Export to ONNX Error #10

ivder · 2019-10-29T08:36:09Z

Hi, thanks for the great work. Following my question here , I tried to convert to ONNX using this repo. But I got several errors.

By inputting this command just like your example, I got segmentation fault error :

sudo python3 onnx_export.py --model mobilenetv3_100 ./mobilenetv3_100.onnx
==> Creating PyTorch mobilenetv3_100 model
==> Exporting model to ONNX format at './mobilenetv3_100.onnx'
==> Loading and checking exported model from './mobilenetv3_100.onnx'
Segmentation fault

When I tried with efficientnet_b0 using checkpoint and not using checkpoint
sudo python3 onnx_export.py --model efficientnet_b0 ./efficientnet.onnx
or
sudo python3 onnx_export.py --model efficientnet_b0 --checkpoint ../train/model_best.pth.tar --num-classes 30 ./efficientnet.onnx

I got Couldn't export Python operator SwishAutoFn error.

==> Creating PyTorch efficientnet_b0 model
=> Loading checkpoint '../train/20191015-Deepeye36k-efficientnet_b0-224/model_best.pth.tar'
=> Loaded checkpoint '../train/20191015-Deepeye36k-efficientnet_b0-224/model_best.pth.tar'
==> Exporting model to ONNX format at './efficientnet.onnx'
Traceback (most recent call last):
  File "onnx_export.py", line 75, in <module>
    main()
  File "onnx_export.py", line 59, in main
    input_names=input_names, output_names=output_names)
  File "/home/ivan/.local/lib/python3.6/site-packages/torch/onnx/__init__.py", line 26, in _export
    result = utils._export(*args, **kwargs)
  File "/home/ivan/.local/lib/python3.6/site-packages/torch/onnx/utils.py", line 394, in _export
    operator_export_type, strip_doc_string, val_keep_init_as_ip)
RuntimeError: ONNX export failed: Couldn't export Python operator SwishAutoFn

Any help would be appreciated, thanks

The text was updated successfully, but these errors were encountered:

ivder · 2019-10-29T08:40:52Z

Full traceback :

==> Creating PyTorch efficientnet_b0 model
==> Exporting model to ONNX format at './efficientnet.onnx'
Traceback (most recent call last):
  File "onnx_export.py", line 75, in <module>
    main()
  File "onnx_export.py", line 59, in main
    input_names=input_names, output_names=output_names)
  File "/home/ivan/.local/lib/python3.6/site-packages/torch/onnx/__init__.py", line 26, in _export
    result = utils._export(*args, **kwargs)
  File "/home/ivan/.local/lib/python3.6/site-packages/torch/onnx/utils.py", line 394, in _export
    operator_export_type, strip_doc_string, val_keep_init_as_ip)
RuntimeError: ONNX export failed: Couldn't export Python operator SwishAutoFn

Defined at:
/ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py(30): swish
/ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py(104): forward
/home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/module.py(525): _slow_forward
/home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/module.py(539): __call__
/ssd/pytorch/gen-efficientnet-pytorch/geffnet/gen_efficientnet.py(155): features
/ssd/pytorch/gen-efficientnet-pytorch/geffnet/gen_efficientnet.py(171): forward
/home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/module.py(525): _slow_forward
/home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/module.py(539): __call__
/home/ivan/.local/lib/python3.6/site-packages/torch/jit/__init__.py(352): forward
/home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/module.py(541): __call__
/home/ivan/.local/lib/python3.6/site-packages/torch/jit/__init__.py(275): get_trace_graph
/home/ivan/.local/lib/python3.6/site-packages/torch/onnx/utils.py(206): _trace_and_get_graph_from_model
/home/ivan/.local/lib/python3.6/site-packages/torch/onnx/utils.py(249): _model_to_graph
/home/ivan/.local/lib/python3.6/site-packages/torch/onnx/utils.py(382): _export
/home/ivan/.local/lib/python3.6/site-packages/torch/onnx/__init__.py(26): _export
onnx_export.py(59): main
onnx_export.py(75): <module>


Graph we tried to export:
graph(%input0 : Float(1, 3, 224, 224),
      %1 : Float(32, 3, 3, 3),
      %2 : Float(32),
      %3 : Float(32),
      %4 : Float(32),
      %5 : Float(32),
      %6 : Long(),
      %7 : Float(32, 1, 3, 3),
      %8 : Float(32),
      %9 : Float(32),
      %10 : Float(32),
      %11 : Float(32),
      %12 : Long(),
      %13 : Float(8, 32, 1, 1),
      %14 : Float(8),
      %15 : Float(32, 8, 1, 1),
      %16 : Float(32),
      %17 : Float(16, 32, 1, 1),
      %18 : Float(16),
      %19 : Float(16),
      %20 : Float(16),
      %21 : Float(16),
      %22 : Long(),
      %23 : Float(96, 16, 1, 1),
      %24 : Float(96),
      %25 : Float(96),
      %26 : Float(96),
      %27 : Float(96),
      %28 : Long(),
      %29 : Float(96, 1, 3, 3),
      %30 : Float(96),
      %31 : Float(96),
      %32 : Float(96),
      %33 : Float(96),
      %34 : Long(),
      %35 : Float(4, 96, 1, 1),
      %36 : Float(4),
      %37 : Float(96, 4, 1, 1),
      %38 : Float(96),
      %39 : Float(24, 96, 1, 1),
      %40 : Float(24),
      %41 : Float(24),
      %42 : Float(24),
      %43 : Float(24),
      %44 : Long(),
      %45 : Float(144, 24, 1, 1),
      %46 : Float(144),
      %47 : Float(144),
      %48 : Float(144),
      %49 : Float(144),
      %50 : Long(),
      %51 : Float(144, 1, 3, 3),
      %52 : Float(144),
      %53 : Float(144),
      %54 : Float(144),
      %55 : Float(144),
      %56 : Long(),
      %57 : Float(6, 144, 1, 1),
      %58 : Float(6),
      %59 : Float(144, 6, 1, 1),
      %60 : Float(144),
      %61 : Float(24, 144, 1, 1),
      %62 : Float(24),
      %63 : Float(24),
      %64 : Float(24),
      %65 : Float(24),
      %66 : Long(),
      %67 : Float(144, 24, 1, 1),
      %68 : Float(144),
      %69 : Float(144),
      %70 : Float(144),
      %71 : Float(144),
      %72 : Long(),
      %73 : Float(144, 1, 5, 5),
      %74 : Float(144),
      %75 : Float(144),
      %76 : Float(144),
      %77 : Float(144),
      %78 : Long(),
      %79 : Float(6, 144, 1, 1),
      %80 : Float(6),
      %81 : Float(144, 6, 1, 1),
      %82 : Float(144),
      %83 : Float(40, 144, 1, 1),
      %84 : Float(40),
      %85 : Float(40),
      %86 : Float(40),
      %87 : Float(40),
      %88 : Long(),
      %89 : Float(240, 40, 1, 1),
      %90 : Float(240),
      %91 : Float(240),
      %92 : Float(240),
      %93 : Float(240),
      %94 : Long(),
      %95 : Float(240, 1, 5, 5),
      %96 : Float(240),
      %97 : Float(240),
      %98 : Float(240),
      %99 : Float(240),
      %100 : Long(),
      %101 : Float(10, 240, 1, 1),
      %102 : Float(10),
      %103 : Float(240, 10, 1, 1),
      %104 : Float(240),
      %105 : Float(40, 240, 1, 1),
      %106 : Float(40),
      %107 : Float(40),
      %108 : Float(40),
      %109 : Float(40),
      %110 : Long(),
      %111 : Float(240, 40, 1, 1),
      %112 : Float(240),
      %113 : Float(240),
      %114 : Float(240),
      %115 : Float(240),
      %116 : Long(),
      %117 : Float(240, 1, 3, 3),
      %118 : Float(240),
      %119 : Float(240),
      %120 : Float(240),
      %121 : Float(240),
      %122 : Long(),
      %123 : Float(10, 240, 1, 1),
      %124 : Float(10),
      %125 : Float(240, 10, 1, 1),
      %126 : Float(240),
      %127 : Float(80, 240, 1, 1),
      %128 : Float(80),
      %129 : Float(80),
      %130 : Float(80),
      %131 : Float(80),
      %132 : Long(),
      %133 : Float(480, 80, 1, 1),
      %134 : Float(480),
      %135 : Float(480),
      %136 : Float(480),
      %137 : Float(480),
      %138 : Long(),
      %139 : Float(480, 1, 3, 3),
      %140 : Float(480),
      %141 : Float(480),
      %142 : Float(480),
      %143 : Float(480),
      %144 : Long(),
      %145 : Float(20, 480, 1, 1),
      %146 : Float(20),
      %147 : Float(480, 20, 1, 1),
      %148 : Float(480),
      %149 : Float(80, 480, 1, 1),
      %150 : Float(80),
      %151 : Float(80),
      %152 : Float(80),
      %153 : Float(80),
      %154 : Long(),
      %155 : Float(480, 80, 1, 1),
      %156 : Float(480),
      %157 : Float(480),
      %158 : Float(480),
      %159 : Float(480),
      %160 : Long(),
      %161 : Float(480, 1, 3, 3),
      %162 : Float(480),
      %163 : Float(480),
      %164 : Float(480),
      %165 : Float(480),
      %166 : Long(),
      %167 : Float(20, 480, 1, 1),
      %168 : Float(20),
      %169 : Float(480, 20, 1, 1),
      %170 : Float(480),
      %171 : Float(80, 480, 1, 1),
      %172 : Float(80),
      %173 : Float(80),
      %174 : Float(80),
      %175 : Float(80),
      %176 : Long(),
      %177 : Float(480, 80, 1, 1),
      %178 : Float(480),
      %179 : Float(480),
      %180 : Float(480),
      %181 : Float(480),
      %182 : Long(),
      %183 : Float(480, 1, 5, 5),
      %184 : Float(480),
      %185 : Float(480),
      %186 : Float(480),
      %187 : Float(480),
      %188 : Long(),
      %189 : Float(20, 480, 1, 1),
      %190 : Float(20),
      %191 : Float(480, 20, 1, 1),
      %192 : Float(480),
      %193 : Float(112, 480, 1, 1),
      %194 : Float(112),
      %195 : Float(112),
      %196 : Float(112),
      %197 : Float(112),
      %198 : Long(),
      %199 : Float(672, 112, 1, 1),
      %200 : Float(672),
      %201 : Float(672),
      %202 : Float(672),
      %203 : Float(672),
      %204 : Long(),
      %205 : Float(672, 1, 5, 5),
      %206 : Float(672),
      %207 : Float(672),
      %208 : Float(672),
      %209 : Float(672),
      %210 : Long(),
      %211 : Float(28, 672, 1, 1),
      %212 : Float(28),
      %213 : Float(672, 28, 1, 1),
      %214 : Float(672),
      %215 : Float(112, 672, 1, 1),
      %216 : Float(112),
      %217 : Float(112),
      %218 : Float(112),
      %219 : Float(112),
      %220 : Long(),
      %221 : Float(672, 112, 1, 1),
      %222 : Float(672),
      %223 : Float(672),
      %224 : Float(672),
      %225 : Float(672),
      %226 : Long(),
      %227 : Float(672, 1, 5, 5),
      %228 : Float(672),
      %229 : Float(672),
      %230 : Float(672),
      %231 : Float(672),
      %232 : Long(),
      %233 : Float(28, 672, 1, 1),
      %234 : Float(28),
      %235 : Float(672, 28, 1, 1),
      %236 : Float(672),
      %237 : Float(112, 672, 1, 1),
      %238 : Float(112),
      %239 : Float(112),
      %240 : Float(112),
      %241 : Float(112),
      %242 : Long(),
      %243 : Float(672, 112, 1, 1),
      %244 : Float(672),
      %245 : Float(672),
      %246 : Float(672),
      %247 : Float(672),
      %248 : Long(),
      %249 : Float(672, 1, 5, 5),
      %250 : Float(672),
      %251 : Float(672),
      %252 : Float(672),
      %253 : Float(672),
      %254 : Long(),
      %255 : Float(28, 672, 1, 1),
      %256 : Float(28),
      %257 : Float(672, 28, 1, 1),
      %258 : Float(672),
      %259 : Float(192, 672, 1, 1),
      %260 : Float(192),
      %261 : Float(192),
      %262 : Float(192),
      %263 : Float(192),
      %264 : Long(),
      %265 : Float(1152, 192, 1, 1),
      %266 : Float(1152),
      %267 : Float(1152),
      %268 : Float(1152),
      %269 : Float(1152),
      %270 : Long(),
      %271 : Float(1152, 1, 5, 5),
      %272 : Float(1152),
      %273 : Float(1152),
      %274 : Float(1152),
      %275 : Float(1152),
      %276 : Long(),
      %277 : Float(48, 1152, 1, 1),
      %278 : Float(48),
      %279 : Float(1152, 48, 1, 1),
      %280 : Float(1152),
      %281 : Float(192, 1152, 1, 1),
      %282 : Float(192),
      %283 : Float(192),
      %284 : Float(192),
      %285 : Float(192),
      %286 : Long(),
      %287 : Float(1152, 192, 1, 1),
      %288 : Float(1152),
      %289 : Float(1152),
      %290 : Float(1152),
      %291 : Float(1152),
      %292 : Long(),
      %293 : Float(1152, 1, 5, 5),
      %294 : Float(1152),
      %295 : Float(1152),
      %296 : Float(1152),
      %297 : Float(1152),
      %298 : Long(),
      %299 : Float(48, 1152, 1, 1),
      %300 : Float(48),
      %301 : Float(1152, 48, 1, 1),
      %302 : Float(1152),
      %303 : Float(192, 1152, 1, 1),
      %304 : Float(192),
      %305 : Float(192),
      %306 : Float(192),
      %307 : Float(192),
      %308 : Long(),
      %309 : Float(1152, 192, 1, 1),
      %310 : Float(1152),
      %311 : Float(1152),
      %312 : Float(1152),
      %313 : Float(1152),
      %314 : Long(),
      %315 : Float(1152, 1, 5, 5),
      %316 : Float(1152),
      %317 : Float(1152),
      %318 : Float(1152),
      %319 : Float(1152),
      %320 : Long(),
      %321 : Float(48, 1152, 1, 1),
      %322 : Float(48),
      %323 : Float(1152, 48, 1, 1),
      %324 : Float(1152),
      %325 : Float(192, 1152, 1, 1),
      %326 : Float(192),
      %327 : Float(192),
      %328 : Float(192),
      %329 : Float(192),
      %330 : Long(),
      %331 : Float(1152, 192, 1, 1),
      %332 : Float(1152),
      %333 : Float(1152),
      %334 : Float(1152),
      %335 : Float(1152),
      %336 : Long(),
      %337 : Float(1152, 1, 3, 3),
      %338 : Float(1152),
      %339 : Float(1152),
      %340 : Float(1152),
      %341 : Float(1152),
      %342 : Long(),
      %343 : Float(48, 1152, 1, 1),
      %344 : Float(48),
      %345 : Float(1152, 48, 1, 1),
      %346 : Float(1152),
      %347 : Float(320, 1152, 1, 1),
      %348 : Float(320),
      %349 : Float(320),
      %350 : Float(320),
      %351 : Float(320),
      %352 : Long(),
      %353 : Float(1280, 320, 1, 1),
      %354 : Float(1280),
      %355 : Float(1280),
      %356 : Float(1280),
      %357 : Float(1280),
      %358 : Long(),
      %359 : Float(1000, 1280),
      %360 : Float(1000)):
  %361 : Float(1, 32, 112, 112) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%input0, %1), scope: GenEfficientNet/Conv2d[conv_stem] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %362 : Float(1, 32, 112, 112) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%361, %2, %3, %4, %5), scope: GenEfficientNet/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.3 : Float(1, 32, 112, 112) = ^SwishAutoFn()(%362), scope: GenEfficientNet/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %364 : Float(1, 32, 112, 112) = onnx::Conv[dilations=[1, 1], group=32, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%input.3, %7), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %365 : Float(1, 32, 112, 112) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%364, %8, %9, %10, %11), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.5 : Float(1, 32, 112, 112) = ^SwishAutoFn()(%365), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %367 : Float(1, 32, 1, 1) = onnx::GlobalAveragePool(%input.5), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %368 : Float(1, 8, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%367, %13, %14), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.7 : Float(1, 8, 1, 1) = ^SwishAutoFn()(%368), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %370 : Float(1, 32, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.7, %15, %16), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %371 : Float(1, 32, 1, 1) = onnx::Sigmoid(%370), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %372 : Float(1, 32, 112, 112) = onnx::Mul(%input.5, %371), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %373 : Float(1, 16, 112, 112) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%372, %17), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %374 : Float(1, 16, 112, 112) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%373, %18, %19, %20, %21), scope: GenEfficientNet/Sequential[blocks]/Sequential[0]/DepthwiseSeparableConv[0]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %375 : Float(1, 96, 112, 112) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%374, %23), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %376 : Float(1, 96, 112, 112) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%375, %24, %25, %26, %27), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.12 : Float(1, 96, 112, 112) = ^SwishAutoFn()(%376), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %378 : Float(1, 96, 56, 56) = onnx::Conv[dilations=[1, 1], group=96, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%input.12, %29), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %379 : Float(1, 96, 56, 56) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%378, %30, %31, %32, %33), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.14 : Float(1, 96, 56, 56) = ^SwishAutoFn()(%379), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %381 : Float(1, 96, 1, 1) = onnx::GlobalAveragePool(%input.14), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %382 : Float(1, 4, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%381, %35, %36), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.16 : Float(1, 4, 1, 1) = ^SwishAutoFn()(%382), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %384 : Float(1, 96, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.16, %37, %38), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %385 : Float(1, 96, 1, 1) = onnx::Sigmoid(%384), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %386 : Float(1, 96, 56, 56) = onnx::Mul(%input.14, %385), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %387 : Float(1, 24, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%386, %39), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %388 : Float(1, 24, 56, 56) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%387, %40, %41, %42, %43), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[0]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %389 : Float(1, 144, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%388, %45), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %390 : Float(1, 144, 56, 56) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%389, %46, %47, %48, %49), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.21 : Float(1, 144, 56, 56) = ^SwishAutoFn()(%390), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %392 : Float(1, 144, 56, 56) = onnx::Conv[dilations=[1, 1], group=144, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%input.21, %51), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %393 : Float(1, 144, 56, 56) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%392, %52, %53, %54, %55), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.23 : Float(1, 144, 56, 56) = ^SwishAutoFn()(%393), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %395 : Float(1, 144, 1, 1) = onnx::GlobalAveragePool(%input.23), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %396 : Float(1, 6, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%395, %57, %58), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.25 : Float(1, 6, 1, 1) = ^SwishAutoFn()(%396), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %398 : Float(1, 144, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.25, %59, %60), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %399 : Float(1, 144, 1, 1) = onnx::Sigmoid(%398), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %400 : Float(1, 144, 56, 56) = onnx::Mul(%input.23, %399), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %401 : Float(1, 24, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%400, %61), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %402 : Float(1, 24, 56, 56) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%401, %62, %63, %64, %65), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %403 : Float(1, 24, 56, 56) = onnx::Add(%402, %388), scope: GenEfficientNet/Sequential[blocks]/Sequential[1]/InvertedResidual[1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:230:0
  %404 : Float(1, 144, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%403, %67), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %405 : Float(1, 144, 56, 56) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%404, %68, %69, %70, %71), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.30 : Float(1, 144, 56, 56) = ^SwishAutoFn()(%405), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %407 : Float(1, 144, 28, 28) = onnx::Conv[dilations=[1, 1], group=144, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[2, 2]](%input.30, %73), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %408 : Float(1, 144, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%407, %74, %75, %76, %77), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.32 : Float(1, 144, 28, 28) = ^SwishAutoFn()(%408), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %410 : Float(1, 144, 1, 1) = onnx::GlobalAveragePool(%input.32), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %411 : Float(1, 6, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%410, %79, %80), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.34 : Float(1, 6, 1, 1) = ^SwishAutoFn()(%411), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %413 : Float(1, 144, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.34, %81, %82), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %414 : Float(1, 144, 1, 1) = onnx::Sigmoid(%413), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %415 : Float(1, 144, 28, 28) = onnx::Mul(%input.32, %414), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %416 : Float(1, 40, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%415, %83), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %417 : Float(1, 40, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%416, %84, %85, %86, %87), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[0]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %418 : Float(1, 240, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%417, %89), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %419 : Float(1, 240, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%418, %90, %91, %92, %93), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.39 : Float(1, 240, 28, 28) = ^SwishAutoFn()(%419), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %421 : Float(1, 240, 28, 28) = onnx::Conv[dilations=[1, 1], group=240, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%input.39, %95), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %422 : Float(1, 240, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%421, %96, %97, %98, %99), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.41 : Float(1, 240, 28, 28) = ^SwishAutoFn()(%422), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %424 : Float(1, 240, 1, 1) = onnx::GlobalAveragePool(%input.41), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %425 : Float(1, 10, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%424, %101, %102), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.43 : Float(1, 10, 1, 1) = ^SwishAutoFn()(%425), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %427 : Float(1, 240, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.43, %103, %104), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %428 : Float(1, 240, 1, 1) = onnx::Sigmoid(%427), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %429 : Float(1, 240, 28, 28) = onnx::Mul(%input.41, %428), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %430 : Float(1, 40, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%429, %105), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %431 : Float(1, 40, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%430, %106, %107, %108, %109), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %432 : Float(1, 40, 28, 28) = onnx::Add(%431, %417), scope: GenEfficientNet/Sequential[blocks]/Sequential[2]/InvertedResidual[1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:230:0
  %433 : Float(1, 240, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%432, %111), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %434 : Float(1, 240, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%433, %112, %113, %114, %115), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.48 : Float(1, 240, 28, 28) = ^SwishAutoFn()(%434), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %436 : Float(1, 240, 14, 14) = onnx::Conv[dilations=[1, 1], group=240, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%input.48, %117), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %437 : Float(1, 240, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%436, %118, %119, %120, %121), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.50 : Float(1, 240, 14, 14) = ^SwishAutoFn()(%437), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %439 : Float(1, 240, 1, 1) = onnx::GlobalAveragePool(%input.50), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %440 : Float(1, 10, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%439, %123, %124), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.52 : Float(1, 10, 1, 1) = ^SwishAutoFn()(%440), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %442 : Float(1, 240, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.52, %125, %126), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %443 : Float(1, 240, 1, 1) = onnx::Sigmoid(%442), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %444 : Float(1, 240, 14, 14) = onnx::Mul(%input.50, %443), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %445 : Float(1, 80, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%444, %127), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %446 : Float(1, 80, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%445, %128, %129, %130, %131), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[0]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %447 : Float(1, 480, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%446, %133), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %448 : Float(1, 480, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%447, %134, %135, %136, %137), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.57 : Float(1, 480, 14, 14) = ^SwishAutoFn()(%448), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %450 : Float(1, 480, 14, 14) = onnx::Conv[dilations=[1, 1], group=480, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%input.57, %139), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %451 : Float(1, 480, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%450, %140, %141, %142, %143), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.59 : Float(1, 480, 14, 14) = ^SwishAutoFn()(%451), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %453 : Float(1, 480, 1, 1) = onnx::GlobalAveragePool(%input.59), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %454 : Float(1, 20, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%453, %145, %146), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.61 : Float(1, 20, 1, 1) = ^SwishAutoFn()(%454), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %456 : Float(1, 480, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.61, %147, %148), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %457 : Float(1, 480, 1, 1) = onnx::Sigmoid(%456), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %458 : Float(1, 480, 14, 14) = onnx::Mul(%input.59, %457), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %459 : Float(1, 80, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%458, %149), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %460 : Float(1, 80, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%459, %150, %151, %152, %153), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %461 : Float(1, 80, 14, 14) = onnx::Add(%460, %446), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:230:0
  %462 : Float(1, 480, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%461, %155), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %463 : Float(1, 480, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%462, %156, %157, %158, %159), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.66 : Float(1, 480, 14, 14) = ^SwishAutoFn()(%463), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %465 : Float(1, 480, 14, 14) = onnx::Conv[dilations=[1, 1], group=480, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%input.66, %161), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %466 : Float(1, 480, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%465, %162, %163, %164, %165), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.68 : Float(1, 480, 14, 14) = ^SwishAutoFn()(%466), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %468 : Float(1, 480, 1, 1) = onnx::GlobalAveragePool(%input.68), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %469 : Float(1, 20, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%468, %167, %168), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.70 : Float(1, 20, 1, 1) = ^SwishAutoFn()(%469), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %471 : Float(1, 480, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.70, %169, %170), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %472 : Float(1, 480, 1, 1) = onnx::Sigmoid(%471), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %473 : Float(1, 480, 14, 14) = onnx::Mul(%input.68, %472), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %474 : Float(1, 80, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%473, %171), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %475 : Float(1, 80, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%474, %172, %173, %174, %175), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %476 : Float(1, 80, 14, 14) = onnx::Add(%475, %461), scope: GenEfficientNet/Sequential[blocks]/Sequential[3]/InvertedResidual[2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:230:0
  %477 : Float(1, 480, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%476, %177), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %478 : Float(1, 480, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%477, %178, %179, %180, %181), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.75 : Float(1, 480, 14, 14) = ^SwishAutoFn()(%478), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %480 : Float(1, 480, 14, 14) = onnx::Conv[dilations=[1, 1], group=480, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%input.75, %183), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %481 : Float(1, 480, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%480, %184, %185, %186, %187), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.77 : Float(1, 480, 14, 14) = ^SwishAutoFn()(%481), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %483 : Float(1, 480, 1, 1) = onnx::GlobalAveragePool(%input.77), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %484 : Float(1, 20, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%483, %189, %190), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.79 : Float(1, 20, 1, 1) = ^SwishAutoFn()(%484), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %486 : Float(1, 480, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.79, %191, %192), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %487 : Float(1, 480, 1, 1) = onnx::Sigmoid(%486), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %488 : Float(1, 480, 14, 14) = onnx::Mul(%input.77, %487), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %489 : Float(1, 112, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%488, %193), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %490 : Float(1, 112, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%489, %194, %195, %196, %197), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[0]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %491 : Float(1, 672, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%490, %199), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %492 : Float(1, 672, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%491, %200, %201, %202, %203), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.84 : Float(1, 672, 14, 14) = ^SwishAutoFn()(%492), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %494 : Float(1, 672, 14, 14) = onnx::Conv[dilations=[1, 1], group=672, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%input.84, %205), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %495 : Float(1, 672, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%494, %206, %207, %208, %209), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.86 : Float(1, 672, 14, 14) = ^SwishAutoFn()(%495), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %497 : Float(1, 672, 1, 1) = onnx::GlobalAveragePool(%input.86), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %498 : Float(1, 28, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%497, %211, %212), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.88 : Float(1, 28, 1, 1) = ^SwishAutoFn()(%498), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %500 : Float(1, 672, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.88, %213, %214), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %501 : Float(1, 672, 1, 1) = onnx::Sigmoid(%500), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %502 : Float(1, 672, 14, 14) = onnx::Mul(%input.86, %501), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %503 : Float(1, 112, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%502, %215), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %504 : Float(1, 112, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%503, %216, %217, %218, %219), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %505 : Float(1, 112, 14, 14) = onnx::Add(%504, %490), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:230:0
  %506 : Float(1, 672, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%505, %221), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %507 : Float(1, 672, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%506, %222, %223, %224, %225), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.93 : Float(1, 672, 14, 14) = ^SwishAutoFn()(%507), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %509 : Float(1, 672, 14, 14) = onnx::Conv[dilations=[1, 1], group=672, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%input.93, %227), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %510 : Float(1, 672, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%509, %228, %229, %230, %231), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.95 : Float(1, 672, 14, 14) = ^SwishAutoFn()(%510), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %512 : Float(1, 672, 1, 1) = onnx::GlobalAveragePool(%input.95), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %513 : Float(1, 28, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%512, %233, %234), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.97 : Float(1, 28, 1, 1) = ^SwishAutoFn()(%513), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %515 : Float(1, 672, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.97, %235, %236), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %516 : Float(1, 672, 1, 1) = onnx::Sigmoid(%515), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %517 : Float(1, 672, 14, 14) = onnx::Mul(%input.95, %516), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %518 : Float(1, 112, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%517, %237), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %519 : Float(1, 112, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%518, %238, %239, %240, %241), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %520 : Float(1, 112, 14, 14) = onnx::Add(%519, %505), scope: GenEfficientNet/Sequential[blocks]/Sequential[4]/InvertedResidual[2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:230:0
  %521 : Float(1, 672, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%520, %243), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %522 : Float(1, 672, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%521, %244, %245, %246, %247), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.102 : Float(1, 672, 14, 14) = ^SwishAutoFn()(%522), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %524 : Float(1, 672, 7, 7) = onnx::Conv[dilations=[1, 1], group=672, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[2, 2]](%input.102, %249), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %525 : Float(1, 672, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%524, %250, %251, %252, %253), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.104 : Float(1, 672, 7, 7) = ^SwishAutoFn()(%525), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %527 : Float(1, 672, 1, 1) = onnx::GlobalAveragePool(%input.104), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %528 : Float(1, 28, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%527, %255, %256), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.106 : Float(1, 28, 1, 1) = ^SwishAutoFn()(%528), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %530 : Float(1, 672, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.106, %257, %258), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %531 : Float(1, 672, 1, 1) = onnx::Sigmoid(%530), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %532 : Float(1, 672, 7, 7) = onnx::Mul(%input.104, %531), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %533 : Float(1, 192, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%532, %259), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %534 : Float(1, 192, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%533, %260, %261, %262, %263), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[0]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %535 : Float(1, 1152, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%534, %265), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %536 : Float(1, 1152, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%535, %266, %267, %268, %269), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.111 : Float(1, 1152, 7, 7) = ^SwishAutoFn()(%536), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %538 : Float(1, 1152, 7, 7) = onnx::Conv[dilations=[1, 1], group=1152, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%input.111, %271), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %539 : Float(1, 1152, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%538, %272, %273, %274, %275), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.113 : Float(1, 1152, 7, 7) = ^SwishAutoFn()(%539), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %541 : Float(1, 1152, 1, 1) = onnx::GlobalAveragePool(%input.113), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %542 : Float(1, 48, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%541, %277, %278), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.115 : Float(1, 48, 1, 1) = ^SwishAutoFn()(%542), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %544 : Float(1, 1152, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.115, %279, %280), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %545 : Float(1, 1152, 1, 1) = onnx::Sigmoid(%544), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %546 : Float(1, 1152, 7, 7) = onnx::Mul(%input.113, %545), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %547 : Float(1, 192, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%546, %281), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %548 : Float(1, 192, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%547, %282, %283, %284, %285), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %549 : Float(1, 192, 7, 7) = onnx::Add(%548, %534), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:230:0
  %550 : Float(1, 1152, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%549, %287), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %551 : Float(1, 1152, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%550, %288, %289, %290, %291), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.120 : Float(1, 1152, 7, 7) = ^SwishAutoFn()(%551), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %553 : Float(1, 1152, 7, 7) = onnx::Conv[dilations=[1, 1], group=1152, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%input.120, %293), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %554 : Float(1, 1152, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%553, %294, %295, %296, %297), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.122 : Float(1, 1152, 7, 7) = ^SwishAutoFn()(%554), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %556 : Float(1, 1152, 1, 1) = onnx::GlobalAveragePool(%input.122), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %557 : Float(1, 48, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%556, %299, %300), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.124 : Float(1, 48, 1, 1) = ^SwishAutoFn()(%557), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %559 : Float(1, 1152, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.124, %301, %302), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %560 : Float(1, 1152, 1, 1) = onnx::Sigmoid(%559), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %561 : Float(1, 1152, 7, 7) = onnx::Mul(%input.122, %560), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %562 : Float(1, 192, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%561, %303), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %563 : Float(1, 192, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%562, %304, %305, %306, %307), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %564 : Float(1, 192, 7, 7) = onnx::Add(%563, %549), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:230:0
  %565 : Float(1, 1152, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%564, %309), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %566 : Float(1, 1152, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%565, %310, %311, %312, %313), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.129 : Float(1, 1152, 7, 7) = ^SwishAutoFn()(%566), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %568 : Float(1, 1152, 7, 7) = onnx::Conv[dilations=[1, 1], group=1152, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%input.129, %315), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %569 : Float(1, 1152, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%568, %316, %317, %318, %319), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.131 : Float(1, 1152, 7, 7) = ^SwishAutoFn()(%569), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %571 : Float(1, 1152, 1, 1) = onnx::GlobalAveragePool(%input.131), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %572 : Float(1, 48, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%571, %321, %322), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.133 : Float(1, 48, 1, 1) = ^SwishAutoFn()(%572), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %574 : Float(1, 1152, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.133, %323, %324), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %575 : Float(1, 1152, 1, 1) = onnx::Sigmoid(%574), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %576 : Float(1, 1152, 7, 7) = onnx::Mul(%input.131, %575), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %577 : Float(1, 192, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%576, %325), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %578 : Float(1, 192, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%577, %326, %327, %328, %329), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %579 : Float(1, 192, 7, 7) = onnx::Add(%578, %564), scope: GenEfficientNet/Sequential[blocks]/Sequential[5]/InvertedResidual[3] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:230:0
  %580 : Float(1, 1152, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%579, %331), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/Conv2d[conv_pw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %581 : Float(1, 1152, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%580, %332, %333, %334, %335), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/BatchNorm2d[bn1] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.138 : Float(1, 1152, 7, 7) = ^SwishAutoFn()(%581), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %583 : Float(1, 1152, 7, 7) = onnx::Conv[dilations=[1, 1], group=1152, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%input.138, %337), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/Conv2d[conv_dw] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %584 : Float(1, 1152, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%583, %338, %339, %340, %341), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.140 : Float(1, 1152, 7, 7) = ^SwishAutoFn()(%584), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %586 : Float(1, 1152, 1, 1) = onnx::GlobalAveragePool(%input.140), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/SqueezeExcite[se]/AdaptiveAvgPool2d[avg_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %587 : Float(1, 48, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%586, %343, %344), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_reduce] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %input.142 : Float(1, 48, 1, 1) = ^SwishAutoFn()(%587), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/SqueezeExcite[se]/Swish[act1] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %589 : Float(1, 1152, 1, 1) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%input.142, %345, %346), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/SqueezeExcite[se]/Conv2d[conv_expand] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %590 : Float(1, 1152, 1, 1) = onnx::Sigmoid(%589), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:126:0
  %591 : Float(1, 1152, 7, 7) = onnx::Mul(%input.140, %590), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/SqueezeExcite[se] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/efficientnet_builder.py:152:0
  %592 : Float(1, 320, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%591, %347), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/Conv2d[conv_pwl] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %593 : Float(1, 320, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%592, %348, %349, %350, %351), scope: GenEfficientNet/Sequential[blocks]/Sequential[6]/InvertedResidual[0]/BatchNorm2d[bn3] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %594 : Float(1, 1280, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%593, %353), scope: GenEfficientNet/Conv2d[conv_head] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py:342:0
  %595 : Float(1, 1280, 7, 7) = onnx::BatchNormalization[epsilon=1e-05, momentum=0.9](%594, %354, %355, %356, %357), scope: GenEfficientNet/BatchNorm2d[bn2] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1670:0
  %input.147 : Float(1, 1280, 7, 7) = ^SwishAutoFn()(%595), scope: GenEfficientNet/Swish[act2] # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/activations.py:30:0
  %597 : Float(1, 1280, 1, 1) = onnx::GlobalAveragePool(%input.147), scope: GenEfficientNet/AdaptiveAvgPool2d[global_pool] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:768:0
  %598 : Float(1, 1280) = onnx::Flatten[axis=1](%597), scope: GenEfficientNet # /ssd/pytorch/gen-efficientnet-pytorch/geffnet/gen_efficientnet.py:173:0
  %output0 : Float(1, 1000) = onnx::Gemm[alpha=1, beta=1, transB=1](%598, %359, %360), scope: GenEfficientNet/Linear[classifier] # /home/ivan/.local/lib/python3.6/site-packages/torch/nn/functional.py:1370:0
  return (%output0)

rwightman · 2019-10-29T15:26:52Z

As per the little comment in the README, you currently need to set the exportable flag to True (int he latest version it's here) https://github.com/rwightman/gen-efficientnet-pytorch/blob/master/geffnet/exportable.py ...

The memory-efficient autograd FN for swish, and the default padded same convolutions for the TF variant models don't export properly

Although... the MobileNetV3 model uses neither of those so it should have worked and looked like it would have aside from the segmentation fault. I just ran it from the master with no crashes/issues. Are you using a fairly new version of PyTorch 1.2+? My ONNX version is 1.5.

rwightman · 2019-10-29T19:21:31Z

@ivder It looks like the opposite of what I expected, PyTorch/ONNX being too new. I get the same segfault with PyTorch 1.3 + ONNX 1.6 installed. I get a different crash with PyTorch 1.3 + ONNX 1.5 installed. It works with PyTorch 1.2 + ONNX 1.5

Some related issues:
onnx/onnx#2417
onnx/onnx#2394

ivder · 2019-10-30T01:32:20Z

@rwightman Thanks for clarifying this issue. I used the latest onnx and pytorch, after downgraded both of them I manage to export the model to ONNX.

Another thing that I want to confirm, the exported onnx is using caffe2 as backend right? Can I use the onnx model in another framework such as TensorFlow, Caffee, TensorRT for inference? Thanks for the help

rwightman · 2019-10-30T07:03:02Z

@ivder Good to hear. You should be able to use the models with any ONNX runtime or conversion tool (https://github.com/microsoft/onnxruntime, https://github.com/onnx/onnx-tensorrt, etc) that supports the same file format and operator versions as you export in. Caffe2 is just the default available runtime if you have PyTorch installed.

rwightman closed this as completed Oct 30, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Export to ONNX Error #10

Export to ONNX Error #10

ivder commented Oct 29, 2019 •

edited

Loading

ivder commented Oct 29, 2019

rwightman commented Oct 29, 2019

rwightman commented Oct 29, 2019

ivder commented Oct 30, 2019

rwightman commented Oct 30, 2019

Export to ONNX Error #10

Export to ONNX Error #10

Comments

ivder commented Oct 29, 2019 • edited Loading

ivder commented Oct 29, 2019

rwightman commented Oct 29, 2019

rwightman commented Oct 29, 2019

ivder commented Oct 30, 2019

rwightman commented Oct 30, 2019

ivder commented Oct 29, 2019 •

edited

Loading