa.reshape"(%37) {new_shape = [1, 3072, 768]} : (tensor<3072x768xf32>) -> tensor<1x3072x768xf32>
  %1296 = "tosa.matmul"(%1294, %1295) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32>
  %1297 = "tosa.reshape"(%1296) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_63 = tensor.cast %1297 : tensor<128x768xf32> to tensor<128x768xf32>
  %1298 = torch.aten.mm %1293, %38 : !torch.vtensor<[128,3072],f32>, !torch.vtensor<[3072,768],f32> -> !torch.vtensor<[128,768],f32>
  %1299 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1300 = "tosa.mul"(%39, %1299) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %1301 = torch.aten.mul.Scalar %40, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %1302 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1303 = "tosa.mul"(%cast_63, %1302) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %1304 = "tosa.add"(%1300, %1303) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %1305 = torch.aten.add.Tensor %1301, %1298, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %1306 = "tosa.reshape"(%1304) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1307 = torch.aten.view %1305, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %1308 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1309 = "tosa.mul"(%1306, %1308) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %1310 = "tosa.add"(%1230, %1309) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1311 = torch.aten.add.Tensor %1231, %1307, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1312 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %1313 = "tosa.reciprocal"(%1312) : (tensor<1xf32>) -> tensor<1xf32>
  %1314 = "tosa.reduce_sum"(%1310) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1315 = "tosa.reshape"(%1314) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1316 = "tosa.mul"(%1315, %1313) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1317 = "tosa.sub"(%1310, %1316) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1318 = "tosa.mul"(%1317, %1317) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1319 = "tosa.reduce_sum"(%1318) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1320 = "tosa.reshape"(%1319) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1321 = "tosa.mul"(%1320, %1313) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1322 = "tosa.reshape"(%35) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1323 = "tosa.reshape"(%33) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1324 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %1325 = "tosa.sub"(%1310, %1316) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1326 = "tosa.add"(%1321, %1324) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %1327 = "tosa.rsqrt"(%1326) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1328 = "tosa.mul"(%1325, %1327) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1329 = "tosa.mul"(%1328, %1322) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %1330 = "tosa.add"(%1329, %1323) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_64, %result1_65, %result2_66 = torch.aten.native_layer_norm %1311, %202, %36, %34, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %1331 = "tosa.reshape"(%1330) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %1332 = torch.aten.view %result0_64, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %1333 = "tosa.reshape"(%1331) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1334 = "tosa.reshape"(%29) {new_shape = [1, 768, 2304]} : (tensor<768x2304xf32>) -> tensor<1x768x2304xf32>
  %1335 = "tosa.matmul"(%1333, %1334) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32>
  %1336 = "tosa.reshape"(%1335) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32>
  %cast_67 = tensor.cast %1336 : tensor<128x2304xf32> to tensor<128x2304xf32>
  %1337 = torch.aten.mm %1332, %30 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,2304],f32> -> !torch.vtensor<[128,2304],f32>
  %1338 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1339 = "tosa.mul"(%31, %1338) {shift = 0 : i32} : (tensor<2304xf32>, tensor<f32>) -> tensor<2304xf32>
  %1340 = torch.aten.mul.Scalar %32, %int1 : !torch.vtensor<[2304],f32>, !torch.int -> !torch.vtensor<[2304],f32>
  %1341 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1342 = "tosa.mul"(%cast_67, %1341) {shift = 0 : i32} : (tensor<128x2304xf32>, tensor<f32>) -> tensor<128x2304xf32>
  %1343 = "tosa.add"(%1339, %1342) : (tensor<2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32>
  %1344 = torch.aten.add.Tensor %1340, %1337, %int1 : !torch.vtensor<[2304],f32>, !torch.vtensor<[128,2304],f32>, !torch.int -> !torch.vtensor<[128,2304],f32>
  %1345 = "tosa.reshape"(%1343) {new_shape = [1, 128, 2304]} : (tensor<128x2304xf32>) -> tensor<1x128x2304xf32>
  %1346 = torch.aten.view %1344, %237 : !torch.vtensor<[128,2304],f32>, !torch.list<int> -> !torch.vtensor<[1,128,2304],f32>
  %1347 = torch.aten.as_strided %1346, %240, %241, %int0 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1348 = builtin.unrealized_conversion_cast %1347 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %1349 = torch.aten.as_strided %1346, %240, %241, %int768 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1350 = builtin.unrealized_conversion_cast %1349 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %1351 = torch.aten.as_strided %1346, %240, %241, %int1536 : !torch.vtensor<[1,128,2304],f32>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1352 = builtin.unrealized_conversion_cast %1351 : !torch.vtensor<[1,128,768],f32> to tensor<1x128x768xf32>
  %1353 = "tosa.reshape"(%1348) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %1354 = torch.aten.view %1347, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %1355 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %1356 = "tosa.transpose"(%1353, %1355) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %1357 = torch.aten.permute %1354, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1358 = "tosa.reshape"(%1350) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %1359 = torch.aten.view %1349, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %1360 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %1361 = "tosa.transpose"(%1358, %1360) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %1362 = torch.aten.permute %1359, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1363 = "tosa.reshape"(%1352) {new_shape = [1, 128, 12, 64]} : (tensor<1x128x768xf32>) -> tensor<1x128x12x64xf32>
  %1364 = torch.aten.view %1351, %248 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %1365 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %1366 = "tosa.transpose"(%1363, %1365) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32>
  %1367 = torch.aten.permute %1364, %251 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1368 = "tosa.const"() {value = dense<[0, 1, 3, 2]> : tensor<4xi32>} : () -> tensor<4xi32>
  %1369 = "tosa.transpose"(%1361, %1368) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32>
  %1370 = torch.aten.transpose.int %1362, %int-1, %int-2 : !torch.vtensor<[1,12,128,64],f32>, !torch.int, !torch.int -> !torch.vtensor<[1,12,64,128],f32>
  %1371 = torch.aten.broadcast_to %1357, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1372 = "tosa.reshape"(%1356) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %1373 = torch.aten.view %1357, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %1374 = torch.aten.broadcast_to %1370, %273 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,64,128],f32>
  %1375 = "tosa.reshape"(%1369) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32>
  %1376 = torch.aten.view %1370, %275 : !torch.vtensor<[1,12,64,128],f32>, !torch.list<int> -> !torch.vtensor<[12,64,128],f32>
  %1377 = "tosa.matmul"(%1372, %1375) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32>
  %cast_68 = tensor.cast %1377 : tensor<12x128x128xf32> to tensor<12x128x128xf32>
  %1378 = torch.aten.bmm %1373, %1376 : !torch.vtensor<[12,128,64],f32>, !torch.vtensor<[12,64,128],f32> -> !torch.vtensor<[12,128,128],f32>
  %1379 = "tosa.reshape"(%cast_68) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %1380 = torch.aten.view %1378, %280 : !torch.vtensor<[12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %1381 = "tosa.cast"(%153) : (tensor<f32>) -> tensor<f32>
  %1382 = torch.aten.clone %154, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %1383 = "tosa.reciprocal"(%1381) : (tensor<f32>) -> tensor<f32>
  %1384 = "tosa.mul"(%1379, %1383) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %1385 = torch.aten.div.Tensor %1380, %1382 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %1386 = torch.aten.as_strided %152, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %1387 = torch.aten.as_strided %1386, %288, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,1024,1024],ui8>
  %1388 = torch.aten.as_strided %1387, %292, %289, %int0 : !torch.vtensor<[1,1,1024,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,1024],ui8>
  %1389 = torch.aten.as_strided %1388, %294, %289, %int0 : !torch.vtensor<[1,1,128,1024],ui8>, !torch.list<int>, !torch.list<int>, !torch.int -> !torch.vtensor<[1,1,128,128],ui8>
  %1390 = builtin.unrealized_conversion_cast %1389 : !torch.vtensor<[1,1,128,128],ui8> to tensor<1x1x128x128xi8>
  %1391 = "tosa.const"() {value = dense<0> : tensor<1x1x128x128xi8>} : () -> tensor<1x1x128x128xi8>
  %1392 = "tosa.equal"(%1390, %1391) : (tensor<1x1x128x128xi8>, tensor<1x1x128x128xi8>) -> tensor<1x1x128x128xi1>
  %1393 = "tosa.logical_not"(%1392) : (tensor<1x1x128x128xi1>) -> tensor<1x1x128x128xi1>
  %1394 = torch.aten.to.dtype %1389, %int11, %false, %false, %none : !torch.vtensor<[1,1,128,128],ui8>, !torch.int, !torch.bool, !torch.bool, !torch.none -> !torch.vtensor<[1,1,128,128],i1>
  %1395 = "tosa.cast"(%149) : (tensor<f32>) -> tensor<f32>
  %1396 = torch.aten.clone %150, %none : !torch.vtensor<[],f32>, !torch.none -> !torch.vtensor<[],f32>
  %1397 = "tosa.select"(%1393, %1384, %1395) : (tensor<1x1x128x128xi1>, tensor<1x12x128x128xf32>, tensor<f32>) -> tensor<1x12x128x128xf32>
  %1398 = torch.aten.where.self %1394, %1385, %1396 : !torch.vtensor<[1,1,128,128],i1>, !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %1399 = "tosa.reduce_max"(%1397) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %1400 = "tosa.argmax"(%1397) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128xi64>
  %1401 = "tosa.reshape"(%1400) {new_shape = [1, 12, 128, 1]} : (tensor<1x12x128xi64>) -> tensor<1x12x128x1xi64>
  %values_69, %indices_70 = torch.aten.max.dim %1398, %int-1, %true : !torch.vtensor<[1,12,128,128],f32>, !torch.int, !torch.bool -> !torch.vtensor<[1,12,128,1],f32>, !torch.vtensor<[1,12,128,1],si64>
  %1402 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1403 = "tosa.mul"(%1399, %1402) {shift = 0 : i32} : (tensor<1x12x128x1xf32>, tensor<f32>) -> tensor<1x12x128x1xf32>
  %1404 = "tosa.sub"(%1397, %1403) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %1405 = torch.aten.sub.Tensor %1398, %values_69, %int1 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32>, !torch.int -> !torch.vtensor<[1,12,128,128],f32>
  %1406 = "tosa.exp"(%1404) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32>
  %1407 = torch.aten.exp %1405 : !torch.vtensor<[1,12,128,128],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %1408 = "tosa.reduce_sum"(%1406) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32>
  %1409 = torch.aten.sum.dim_IntList %1407, %305, %true, %none : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[1,12,128,1],f32>
  %1410 = "tosa.reciprocal"(%1408) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32>
  %1411 = "tosa.mul"(%1406, %1410) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32>
  %1412 = torch.aten.div.Tensor %1407, %1409 : !torch.vtensor<[1,12,128,128],f32>, !torch.vtensor<[1,12,128,1],f32> -> !torch.vtensor<[1,12,128,128],f32>
  %1413 = torch.aten.broadcast_to %1412, %280 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,128],f32>
  %1414 = "tosa.reshape"(%1411) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32>
  %1415 = torch.aten.view %1412, %321 : !torch.vtensor<[1,12,128,128],f32>, !torch.list<int> -> !torch.vtensor<[12,128,128],f32>
  %1416 = torch.aten.broadcast_to %1367, %268 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1417 = "tosa.reshape"(%1366) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32>
  %1418 = torch.aten.view %1367, %270 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[12,128,64],f32>
  %1419 = "tosa.matmul"(%1414, %1417) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32>
  %cast_71 = tensor.cast %1419 : tensor<12x128x64xf32> to tensor<12x128x64xf32>
  %1420 = torch.aten.bmm %1415, %1418 : !torch.vtensor<[12,128,128],f32>, !torch.vtensor<[12,128,64],f32> -> !torch.vtensor<[12,128,64],f32>
  %1421 = "tosa.reshape"(%cast_71) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32>
  %1422 = torch.aten.view %1420, %268 : !torch.vtensor<[12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,12,128,64],f32>
  %1423 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64>
  %1424 = "tosa.transpose"(%1421, %1423) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32>
  %1425 = torch.aten.permute %1422, %251 : !torch.vtensor<[1,12,128,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,12,64],f32>
  %1426 = "tosa.cast"(%1424) : (tensor<1x128x12x64xf32>) -> tensor<1x128x12x64xf32>
  %1427 = torch.aten.clone %1425, %int0 : !torch.vtensor<[1,128,12,64],f32>, !torch.int -> !torch.vtensor<[1,128,12,64],f32>
  %1428 = "tosa.reshape"(%1426) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32>
  %1429 = torch.aten.view %1427, %240 : !torch.vtensor<[1,128,12,64],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %1430 = "tosa.reshape"(%1428) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %1431 = torch.aten.view %1429, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %1432 = "tosa.reshape"(%1430) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1433 = "tosa.reshape"(%25) {new_shape = [1, 768, 768]} : (tensor<768x768xf32>) -> tensor<1x768x768xf32>
  %1434 = "tosa.matmul"(%1432, %1433) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32>
  %1435 = "tosa.reshape"(%1434) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_72 = tensor.cast %1435 : tensor<128x768xf32> to tensor<128x768xf32>
  %1436 = torch.aten.mm %1431, %26 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,768],f32> -> !torch.vtensor<[128,768],f32>
  %1437 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1438 = "tosa.mul"(%27, %1437) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %1439 = torch.aten.mul.Scalar %28, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %1440 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1441 = "tosa.mul"(%cast_72, %1440) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %1442 = "tosa.add"(%1438, %1441) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %1443 = torch.aten.add.Tensor %1439, %1436, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %1444 = "tosa.reshape"(%1442) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1445 = torch.aten.view %1443, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %1446 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1447 = "tosa.mul"(%1310, %1446) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %1448 = "tosa.add"(%1444, %1447) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1449 = torch.aten.add.Tensor %1445, %1311, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1450 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %1451 = "tosa.reciprocal"(%1450) : (tensor<1xf32>) -> tensor<1xf32>
  %1452 = "tosa.reduce_sum"(%1448) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1453 = "tosa.reshape"(%1452) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1454 = "tosa.mul"(%1453, %1451) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1455 = "tosa.sub"(%1448, %1454) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1456 = "tosa.mul"(%1455, %1455) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1457 = "tosa.reduce_sum"(%1456) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1458 = "tosa.reshape"(%1457) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1459 = "tosa.mul"(%1458, %1451) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1460 = "tosa.reshape"(%23) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1461 = "tosa.reshape"(%21) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1462 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %1463 = "tosa.sub"(%1448, %1454) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1464 = "tosa.add"(%1459, %1462) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %1465 = "tosa.rsqrt"(%1464) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1466 = "tosa.mul"(%1463, %1465) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1467 = "tosa.mul"(%1466, %1460) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %1468 = "tosa.add"(%1467, %1461) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_73, %result1_74, %result2_75 = torch.aten.native_layer_norm %1449, %202, %24, %22, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %1469 = "tosa.reshape"(%1468) {new_shape = [-1, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %1470 = torch.aten.view %result0_73, %222 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %1471 = "tosa.reshape"(%1469) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1472 = "tosa.reshape"(%17) {new_shape = [1, 768, 3072]} : (tensor<768x3072xf32>) -> tensor<1x768x3072xf32>
  %1473 = "tosa.matmul"(%1471, %1472) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32>
  %1474 = "tosa.reshape"(%1473) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %cast_76 = tensor.cast %1474 : tensor<128x3072xf32> to tensor<128x3072xf32>
  %1475 = torch.aten.mm %1470, %18 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,3072],f32> -> !torch.vtensor<[128,3072],f32>
  %1476 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1477 = "tosa.mul"(%19, %1476) {shift = 0 : i32} : (tensor<3072xf32>, tensor<f32>) -> tensor<3072xf32>
  %1478 = torch.aten.mul.Scalar %20, %int1 : !torch.vtensor<[3072],f32>, !torch.int -> !torch.vtensor<[3072],f32>
  %1479 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1480 = "tosa.mul"(%cast_76, %1479) {shift = 0 : i32} : (tensor<128x3072xf32>, tensor<f32>) -> tensor<128x3072xf32>
  %1481 = "tosa.add"(%1477, %1480) : (tensor<3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32>
  %1482 = torch.aten.add.Tensor %1478, %1475, %int1 : !torch.vtensor<[3072],f32>, !torch.vtensor<[128,3072],f32>, !torch.int -> !torch.vtensor<[128,3072],f32>
  %1483 = "tosa.reshape"(%1481) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %1484 = torch.aten.view %1482, %391 : !torch.vtensor<[128,3072],f32>, !torch.list<int> -> !torch.vtensor<[1,128,3072],f32>
  %1485 = "tosa.const"() {value = dense<5.000000e-01> : tensor<f32>} : () -> tensor<f32>
  %1486 = "tosa.mul"(%1483, %1485) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1487 = torch.aten.mul.Scalar %1484, %float5.000000e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1488 = "tosa.const"() {value = dense<3.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1489 = "tosa.pow"(%1483, %1488) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1490 = torch.aten.pow.Tensor_Scalar %1484, %float3.000000e00 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1491 = "tosa.const"() {value = dense<4.471500e-02> : tensor<f32>} : () -> tensor<f32>
  %1492 = "tosa.mul"(%1489, %1491) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1493 = torch.aten.mul.Scalar %1490, %float4.471500e-02 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1494 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1495 = "tosa.mul"(%1492, %1494) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1496 = "tosa.add"(%1483, %1495) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %1497 = torch.aten.add.Tensor %1484, %1493, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32>, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %1498 = "tosa.const"() {value = dense<0.797884583> : tensor<f32>} : () -> tensor<f32>
  %1499 = "tosa.mul"(%1496, %1498) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1500 = torch.aten.mul.Scalar %1497, %float7.978850e-01 : !torch.vtensor<[1,128,3072],f32>, !torch.float -> !torch.vtensor<[1,128,3072],f32>
  %1501 = "tosa.tanh"(%1499) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %1502 = torch.aten.tanh %1500 : !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %1503 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1504 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1505 = "tosa.mul"(%1503, %1504) {shift = 0 : i32} : (tensor<f32>, tensor<f32>) -> tensor<f32>
  %1506 = "tosa.add"(%1501, %1505) : (tensor<1x128x3072xf32>, tensor<f32>) -> tensor<1x128x3072xf32>
  %1507 = torch.aten.add.Scalar %1502, %float1.000000e00, %int1 : !torch.vtensor<[1,128,3072],f32>, !torch.float, !torch.int -> !torch.vtensor<[1,128,3072],f32>
  %1508 = "tosa.mul"(%1486, %1506) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32>
  %1509 = torch.aten.mul.Tensor %1487, %1507 : !torch.vtensor<[1,128,3072],f32>, !torch.vtensor<[1,128,3072],f32> -> !torch.vtensor<[1,128,3072],f32>
  %1510 = "tosa.reshape"(%1508) {new_shape = [-1, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32>
  %1511 = torch.aten.view %1509, %419 : !torch.vtensor<[1,128,3072],f32>, !torch.list<int> -> !torch.vtensor<[128,3072],f32>
  %1512 = "tosa.reshape"(%1510) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32>
  %1513 = "tosa.reshape"(%13) {new_shape = [1, 3072, 768]} : (tensor<3072x768xf32>) -> tensor<1x3072x768xf32>
  %1514 = "tosa.matmul"(%1512, %1513) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32>
  %1515 = "tosa.reshape"(%1514) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %cast_77 = tensor.cast %1515 : tensor<128x768xf32> to tensor<128x768xf32>
  %1516 = torch.aten.mm %1511, %14 : !torch.vtensor<[128,3072],f32>, !torch.vtensor<[3072,768],f32> -> !torch.vtensor<[128,768],f32>
  %1517 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1518 = "tosa.mul"(%15, %1517) {shift = 0 : i32} : (tensor<768xf32>, tensor<f32>) -> tensor<768xf32>
  %1519 = torch.aten.mul.Scalar %16, %int1 : !torch.vtensor<[768],f32>, !torch.int -> !torch.vtensor<[768],f32>
  %1520 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1521 = "tosa.mul"(%cast_77, %1520) {shift = 0 : i32} : (tensor<128x768xf32>, tensor<f32>) -> tensor<128x768xf32>
  %1522 = "tosa.add"(%1518, %1521) : (tensor<768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32>
  %1523 = torch.aten.add.Tensor %1519, %1516, %int1 : !torch.vtensor<[768],f32>, !torch.vtensor<[128,768],f32>, !torch.int -> !torch.vtensor<[128,768],f32>
  %1524 = "tosa.reshape"(%1522) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1525 = torch.aten.view %1523, %240 : !torch.vtensor<[128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %1526 = "tosa.const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %1527 = "tosa.mul"(%1524, %1526) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<f32>) -> tensor<1x128x768xf32>
  %1528 = "tosa.add"(%1448, %1527) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1529 = torch.aten.add.Tensor %1449, %1525, %int1 : !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,768],f32>, !torch.int -> !torch.vtensor<[1,128,768],f32>
  %1530 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32>
  %1531 = "tosa.reciprocal"(%1530) : (tensor<1xf32>) -> tensor<1xf32>
  %1532 = "tosa.reduce_sum"(%1528) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1533 = "tosa.reshape"(%1532) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1534 = "tosa.mul"(%1533, %1531) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1535 = "tosa.sub"(%1528, %1534) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1536 = "tosa.mul"(%1535, %1535) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1537 = "tosa.reduce_sum"(%1536) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32>
  %1538 = "tosa.reshape"(%1537) {new_shape = [1, 128, 1]} : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1539 = "tosa.mul"(%1538, %1531) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1xf32>) -> tensor<1x128x1xf32>
  %1540 = "tosa.reshape"(%11) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1541 = "tosa.reshape"(%9) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32>
  %1542 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<f32>} : () -> tensor<f32>
  %1543 = "tosa.sub"(%1528, %1534) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1544 = "tosa.add"(%1539, %1542) : (tensor<1x128x1xf32>, tensor<f32>) -> tensor<1x128x1xf32>
  %1545 = "tosa.rsqrt"(%1544) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32>
  %1546 = "tosa.mul"(%1543, %1545) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32>
  %1547 = "tosa.mul"(%1546, %1540) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %1548 = "tosa.add"(%1547, %1541) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32>
  %result0_78, %result1_79, %result2_80 = torch.aten.native_layer_norm %1529, %202, %12, %10, %float1.000000e-05 : !torch.vtensor<[1,128,768],f32>, !torch.list<int>, !torch.vtensor<[768],f32>, !torch.vtensor<[768],f32>, !torch.float -> !torch.vtensor<[1,128,768],f32>, !torch.vtensor<[1,128,1],f32>, !torch.vtensor<[1,128,1],f32>
  %1549 = "tosa.reshape"(%1548) {new_shape = [1, 128, 768]} : (tensor<1x128x768xf32>) -> tensor<1x128x768xf32>
  %1550 = torch.aten.view %result0_78, %240 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[1,128,768],f32>
  %1551 = "tosa.const"() {value = dense<[1, 0]> : tensor<2xi32>} : () -> tensor<2xi32>
  %1552 = "tosa.transpose"(%7, %1551) : (tensor<2x768xf32>, tensor<2xi32>) -> tensor<768x2xf32>
  %1553 = torch.aten.transpose.int %8, %int0, %int1 : !torch.vtensor<[2,768],f32>, !torch.int, !torch.int -> !torch.vtensor<[768,2],f32>
  %1554 = torch.prim.ListConstruct %int128, %int768 : (!torch.int, !torch.int) -> !torch.list<int>
  %1555 = "tosa.reshape"(%1549) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32>
  %1556 = torch.aten.view %1550, %1554 : !torch.vtensor<[1,128,768],f32>, !torch.list<int> -> !torch.vtensor<[128,768],f32>
  %1557 = "tosa.reshape"(%1555) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32>
  %1558 = "tosa.reshape"(%1552) {new_shape = [1, 768, 2]} : (tensor<768x2xf32>) -> tensor<1x768x2xf32>
  %1559 = "tosa.matmul"(%1557, %1558) : (tensor<1x128x768xf32>, tensor<1x768x2xf32>) -> tensor<1x128x2xf32>
  %1560 = "tosa.reshape"(%1559) {new_shape = [128, 2]} : (tensor<1x128x2xf32>) -> tensor<128x2xf32>
  %cast_81 = tensor.cast %1560 : tensor<128x2xf32> to tensor<128x2xf32>
  %1561 = torch.aten.mm %1556, %1553 : !torch.vtensor<[128,768],f32>, !torch.vtensor<[768,2],f32> -> !torch.vtensor<[128,2],f32>
  %1562 = torch.prim.ListConstruct %int1, %int128, %int2 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
  %1563 = "tosa.reshape"(%cast_81) {new_shape = [1, 128, 2]} : (tensor<128x2xf32>) -> tensor<1x128x2xf32>
  %1564 = torch.aten.view %1561, %1562 : !torch.vtensor<[128,2],f32>, !torch.list<int> -> !torch.vtensor<[1,128,2],f32>
  %1565 = "tosa.const"() {value = dense<0> : tensor<1xi64>} : () -> tensor<1xi64>
  %1566 = "tosa.cast"(%1565) : (tensor<1xi64>) -> tensor<1xi64>
  %1567 = torch.aten.arange.start_step %int0, %int1, %int1, %none, %none, %cpu, %false : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[1],si64>
  %1568 = torch.aten.slice.Tensor %1564, %int1, %int-1, %int0, %int1 : !torch.vtensor<[1,128,2],f32>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,1,2],f32>
  %1569 = torch.aten.squeeze.dim %1568, %int1 : !torch.vtensor<[1,1,2],f32>, !torch.int -> !torch.vtensor<[1,2],f32>
  %1570 = torch.prim.ListConstruct %1567 : (!torch.vtensor<[1],si64>) -> !torch.list<vtensor>
  %1571 = torch.aten.index.Tensor %1569, %1570 : !torch.vtensor<[1,2],f32>, !torch.list<vtensor> -> !torch.vtensor<[1,2],f32>
  return %1571 : !torch.vtensor<[1,2],f32>
}

torch-mlir-opt -pass-pipeline='builtin.module(torch-backend-to-tosa-backend-pipeline)' /tmp/_lambda.mlir -mlir-print-ir-after-all -mlir-disable-threading --debug