Created
January 3, 2023 07:35
-
-
Save AmosLewis/f2d8fb5d121d38a18577aea5234b8d45 to your computer and use it in GitHub Desktop.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| func.func @forward(%arg0: tensor<1x128xi64>) -> tensor<1x2xf32> { | |
| %0 = "tosa.const"() {value = dense_resource<__elided__> : tensor<2x768xf32>} : () -> tensor<2x768xf32> | |
| %1 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768xf32>} : () -> tensor<768xf32> | |
| %2 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072x768xf32>} : () -> tensor<3072x768xf32> | |
| %3 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x3072xf32>} : () -> tensor<768x3072xf32> | |
| %4 = "tosa.const"() {value = dense_resource<__elided__> : tensor<3072xf32>} : () -> tensor<3072xf32> | |
| %5 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x768xf32>} : () -> tensor<768x768xf32> | |
| %6 = "tosa.const"() {value = dense_resource<__elided__> : tensor<768x2304xf32>} : () -> tensor<768x2304xf32> | |
| %7 = "tosa.const"() {value = dense_resource<__elided__> : tensor<2304xf32>} : () -> tensor<2304xf32> | |
| %8 = "tosa.const"() {value = dense<-3.40282347E+38> : tensor<f32>} : () -> tensor<f32> | |
| %9 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1x1x1024x1024xui8>} : () -> tensor<1x1x1024x1024xi8> | |
| %10 = "tosa.const"() {value = dense<8.000000e+00> : tensor<f32>} : () -> tensor<f32> | |
| %11 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1024x768xf32>} : () -> tensor<1024x768xf32> | |
| %12 = "tosa.const"() {value = dense_resource<__elided__> : tensor<50257x768xf32>} : () -> tensor<50257x768xf32> | |
| %13 = "tosa.const"() {value = dense<7.680000e+02> : tensor<1xf32>} : () -> tensor<1xf32> | |
| %14 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1x98304xi32>} : () -> tensor<1x98304xi32> | |
| %15 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1x98304xi32>} : () -> tensor<1x98304xi32> | |
| %16 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1x98304xi32>} : () -> tensor<1x98304xi32> | |
| %17 = "tosa.const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi64>} : () -> tensor<4xi64> | |
| %18 = "tosa.const"() {value = dense<[0, 1, 3, 2]> : tensor<4xi32>} : () -> tensor<4xi32> | |
| %19 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1x1048576xi32>} : () -> tensor<1x1048576xi32> | |
| %20 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1x131072xi32>} : () -> tensor<1x131072xi32> | |
| %21 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1x16384xi32>} : () -> tensor<1x16384xi32> | |
| %22 = "tosa.const"() {value = dense<0> : tensor<1x1x128x128xi8>} : () -> tensor<1x1x128x128xi8> | |
| %23 = "tosa.const"() {value = dense<[1, 0]> : tensor<2xi32>} : () -> tensor<2xi32> | |
| %24 = "tosa.const"() {value = dense<0> : tensor<1xi64>} : () -> tensor<1xi64> | |
| %25 = "tosa.const"() {value = dense<0.797884583> : tensor<1x1x1xf32>} : () -> tensor<1x1x1xf32> | |
| %26 = "tosa.const"() {value = dense<4.471500e-02> : tensor<1x1x1xf32>} : () -> tensor<1x1x1xf32> | |
| %27 = "tosa.const"() {value = dense<3.000000e+00> : tensor<1x1x1xf32>} : () -> tensor<1x1x1xf32> | |
| %28 = "tosa.const"() {value = dense<5.000000e-01> : tensor<1x1x1xf32>} : () -> tensor<1x1x1xf32> | |
| %29 = "tosa.const"() {value = dense<9.99999974E-6> : tensor<1x1x1xf32>} : () -> tensor<1x1x1xf32> | |
| %30 = "tosa.const"() {value = dense<1.000000e+00> : tensor<1x1x1xf32>} : () -> tensor<1x1x1xf32> | |
| %31 = "tosa.const"() {value = dense_resource<__elided__> : tensor<1x128xi64>} : () -> tensor<1x128xi64> | |
| %32 = torch_c.from_builtin_tensor %arg0 : tensor<1x128xi64> -> !torch.vtensor<[1,128],si64> | |
| %33 = torch_c.to_builtin_tensor %32 : !torch.vtensor<[1,128],si64> -> tensor<1x128xi64> | |
| %34 = "tosa.reshape"(%12) {new_shape = [1, 50257, 768]} : (tensor<50257x768xf32>) -> tensor<1x50257x768xf32> | |
| %35 = "tosa.cast"(%33) : (tensor<1x128xi64>) -> tensor<1x128xi32> | |
| %36 = "tosa.gather"(%34, %35) : (tensor<1x50257x768xf32>, tensor<1x128xi32>) -> tensor<1x128x768xf32> | |
| %37 = "tosa.reshape"(%11) {new_shape = [1, 1024, 768]} : (tensor<1024x768xf32>) -> tensor<1x1024x768xf32> | |
| %38 = "tosa.cast"(%31) : (tensor<1x128xi64>) -> tensor<1x128xi32> | |
| %39 = "tosa.gather"(%37, %38) : (tensor<1x1024x768xf32>, tensor<1x128xi32>) -> tensor<1x128x768xf32> | |
| %40 = "tosa.add"(%36, %39) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %41 = "tosa.reciprocal"(%13) : (tensor<1xf32>) -> tensor<1xf32> | |
| %42 = "tosa.reduce_sum"(%40) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %43 = "tosa.reshape"(%41) {new_shape = [1, 1, 1]} : (tensor<1xf32>) -> tensor<1x1x1xf32> | |
| %44 = "tosa.mul"(%42, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %45 = "tosa.sub"(%40, %44) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %46 = "tosa.mul"(%45, %45) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %47 = "tosa.reduce_sum"(%46) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %48 = "tosa.mul"(%47, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %49 = "tosa.reshape"(%1) {new_shape = [1, 1, 768]} : (tensor<768xf32>) -> tensor<1x1x768xf32> | |
| %50 = "tosa.add"(%48, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %51 = "tosa.rsqrt"(%50) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %52 = "tosa.mul"(%45, %51) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %53 = "tosa.mul"(%52, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %54 = "tosa.add"(%53, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %55 = "tosa.reshape"(%6) {new_shape = [1, 768, 2304]} : (tensor<768x2304xf32>) -> tensor<1x768x2304xf32> | |
| %56 = "tosa.matmul"(%54, %55) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32> | |
| %57 = "tosa.reshape"(%56) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32> | |
| %58 = "tosa.reshape"(%7) {new_shape = [1, 2304]} : (tensor<2304xf32>) -> tensor<1x2304xf32> | |
| %59 = "tosa.add"(%58, %57) : (tensor<1x2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32> | |
| %60 = "tosa.reshape"(%59) {new_shape = [1, 294912, 1]} : (tensor<128x2304xf32>) -> tensor<1x294912x1xf32> | |
| %61 = "tosa.gather"(%60, %14) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %62 = "tosa.gather"(%60, %15) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %63 = "tosa.gather"(%60, %16) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %64 = "tosa.reshape"(%61) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %65 = "tosa.transpose"(%64, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %66 = "tosa.reshape"(%62) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %67 = "tosa.transpose"(%66, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %68 = "tosa.reshape"(%63) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %69 = "tosa.transpose"(%68, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %70 = "tosa.transpose"(%67, %18) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32> | |
| %71 = "tosa.reshape"(%65) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %72 = "tosa.reshape"(%70) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32> | |
| %73 = "tosa.matmul"(%71, %72) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32> | |
| %74 = "tosa.reshape"(%73) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %75 = "tosa.reciprocal"(%10) : (tensor<f32>) -> tensor<f32> | |
| %76 = "tosa.reshape"(%75) {new_shape = [1, 1, 1, 1]} : (tensor<f32>) -> tensor<1x1x1x1xf32> | |
| %77 = "tosa.mul"(%74, %76) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x1x1x1xf32>) -> tensor<1x12x128x128xf32> | |
| %78 = "tosa.reshape"(%9) {new_shape = [1, 1048576, 1]} : (tensor<1x1x1024x1024xi8>) -> tensor<1x1048576x1xi8> | |
| %79 = "tosa.gather"(%78, %19) : (tensor<1x1048576x1xi8>, tensor<1x1048576xi32>) -> tensor<1x1048576x1xi8> | |
| %80 = "tosa.gather"(%79, %19) : (tensor<1x1048576x1xi8>, tensor<1x1048576xi32>) -> tensor<1x1048576x1xi8> | |
| %81 = "tosa.gather"(%80, %20) : (tensor<1x1048576x1xi8>, tensor<1x131072xi32>) -> tensor<1x131072x1xi8> | |
| %82 = "tosa.gather"(%81, %21) : (tensor<1x131072x1xi8>, tensor<1x16384xi32>) -> tensor<1x16384x1xi8> | |
| %83 = "tosa.reshape"(%82) {new_shape = [1, 1, 128, 128]} : (tensor<1x16384x1xi8>) -> tensor<1x1x128x128xi8> | |
| %84 = "tosa.equal"(%83, %22) : (tensor<1x1x128x128xi8>, tensor<1x1x128x128xi8>) -> tensor<1x1x128x128xi1> | |
| %85 = "tosa.select"(%84, %8, %77) : (tensor<1x1x128x128xi1>, tensor<f32>, tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %86 = "tosa.reduce_max"(%85) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %87 = "tosa.sub"(%85, %86) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %88 = "tosa.exp"(%87) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %89 = "tosa.reduce_sum"(%88) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %90 = "tosa.reciprocal"(%89) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32> | |
| %91 = "tosa.mul"(%88, %90) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %92 = "tosa.reshape"(%91) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32> | |
| %93 = "tosa.reshape"(%69) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %94 = "tosa.matmul"(%92, %93) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %95 = "tosa.reshape"(%94) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32> | |
| %96 = "tosa.transpose"(%95, %17) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32> | |
| %97 = "tosa.reshape"(%96) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32> | |
| %98 = "tosa.reshape"(%5) {new_shape = [1, 768, 768]} : (tensor<768x768xf32>) -> tensor<1x768x768xf32> | |
| %99 = "tosa.matmul"(%97, %98) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32> | |
| %100 = "tosa.reshape"(%99) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %101 = "tosa.reshape"(%1) {new_shape = [1, 768]} : (tensor<768xf32>) -> tensor<1x768xf32> | |
| %102 = "tosa.add"(%101, %100) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %103 = "tosa.reshape"(%102) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %104 = "tosa.add"(%103, %40) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %105 = "tosa.reduce_sum"(%104) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %106 = "tosa.mul"(%105, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %107 = "tosa.sub"(%104, %106) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %108 = "tosa.mul"(%107, %107) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %109 = "tosa.reduce_sum"(%108) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %110 = "tosa.mul"(%109, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %111 = "tosa.add"(%110, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %112 = "tosa.rsqrt"(%111) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %113 = "tosa.mul"(%107, %112) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %114 = "tosa.mul"(%113, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %115 = "tosa.add"(%114, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %116 = "tosa.reshape"(%3) {new_shape = [1, 768, 3072]} : (tensor<768x3072xf32>) -> tensor<1x768x3072xf32> | |
| %117 = "tosa.matmul"(%115, %116) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32> | |
| %118 = "tosa.reshape"(%117) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32> | |
| %119 = "tosa.reshape"(%4) {new_shape = [1, 3072]} : (tensor<3072xf32>) -> tensor<1x3072xf32> | |
| %120 = "tosa.add"(%119, %118) : (tensor<1x3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32> | |
| %121 = "tosa.reshape"(%120) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %122 = "tosa.mul"(%121, %28) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %123 = "tosa.pow"(%121, %27) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %124 = "tosa.mul"(%123, %26) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %125 = "tosa.add"(%121, %124) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %126 = "tosa.mul"(%125, %25) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %127 = "tosa.tanh"(%126) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %128 = "tosa.add"(%127, %30) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %129 = "tosa.mul"(%122, %128) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %130 = "tosa.reshape"(%2) {new_shape = [1, 3072, 768]} : (tensor<3072x768xf32>) -> tensor<1x3072x768xf32> | |
| %131 = "tosa.matmul"(%129, %130) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32> | |
| %132 = "tosa.reshape"(%131) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %133 = "tosa.add"(%101, %132) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %134 = "tosa.reshape"(%133) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %135 = "tosa.add"(%104, %134) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %136 = "tosa.reduce_sum"(%135) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %137 = "tosa.mul"(%136, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %138 = "tosa.sub"(%135, %137) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %139 = "tosa.mul"(%138, %138) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %140 = "tosa.reduce_sum"(%139) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %141 = "tosa.mul"(%140, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %142 = "tosa.add"(%141, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %143 = "tosa.rsqrt"(%142) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %144 = "tosa.mul"(%138, %143) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %145 = "tosa.mul"(%144, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %146 = "tosa.add"(%145, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %147 = "tosa.matmul"(%146, %55) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32> | |
| %148 = "tosa.reshape"(%147) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32> | |
| %149 = "tosa.add"(%58, %148) : (tensor<1x2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32> | |
| %150 = "tosa.reshape"(%149) {new_shape = [1, 294912, 1]} : (tensor<128x2304xf32>) -> tensor<1x294912x1xf32> | |
| %151 = "tosa.gather"(%150, %14) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %152 = "tosa.gather"(%150, %15) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %153 = "tosa.gather"(%150, %16) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %154 = "tosa.reshape"(%151) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %155 = "tosa.transpose"(%154, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %156 = "tosa.reshape"(%152) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %157 = "tosa.transpose"(%156, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %158 = "tosa.reshape"(%153) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %159 = "tosa.transpose"(%158, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %160 = "tosa.transpose"(%157, %18) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32> | |
| %161 = "tosa.reshape"(%155) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %162 = "tosa.reshape"(%160) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32> | |
| %163 = "tosa.matmul"(%161, %162) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32> | |
| %164 = "tosa.reshape"(%163) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %165 = "tosa.mul"(%164, %76) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x1x1x1xf32>) -> tensor<1x12x128x128xf32> | |
| %166 = "tosa.select"(%84, %8, %165) : (tensor<1x1x128x128xi1>, tensor<f32>, tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %167 = "tosa.reduce_max"(%166) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %168 = "tosa.sub"(%166, %167) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %169 = "tosa.exp"(%168) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %170 = "tosa.reduce_sum"(%169) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %171 = "tosa.reciprocal"(%170) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32> | |
| %172 = "tosa.mul"(%169, %171) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %173 = "tosa.reshape"(%172) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32> | |
| %174 = "tosa.reshape"(%159) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %175 = "tosa.matmul"(%173, %174) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %176 = "tosa.reshape"(%175) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32> | |
| %177 = "tosa.transpose"(%176, %17) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32> | |
| %178 = "tosa.reshape"(%177) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32> | |
| %179 = "tosa.matmul"(%178, %98) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32> | |
| %180 = "tosa.reshape"(%179) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %181 = "tosa.add"(%101, %180) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %182 = "tosa.reshape"(%181) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %183 = "tosa.add"(%182, %135) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %184 = "tosa.reduce_sum"(%183) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %185 = "tosa.mul"(%184, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %186 = "tosa.sub"(%183, %185) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %187 = "tosa.mul"(%186, %186) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %188 = "tosa.reduce_sum"(%187) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %189 = "tosa.mul"(%188, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %190 = "tosa.add"(%189, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %191 = "tosa.rsqrt"(%190) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %192 = "tosa.mul"(%186, %191) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %193 = "tosa.mul"(%192, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %194 = "tosa.add"(%193, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %195 = "tosa.matmul"(%194, %116) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32> | |
| %196 = "tosa.reshape"(%195) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32> | |
| %197 = "tosa.add"(%119, %196) : (tensor<1x3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32> | |
| %198 = "tosa.reshape"(%197) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %199 = "tosa.mul"(%198, %28) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %200 = "tosa.pow"(%198, %27) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %201 = "tosa.mul"(%200, %26) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %202 = "tosa.add"(%198, %201) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %203 = "tosa.mul"(%202, %25) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %204 = "tosa.tanh"(%203) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %205 = "tosa.add"(%204, %30) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %206 = "tosa.mul"(%199, %205) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %207 = "tosa.matmul"(%206, %130) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32> | |
| %208 = "tosa.reshape"(%207) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %209 = "tosa.add"(%101, %208) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %210 = "tosa.reshape"(%209) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %211 = "tosa.add"(%183, %210) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %212 = "tosa.reduce_sum"(%211) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %213 = "tosa.mul"(%212, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %214 = "tosa.sub"(%211, %213) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %215 = "tosa.mul"(%214, %214) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %216 = "tosa.reduce_sum"(%215) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %217 = "tosa.mul"(%216, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %218 = "tosa.add"(%217, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %219 = "tosa.rsqrt"(%218) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %220 = "tosa.mul"(%214, %219) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %221 = "tosa.mul"(%220, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %222 = "tosa.add"(%221, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %223 = "tosa.matmul"(%222, %55) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32> | |
| %224 = "tosa.reshape"(%223) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32> | |
| %225 = "tosa.add"(%58, %224) : (tensor<1x2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32> | |
| %226 = "tosa.reshape"(%225) {new_shape = [1, 294912, 1]} : (tensor<128x2304xf32>) -> tensor<1x294912x1xf32> | |
| %227 = "tosa.gather"(%226, %14) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %228 = "tosa.gather"(%226, %15) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %229 = "tosa.gather"(%226, %16) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %230 = "tosa.reshape"(%227) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %231 = "tosa.transpose"(%230, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %232 = "tosa.reshape"(%228) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %233 = "tosa.transpose"(%232, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %234 = "tosa.reshape"(%229) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %235 = "tosa.transpose"(%234, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %236 = "tosa.transpose"(%233, %18) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32> | |
| %237 = "tosa.reshape"(%231) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %238 = "tosa.reshape"(%236) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32> | |
| %239 = "tosa.matmul"(%237, %238) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32> | |
| %240 = "tosa.reshape"(%239) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %241 = "tosa.mul"(%240, %76) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x1x1x1xf32>) -> tensor<1x12x128x128xf32> | |
| %242 = "tosa.select"(%84, %8, %241) : (tensor<1x1x128x128xi1>, tensor<f32>, tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %243 = "tosa.reduce_max"(%242) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %244 = "tosa.sub"(%242, %243) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %245 = "tosa.exp"(%244) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %246 = "tosa.reduce_sum"(%245) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %247 = "tosa.reciprocal"(%246) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32> | |
| %248 = "tosa.mul"(%245, %247) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %249 = "tosa.reshape"(%248) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32> | |
| %250 = "tosa.reshape"(%235) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %251 = "tosa.matmul"(%249, %250) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %252 = "tosa.reshape"(%251) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32> | |
| %253 = "tosa.transpose"(%252, %17) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32> | |
| %254 = "tosa.reshape"(%253) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32> | |
| %255 = "tosa.matmul"(%254, %98) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32> | |
| %256 = "tosa.reshape"(%255) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %257 = "tosa.add"(%101, %256) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %258 = "tosa.reshape"(%257) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %259 = "tosa.add"(%258, %211) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %260 = "tosa.reduce_sum"(%259) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %261 = "tosa.mul"(%260, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %262 = "tosa.sub"(%259, %261) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %263 = "tosa.mul"(%262, %262) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %264 = "tosa.reduce_sum"(%263) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %265 = "tosa.mul"(%264, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %266 = "tosa.add"(%265, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %267 = "tosa.rsqrt"(%266) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %268 = "tosa.mul"(%262, %267) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %269 = "tosa.mul"(%268, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %270 = "tosa.add"(%269, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %271 = "tosa.matmul"(%270, %116) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32> | |
| %272 = "tosa.reshape"(%271) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32> | |
| %273 = "tosa.add"(%119, %272) : (tensor<1x3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32> | |
| %274 = "tosa.reshape"(%273) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %275 = "tosa.mul"(%274, %28) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %276 = "tosa.pow"(%274, %27) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %277 = "tosa.mul"(%276, %26) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %278 = "tosa.add"(%274, %277) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %279 = "tosa.mul"(%278, %25) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %280 = "tosa.tanh"(%279) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %281 = "tosa.add"(%280, %30) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %282 = "tosa.mul"(%275, %281) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %283 = "tosa.matmul"(%282, %130) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32> | |
| %284 = "tosa.reshape"(%283) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %285 = "tosa.add"(%101, %284) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %286 = "tosa.reshape"(%285) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %287 = "tosa.add"(%259, %286) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %288 = "tosa.reduce_sum"(%287) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %289 = "tosa.mul"(%288, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %290 = "tosa.sub"(%287, %289) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %291 = "tosa.mul"(%290, %290) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %292 = "tosa.reduce_sum"(%291) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %293 = "tosa.mul"(%292, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %294 = "tosa.add"(%293, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %295 = "tosa.rsqrt"(%294) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %296 = "tosa.mul"(%290, %295) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %297 = "tosa.mul"(%296, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %298 = "tosa.add"(%297, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %299 = "tosa.matmul"(%298, %55) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32> | |
| %300 = "tosa.reshape"(%299) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32> | |
| %301 = "tosa.add"(%58, %300) : (tensor<1x2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32> | |
| %302 = "tosa.reshape"(%301) {new_shape = [1, 294912, 1]} : (tensor<128x2304xf32>) -> tensor<1x294912x1xf32> | |
| %303 = "tosa.gather"(%302, %14) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %304 = "tosa.gather"(%302, %15) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %305 = "tosa.gather"(%302, %16) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %306 = "tosa.reshape"(%303) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %307 = "tosa.transpose"(%306, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %308 = "tosa.reshape"(%304) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %309 = "tosa.transpose"(%308, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %310 = "tosa.reshape"(%305) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %311 = "tosa.transpose"(%310, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %312 = "tosa.transpose"(%309, %18) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32> | |
| %313 = "tosa.reshape"(%307) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %314 = "tosa.reshape"(%312) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32> | |
| %315 = "tosa.matmul"(%313, %314) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32> | |
| %316 = "tosa.reshape"(%315) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %317 = "tosa.mul"(%316, %76) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x1x1x1xf32>) -> tensor<1x12x128x128xf32> | |
| %318 = "tosa.select"(%84, %8, %317) : (tensor<1x1x128x128xi1>, tensor<f32>, tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %319 = "tosa.reduce_max"(%318) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %320 = "tosa.sub"(%318, %319) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %321 = "tosa.exp"(%320) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %322 = "tosa.reduce_sum"(%321) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %323 = "tosa.reciprocal"(%322) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32> | |
| %324 = "tosa.mul"(%321, %323) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %325 = "tosa.reshape"(%324) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32> | |
| %326 = "tosa.reshape"(%311) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %327 = "tosa.matmul"(%325, %326) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %328 = "tosa.reshape"(%327) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32> | |
| %329 = "tosa.transpose"(%328, %17) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32> | |
| %330 = "tosa.reshape"(%329) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32> | |
| %331 = "tosa.matmul"(%330, %98) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32> | |
| %332 = "tosa.reshape"(%331) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %333 = "tosa.add"(%101, %332) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %334 = "tosa.reshape"(%333) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %335 = "tosa.add"(%334, %287) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %336 = "tosa.reduce_sum"(%335) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %337 = "tosa.mul"(%336, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %338 = "tosa.sub"(%335, %337) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %339 = "tosa.mul"(%338, %338) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %340 = "tosa.reduce_sum"(%339) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %341 = "tosa.mul"(%340, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %342 = "tosa.add"(%341, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %343 = "tosa.rsqrt"(%342) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %344 = "tosa.mul"(%338, %343) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %345 = "tosa.mul"(%344, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %346 = "tosa.add"(%345, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %347 = "tosa.matmul"(%346, %116) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32> | |
| %348 = "tosa.reshape"(%347) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32> | |
| %349 = "tosa.add"(%119, %348) : (tensor<1x3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32> | |
| %350 = "tosa.reshape"(%349) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %351 = "tosa.mul"(%350, %28) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %352 = "tosa.pow"(%350, %27) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %353 = "tosa.mul"(%352, %26) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %354 = "tosa.add"(%350, %353) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %355 = "tosa.mul"(%354, %25) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %356 = "tosa.tanh"(%355) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %357 = "tosa.add"(%356, %30) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %358 = "tosa.mul"(%351, %357) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %359 = "tosa.matmul"(%358, %130) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32> | |
| %360 = "tosa.reshape"(%359) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %361 = "tosa.add"(%101, %360) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %362 = "tosa.reshape"(%361) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %363 = "tosa.add"(%335, %362) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %364 = "tosa.reduce_sum"(%363) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %365 = "tosa.mul"(%364, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %366 = "tosa.sub"(%363, %365) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %367 = "tosa.mul"(%366, %366) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %368 = "tosa.reduce_sum"(%367) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %369 = "tosa.mul"(%368, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %370 = "tosa.add"(%369, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %371 = "tosa.rsqrt"(%370) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %372 = "tosa.mul"(%366, %371) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %373 = "tosa.mul"(%372, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %374 = "tosa.add"(%373, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %375 = "tosa.matmul"(%374, %55) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32> | |
| %376 = "tosa.reshape"(%375) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32> | |
| %377 = "tosa.add"(%58, %376) : (tensor<1x2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32> | |
| %378 = "tosa.reshape"(%377) {new_shape = [1, 294912, 1]} : (tensor<128x2304xf32>) -> tensor<1x294912x1xf32> | |
| %379 = "tosa.gather"(%378, %14) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %380 = "tosa.gather"(%378, %15) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %381 = "tosa.gather"(%378, %16) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %382 = "tosa.reshape"(%379) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %383 = "tosa.transpose"(%382, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %384 = "tosa.reshape"(%380) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %385 = "tosa.transpose"(%384, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %386 = "tosa.reshape"(%381) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %387 = "tosa.transpose"(%386, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %388 = "tosa.transpose"(%385, %18) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32> | |
| %389 = "tosa.reshape"(%383) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %390 = "tosa.reshape"(%388) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32> | |
| %391 = "tosa.matmul"(%389, %390) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32> | |
| %392 = "tosa.reshape"(%391) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %393 = "tosa.mul"(%392, %76) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x1x1x1xf32>) -> tensor<1x12x128x128xf32> | |
| %394 = "tosa.select"(%84, %8, %393) : (tensor<1x1x128x128xi1>, tensor<f32>, tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %395 = "tosa.reduce_max"(%394) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %396 = "tosa.sub"(%394, %395) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %397 = "tosa.exp"(%396) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %398 = "tosa.reduce_sum"(%397) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %399 = "tosa.reciprocal"(%398) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32> | |
| %400 = "tosa.mul"(%397, %399) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %401 = "tosa.reshape"(%400) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32> | |
| %402 = "tosa.reshape"(%387) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %403 = "tosa.matmul"(%401, %402) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %404 = "tosa.reshape"(%403) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32> | |
| %405 = "tosa.transpose"(%404, %17) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32> | |
| %406 = "tosa.reshape"(%405) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32> | |
| %407 = "tosa.matmul"(%406, %98) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32> | |
| %408 = "tosa.reshape"(%407) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %409 = "tosa.add"(%101, %408) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %410 = "tosa.reshape"(%409) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %411 = "tosa.add"(%410, %363) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %412 = "tosa.reduce_sum"(%411) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %413 = "tosa.mul"(%412, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %414 = "tosa.sub"(%411, %413) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %415 = "tosa.mul"(%414, %414) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %416 = "tosa.reduce_sum"(%415) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %417 = "tosa.mul"(%416, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %418 = "tosa.add"(%417, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %419 = "tosa.rsqrt"(%418) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %420 = "tosa.mul"(%414, %419) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %421 = "tosa.mul"(%420, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %422 = "tosa.add"(%421, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %423 = "tosa.matmul"(%422, %116) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32> | |
| %424 = "tosa.reshape"(%423) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32> | |
| %425 = "tosa.add"(%119, %424) : (tensor<1x3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32> | |
| %426 = "tosa.reshape"(%425) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %427 = "tosa.mul"(%426, %28) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %428 = "tosa.pow"(%426, %27) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %429 = "tosa.mul"(%428, %26) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %430 = "tosa.add"(%426, %429) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %431 = "tosa.mul"(%430, %25) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %432 = "tosa.tanh"(%431) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %433 = "tosa.add"(%432, %30) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %434 = "tosa.mul"(%427, %433) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %435 = "tosa.matmul"(%434, %130) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32> | |
| %436 = "tosa.reshape"(%435) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %437 = "tosa.add"(%101, %436) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %438 = "tosa.reshape"(%437) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %439 = "tosa.add"(%411, %438) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %440 = "tosa.reduce_sum"(%439) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %441 = "tosa.mul"(%440, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %442 = "tosa.sub"(%439, %441) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %443 = "tosa.mul"(%442, %442) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %444 = "tosa.reduce_sum"(%443) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %445 = "tosa.mul"(%444, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %446 = "tosa.add"(%445, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %447 = "tosa.rsqrt"(%446) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %448 = "tosa.mul"(%442, %447) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %449 = "tosa.mul"(%448, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %450 = "tosa.add"(%449, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %451 = "tosa.matmul"(%450, %55) : (tensor<1x128x768xf32>, tensor<1x768x2304xf32>) -> tensor<1x128x2304xf32> | |
| %452 = "tosa.reshape"(%451) {new_shape = [128, 2304]} : (tensor<1x128x2304xf32>) -> tensor<128x2304xf32> | |
| %453 = "tosa.add"(%58, %452) : (tensor<1x2304xf32>, tensor<128x2304xf32>) -> tensor<128x2304xf32> | |
| %454 = "tosa.reshape"(%453) {new_shape = [1, 294912, 1]} : (tensor<128x2304xf32>) -> tensor<1x294912x1xf32> | |
| %455 = "tosa.gather"(%454, %14) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %456 = "tosa.gather"(%454, %15) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %457 = "tosa.gather"(%454, %16) : (tensor<1x294912x1xf32>, tensor<1x98304xi32>) -> tensor<1x98304x1xf32> | |
| %458 = "tosa.reshape"(%455) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %459 = "tosa.transpose"(%458, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %460 = "tosa.reshape"(%456) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %461 = "tosa.transpose"(%460, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %462 = "tosa.reshape"(%457) {new_shape = [1, 128, 12, 64]} : (tensor<1x98304x1xf32>) -> tensor<1x128x12x64xf32> | |
| %463 = "tosa.transpose"(%462, %17) : (tensor<1x128x12x64xf32>, tensor<4xi64>) -> tensor<1x12x128x64xf32> | |
| %464 = "tosa.transpose"(%461, %18) : (tensor<1x12x128x64xf32>, tensor<4xi32>) -> tensor<1x12x64x128xf32> | |
| %465 = "tosa.reshape"(%459) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %466 = "tosa.reshape"(%464) {new_shape = [12, 64, 128]} : (tensor<1x12x64x128xf32>) -> tensor<12x64x128xf32> | |
| %467 = "tosa.matmul"(%465, %466) : (tensor<12x128x64xf32>, tensor<12x64x128xf32>) -> tensor<12x128x128xf32> | |
| %468 = "tosa.reshape"(%467) {new_shape = [1, 12, 128, 128]} : (tensor<12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %469 = "tosa.mul"(%468, %76) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x1x1x1xf32>) -> tensor<1x12x128x128xf32> | |
| %470 = "tosa.select"(%84, %8, %469) : (tensor<1x1x128x128xi1>, tensor<f32>, tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %471 = "tosa.reduce_max"(%470) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %472 = "tosa.sub"(%470, %471) : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %473 = "tosa.exp"(%472) : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x128xf32> | |
| %474 = "tosa.reduce_sum"(%473) {axis = 3 : i64} : (tensor<1x12x128x128xf32>) -> tensor<1x12x128x1xf32> | |
| %475 = "tosa.reciprocal"(%474) : (tensor<1x12x128x1xf32>) -> tensor<1x12x128x1xf32> | |
| %476 = "tosa.mul"(%473, %475) {shift = 0 : i32} : (tensor<1x12x128x128xf32>, tensor<1x12x128x1xf32>) -> tensor<1x12x128x128xf32> | |
| %477 = "tosa.reshape"(%476) {new_shape = [12, 128, 128]} : (tensor<1x12x128x128xf32>) -> tensor<12x128x128xf32> | |
| %478 = "tosa.reshape"(%463) {new_shape = [12, 128, 64]} : (tensor<1x12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %479 = "tosa.matmul"(%477, %478) : (tensor<12x128x128xf32>, tensor<12x128x64xf32>) -> tensor<12x128x64xf32> | |
| %480 = "tosa.reshape"(%479) {new_shape = [1, 12, 128, 64]} : (tensor<12x128x64xf32>) -> tensor<1x12x128x64xf32> | |
| %481 = "tosa.transpose"(%480, %17) : (tensor<1x12x128x64xf32>, tensor<4xi64>) -> tensor<1x128x12x64xf32> | |
| %482 = "tosa.reshape"(%481) {new_shape = [1, 128, 768]} : (tensor<1x128x12x64xf32>) -> tensor<1x128x768xf32> | |
| %483 = "tosa.matmul"(%482, %98) : (tensor<1x128x768xf32>, tensor<1x768x768xf32>) -> tensor<1x128x768xf32> | |
| %484 = "tosa.reshape"(%483) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %485 = "tosa.add"(%101, %484) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %486 = "tosa.reshape"(%485) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %487 = "tosa.add"(%486, %439) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %488 = "tosa.reduce_sum"(%487) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %489 = "tosa.mul"(%488, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %490 = "tosa.sub"(%487, %489) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %491 = "tosa.mul"(%490, %490) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %492 = "tosa.reduce_sum"(%491) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %493 = "tosa.mul"(%492, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %494 = "tosa.add"(%493, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %495 = "tosa.rsqrt"(%494) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %496 = "tosa.mul"(%490, %495) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %497 = "tosa.mul"(%496, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %498 = "tosa.add"(%497, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %499 = "tosa.matmul"(%498, %116) : (tensor<1x128x768xf32>, tensor<1x768x3072xf32>) -> tensor<1x128x3072xf32> | |
| %500 = "tosa.reshape"(%499) {new_shape = [128, 3072]} : (tensor<1x128x3072xf32>) -> tensor<128x3072xf32> | |
| %501 = "tosa.add"(%119, %500) : (tensor<1x3072xf32>, tensor<128x3072xf32>) -> tensor<128x3072xf32> | |
| %502 = "tosa.reshape"(%501) {new_shape = [1, 128, 3072]} : (tensor<128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %503 = "tosa.mul"(%502, %28) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %504 = "tosa.pow"(%502, %27) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %505 = "tosa.mul"(%504, %26) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %506 = "tosa.add"(%502, %505) : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %507 = "tosa.mul"(%506, %25) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %508 = "tosa.tanh"(%507) : (tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %509 = "tosa.add"(%508, %30) : (tensor<1x128x3072xf32>, tensor<1x1x1xf32>) -> tensor<1x128x3072xf32> | |
| %510 = "tosa.mul"(%503, %509) {shift = 0 : i32} : (tensor<1x128x3072xf32>, tensor<1x128x3072xf32>) -> tensor<1x128x3072xf32> | |
| %511 = "tosa.matmul"(%510, %130) : (tensor<1x128x3072xf32>, tensor<1x3072x768xf32>) -> tensor<1x128x768xf32> | |
| %512 = "tosa.reshape"(%511) {new_shape = [128, 768]} : (tensor<1x128x768xf32>) -> tensor<128x768xf32> | |
| %513 = "tosa.add"(%101, %512) : (tensor<1x768xf32>, tensor<128x768xf32>) -> tensor<128x768xf32> | |
| %514 = "tosa.reshape"(%513) {new_shape = [1, 128, 768]} : (tensor<128x768xf32>) -> tensor<1x128x768xf32> | |
| %515 = "tosa.add"(%487, %514) : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %516 = "tosa.reduce_sum"(%515) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %517 = "tosa.mul"(%516, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %518 = "tosa.sub"(%515, %517) : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %519 = "tosa.mul"(%518, %518) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x768xf32>) -> tensor<1x128x768xf32> | |
| %520 = "tosa.reduce_sum"(%519) {axis = 2 : i64} : (tensor<1x128x768xf32>) -> tensor<1x128x1xf32> | |
| %521 = "tosa.mul"(%520, %43) {shift = 0 : i32} : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %522 = "tosa.add"(%521, %29) : (tensor<1x128x1xf32>, tensor<1x1x1xf32>) -> tensor<1x128x1xf32> | |
| %523 = "tosa.rsqrt"(%522) : (tensor<1x128x1xf32>) -> tensor<1x128x1xf32> | |
| %524 = "tosa.mul"(%518, %523) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x128x1xf32>) -> tensor<1x128x768xf32> | |
| %525 = "tosa.mul"(%524, %49) {shift = 0 : i32} : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %526 = "tosa.add"(%525, %49) : (tensor<1x128x768xf32>, tensor<1x1x768xf32>) -> tensor<1x128x768xf32> | |
| %527 = "tosa.transpose"(%0, %23) : (tensor<2x768xf32>, tensor<2xi32>) -> tensor<768x2xf32> | |
| %528 = "tosa.reshape"(%527) {new_shape = [1, 768, 2]} : (tensor<768x2xf32>) -> tensor<1x768x2xf32> | |
| %529 = "tosa.matmul"(%526, %528) : (tensor<1x128x768xf32>, tensor<1x768x2xf32>) -> tensor<1x128x2xf32> | |
| %530 = torch_c.from_builtin_tensor %24 : tensor<1xi64> -> !torch.vtensor<[1],si64> | |
| %531 = "tosa.slice"(%529) {size = [1, 1, 2], start = [0, 127, 0]} : (tensor<1x128x2xf32>) -> tensor<1x1x2xf32> | |
| %532 = "tosa.reshape"(%531) {new_shape = [1, 2]} : (tensor<1x1x2xf32>) -> tensor<1x2xf32> | |
| %533 = torch_c.from_builtin_tensor %532 : tensor<1x2xf32> -> !torch.vtensor<[1,2],f32> | |
| %534 = torch.prim.ListConstruct %530 : (!torch.vtensor<[1],si64>) -> !torch.list<vtensor> | |
| %535 = torch.aten.index.Tensor %533, %534 : !torch.vtensor<[1,2],f32>, !torch.list<vtensor> -> !torch.vtensor<[1,2],f32> | |
| %536 = torch_c.to_builtin_tensor %535 : !torch.vtensor<[1,2],f32> -> tensor<1x2xf32> | |
| return %536 : tensor<1x2xf32> | |
| } |
Author
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
//===-------------------------------------------===//
Legalizing operation : 'torch.prim.ListConstruct'(0x93a7830) {
%534 = "torch.prim.ListConstruct"(%530) : (!torch.vtensor<[1],si64>) -> !torch.list
} -> FAILURE : unable to fold
} -> FAILURE : no matched legalization pattern
//===-------------------------------------------===//
<eval_with_key>.2:544:12: error: failed to legalize operation 'torch.prim.ListConstruct'
<eval_with_key>.2:544:12: note: see current operation: %534 = "torch.prim.ListConstruct"(%530) : (!torch.vtensor<[1],si64>) -> !torch.list
// -----// IR Dump After FinalizingBackendTypeConversion Failed (torch-finalizing-backend-type-conversion) //----- //
mlir-asm-printer: Verifying operation: func.func