Skip to content

Instantly share code, notes, and snippets.

@Lyken17

Lyken17/proxylessnas_on_tvm.txt

Created June 14, 2019 04:14
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save Lyken17/9f74f71998c1f4761f04b8147e546eb2 to your computer and use it in GitHub Desktop.
Save Lyken17/9f74f71998c1f4761f04b8147e546eb2 to your computer and use it in GitHub Desktop.
Download ZIP
ProxylessNAS deployment on TVM
Raw
proxylessnas_on_tvm.txt
/Users/ligeng/anaconda3/bin/python /Users/ligeng/Workspace/ProxylessNAS/load_onnx.py
File /Users/ligeng/.tvm_test_data/data/cat.png exists, skip.
graph(%actual_input_1 : Float(1, 3, 224, 224),
%first_conv.bn.weight : Float(40),
%first_conv.bn.bias : Float(40),
%first_conv.bn.running_mean : Float(40),
%first_conv.bn.running_var : Float(40),
%first_conv.bn.num_batches_tracked : Long(),
%first_conv.conv.weight : Float(40, 3, 3, 3),
%blocks.0.mobile_inverted_conv.depth_conv.conv.weight : Float(40, 1, 3, 3),
%blocks.0.mobile_inverted_conv.depth_conv.bn.weight : Float(40),
%blocks.0.mobile_inverted_conv.depth_conv.bn.bias : Float(40),
%blocks.0.mobile_inverted_conv.depth_conv.bn.running_mean : Float(40),
%blocks.0.mobile_inverted_conv.depth_conv.bn.running_var : Float(40),
%blocks.0.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.0.mobile_inverted_conv.point_linear.conv.weight : Float(24, 40, 1, 1),
%blocks.0.mobile_inverted_conv.point_linear.bn.weight : Float(24),
%blocks.0.mobile_inverted_conv.point_linear.bn.bias : Float(24),
%blocks.0.mobile_inverted_conv.point_linear.bn.running_mean : Float(24),
%blocks.0.mobile_inverted_conv.point_linear.bn.running_var : Float(24),
%blocks.0.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.1.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(144, 24, 1, 1),
%blocks.1.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(144),
%blocks.1.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(144),
%blocks.1.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(144),
%blocks.1.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(144),
%blocks.1.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.1.mobile_inverted_conv.depth_conv.conv.weight : Float(144, 1, 3, 3),
%blocks.1.mobile_inverted_conv.depth_conv.bn.weight : Float(144),
%blocks.1.mobile_inverted_conv.depth_conv.bn.bias : Float(144),
%blocks.1.mobile_inverted_conv.depth_conv.bn.running_mean : Float(144),
%blocks.1.mobile_inverted_conv.depth_conv.bn.running_var : Float(144),
%blocks.1.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.1.mobile_inverted_conv.point_linear.conv.weight : Float(32, 144, 1, 1),
%blocks.1.mobile_inverted_conv.point_linear.bn.weight : Float(32),
%blocks.1.mobile_inverted_conv.point_linear.bn.bias : Float(32),
%blocks.1.mobile_inverted_conv.point_linear.bn.running_mean : Float(32),
%blocks.1.mobile_inverted_conv.point_linear.bn.running_var : Float(32),
%blocks.1.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.2.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(96, 32, 1, 1),
%blocks.2.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(96),
%blocks.2.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(96),
%blocks.2.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(96),
%blocks.2.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(96),
%blocks.2.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.2.mobile_inverted_conv.depth_conv.conv.weight : Float(96, 1, 3, 3),
%blocks.2.mobile_inverted_conv.depth_conv.bn.weight : Float(96),
%blocks.2.mobile_inverted_conv.depth_conv.bn.bias : Float(96),
%blocks.2.mobile_inverted_conv.depth_conv.bn.running_mean : Float(96),
%blocks.2.mobile_inverted_conv.depth_conv.bn.running_var : Float(96),
%blocks.2.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.2.mobile_inverted_conv.point_linear.conv.weight : Float(32, 96, 1, 1),
%blocks.2.mobile_inverted_conv.point_linear.bn.weight : Float(32),
%blocks.2.mobile_inverted_conv.point_linear.bn.bias : Float(32),
%blocks.2.mobile_inverted_conv.point_linear.bn.running_mean : Float(32),
%blocks.2.mobile_inverted_conv.point_linear.bn.running_var : Float(32),
%blocks.2.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.3.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(96, 32, 1, 1),
%blocks.3.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(96),
%blocks.3.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(96),
%blocks.3.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(96),
%blocks.3.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(96),
%blocks.3.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.3.mobile_inverted_conv.depth_conv.conv.weight : Float(96, 1, 3, 3),
%blocks.3.mobile_inverted_conv.depth_conv.bn.weight : Float(96),
%blocks.3.mobile_inverted_conv.depth_conv.bn.bias : Float(96),
%blocks.3.mobile_inverted_conv.depth_conv.bn.running_mean : Float(96),
%blocks.3.mobile_inverted_conv.depth_conv.bn.running_var : Float(96),
%blocks.3.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.3.mobile_inverted_conv.point_linear.conv.weight : Float(32, 96, 1, 1),
%blocks.3.mobile_inverted_conv.point_linear.bn.weight : Float(32),
%blocks.3.mobile_inverted_conv.point_linear.bn.bias : Float(32),
%blocks.3.mobile_inverted_conv.point_linear.bn.running_mean : Float(32),
%blocks.3.mobile_inverted_conv.point_linear.bn.running_var : Float(32),
%blocks.3.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.4.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(96, 32, 1, 1),
%blocks.4.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(96),
%blocks.4.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(96),
%blocks.4.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(96),
%blocks.4.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(96),
%blocks.4.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.4.mobile_inverted_conv.depth_conv.conv.weight : Float(96, 1, 3, 3),
%blocks.4.mobile_inverted_conv.depth_conv.bn.weight : Float(96),
%blocks.4.mobile_inverted_conv.depth_conv.bn.bias : Float(96),
%blocks.4.mobile_inverted_conv.depth_conv.bn.running_mean : Float(96),
%blocks.4.mobile_inverted_conv.depth_conv.bn.running_var : Float(96),
%blocks.4.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.4.mobile_inverted_conv.point_linear.conv.weight : Float(32, 96, 1, 1),
%blocks.4.mobile_inverted_conv.point_linear.bn.weight : Float(32),
%blocks.4.mobile_inverted_conv.point_linear.bn.bias : Float(32),
%blocks.4.mobile_inverted_conv.point_linear.bn.running_mean : Float(32),
%blocks.4.mobile_inverted_conv.point_linear.bn.running_var : Float(32),
%blocks.4.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.5.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(192, 32, 1, 1),
%blocks.5.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(192),
%blocks.5.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(192),
%blocks.5.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(192),
%blocks.5.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(192),
%blocks.5.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.5.mobile_inverted_conv.depth_conv.conv.weight : Float(192, 1, 3, 3),
%blocks.5.mobile_inverted_conv.depth_conv.bn.weight : Float(192),
%blocks.5.mobile_inverted_conv.depth_conv.bn.bias : Float(192),
%blocks.5.mobile_inverted_conv.depth_conv.bn.running_mean : Float(192),
%blocks.5.mobile_inverted_conv.depth_conv.bn.running_var : Float(192),
%blocks.5.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.5.mobile_inverted_conv.point_linear.conv.weight : Float(48, 192, 1, 1),
%blocks.5.mobile_inverted_conv.point_linear.bn.weight : Float(48),
%blocks.5.mobile_inverted_conv.point_linear.bn.bias : Float(48),
%blocks.5.mobile_inverted_conv.point_linear.bn.running_mean : Float(48),
%blocks.5.mobile_inverted_conv.point_linear.bn.running_var : Float(48),
%blocks.5.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.6.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(144, 48, 1, 1),
%blocks.6.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(144),
%blocks.6.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(144),
%blocks.6.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(144),
%blocks.6.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(144),
%blocks.6.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.6.mobile_inverted_conv.depth_conv.conv.weight : Float(144, 1, 3, 3),
%blocks.6.mobile_inverted_conv.depth_conv.bn.weight : Float(144),
%blocks.6.mobile_inverted_conv.depth_conv.bn.bias : Float(144),
%blocks.6.mobile_inverted_conv.depth_conv.bn.running_mean : Float(144),
%blocks.6.mobile_inverted_conv.depth_conv.bn.running_var : Float(144),
%blocks.6.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.6.mobile_inverted_conv.point_linear.conv.weight : Float(48, 144, 1, 1),
%blocks.6.mobile_inverted_conv.point_linear.bn.weight : Float(48),
%blocks.6.mobile_inverted_conv.point_linear.bn.bias : Float(48),
%blocks.6.mobile_inverted_conv.point_linear.bn.running_mean : Float(48),
%blocks.6.mobile_inverted_conv.point_linear.bn.running_var : Float(48),
%blocks.6.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.7.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(144, 48, 1, 1),
%blocks.7.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(144),
%blocks.7.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(144),
%blocks.7.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(144),
%blocks.7.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(144),
%blocks.7.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.7.mobile_inverted_conv.depth_conv.conv.weight : Float(144, 1, 3, 3),
%blocks.7.mobile_inverted_conv.depth_conv.bn.weight : Float(144),
%blocks.7.mobile_inverted_conv.depth_conv.bn.bias : Float(144),
%blocks.7.mobile_inverted_conv.depth_conv.bn.running_mean : Float(144),
%blocks.7.mobile_inverted_conv.depth_conv.bn.running_var : Float(144),
%blocks.7.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.7.mobile_inverted_conv.point_linear.conv.weight : Float(48, 144, 1, 1),
%blocks.7.mobile_inverted_conv.point_linear.bn.weight : Float(48),
%blocks.7.mobile_inverted_conv.point_linear.bn.bias : Float(48),
%blocks.7.mobile_inverted_conv.point_linear.bn.running_mean : Float(48),
%blocks.7.mobile_inverted_conv.point_linear.bn.running_var : Float(48),
%blocks.7.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.8.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(144, 48, 1, 1),
%blocks.8.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(144),
%blocks.8.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(144),
%blocks.8.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(144),
%blocks.8.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(144),
%blocks.8.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.8.mobile_inverted_conv.depth_conv.conv.weight : Float(144, 1, 5, 5),
%blocks.8.mobile_inverted_conv.depth_conv.bn.weight : Float(144),
%blocks.8.mobile_inverted_conv.depth_conv.bn.bias : Float(144),
%blocks.8.mobile_inverted_conv.depth_conv.bn.running_mean : Float(144),
%blocks.8.mobile_inverted_conv.depth_conv.bn.running_var : Float(144),
%blocks.8.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.8.mobile_inverted_conv.point_linear.conv.weight : Float(48, 144, 1, 1),
%blocks.8.mobile_inverted_conv.point_linear.bn.weight : Float(48),
%blocks.8.mobile_inverted_conv.point_linear.bn.bias : Float(48),
%blocks.8.mobile_inverted_conv.point_linear.bn.running_mean : Float(48),
%blocks.8.mobile_inverted_conv.point_linear.bn.running_var : Float(48),
%blocks.8.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.9.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(288, 48, 1, 1),
%blocks.9.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(288),
%blocks.9.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(288),
%blocks.9.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(288),
%blocks.9.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(288),
%blocks.9.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.9.mobile_inverted_conv.depth_conv.conv.weight : Float(288, 1, 3, 3),
%blocks.9.mobile_inverted_conv.depth_conv.bn.weight : Float(288),
%blocks.9.mobile_inverted_conv.depth_conv.bn.bias : Float(288),
%blocks.9.mobile_inverted_conv.depth_conv.bn.running_mean : Float(288),
%blocks.9.mobile_inverted_conv.depth_conv.bn.running_var : Float(288),
%blocks.9.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.9.mobile_inverted_conv.point_linear.conv.weight : Float(88, 288, 1, 1),
%blocks.9.mobile_inverted_conv.point_linear.bn.weight : Float(88),
%blocks.9.mobile_inverted_conv.point_linear.bn.bias : Float(88),
%blocks.9.mobile_inverted_conv.point_linear.bn.running_mean : Float(88),
%blocks.9.mobile_inverted_conv.point_linear.bn.running_var : Float(88),
%blocks.9.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.12.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(264, 88, 1, 1),
%blocks.12.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(264),
%blocks.12.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(264),
%blocks.12.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(264),
%blocks.12.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(264),
%blocks.12.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.12.mobile_inverted_conv.depth_conv.conv.weight : Float(264, 1, 3, 3),
%blocks.12.mobile_inverted_conv.depth_conv.bn.weight : Float(264),
%blocks.12.mobile_inverted_conv.depth_conv.bn.bias : Float(264),
%blocks.12.mobile_inverted_conv.depth_conv.bn.running_mean : Float(264),
%blocks.12.mobile_inverted_conv.depth_conv.bn.running_var : Float(264),
%blocks.12.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.12.mobile_inverted_conv.point_linear.conv.weight : Float(88, 264, 1, 1),
%blocks.12.mobile_inverted_conv.point_linear.bn.weight : Float(88),
%blocks.12.mobile_inverted_conv.point_linear.bn.bias : Float(88),
%blocks.12.mobile_inverted_conv.point_linear.bn.running_mean : Float(88),
%blocks.12.mobile_inverted_conv.point_linear.bn.running_var : Float(88),
%blocks.12.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.13.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(528, 88, 1, 1),
%blocks.13.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(528),
%blocks.13.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(528),
%blocks.13.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(528),
%blocks.13.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(528),
%blocks.13.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.13.mobile_inverted_conv.depth_conv.conv.weight : Float(528, 1, 5, 5),
%blocks.13.mobile_inverted_conv.depth_conv.bn.weight : Float(528),
%blocks.13.mobile_inverted_conv.depth_conv.bn.bias : Float(528),
%blocks.13.mobile_inverted_conv.depth_conv.bn.running_mean : Float(528),
%blocks.13.mobile_inverted_conv.depth_conv.bn.running_var : Float(528),
%blocks.13.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.13.mobile_inverted_conv.point_linear.conv.weight : Float(104, 528, 1, 1),
%blocks.13.mobile_inverted_conv.point_linear.bn.weight : Float(104),
%blocks.13.mobile_inverted_conv.point_linear.bn.bias : Float(104),
%blocks.13.mobile_inverted_conv.point_linear.bn.running_mean : Float(104),
%blocks.13.mobile_inverted_conv.point_linear.bn.running_var : Float(104),
%blocks.13.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.14.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(312, 104, 1, 1),
%blocks.14.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(312),
%blocks.14.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(312),
%blocks.14.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(312),
%blocks.14.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(312),
%blocks.14.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.14.mobile_inverted_conv.depth_conv.conv.weight : Float(312, 1, 3, 3),
%blocks.14.mobile_inverted_conv.depth_conv.bn.weight : Float(312),
%blocks.14.mobile_inverted_conv.depth_conv.bn.bias : Float(312),
%blocks.14.mobile_inverted_conv.depth_conv.bn.running_mean : Float(312),
%blocks.14.mobile_inverted_conv.depth_conv.bn.running_var : Float(312),
%blocks.14.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.14.mobile_inverted_conv.point_linear.conv.weight : Float(104, 312, 1, 1),
%blocks.14.mobile_inverted_conv.point_linear.bn.weight : Float(104),
%blocks.14.mobile_inverted_conv.point_linear.bn.bias : Float(104),
%blocks.14.mobile_inverted_conv.point_linear.bn.running_mean : Float(104),
%blocks.14.mobile_inverted_conv.point_linear.bn.running_var : Float(104),
%blocks.14.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.15.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(312, 104, 1, 1),
%blocks.15.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(312),
%blocks.15.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(312),
%blocks.15.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(312),
%blocks.15.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(312),
%blocks.15.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.15.mobile_inverted_conv.depth_conv.conv.weight : Float(312, 1, 3, 3),
%blocks.15.mobile_inverted_conv.depth_conv.bn.weight : Float(312),
%blocks.15.mobile_inverted_conv.depth_conv.bn.bias : Float(312),
%blocks.15.mobile_inverted_conv.depth_conv.bn.running_mean : Float(312),
%blocks.15.mobile_inverted_conv.depth_conv.bn.running_var : Float(312),
%blocks.15.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.15.mobile_inverted_conv.point_linear.conv.weight : Float(104, 312, 1, 1),
%blocks.15.mobile_inverted_conv.point_linear.bn.weight : Float(104),
%blocks.15.mobile_inverted_conv.point_linear.bn.bias : Float(104),
%blocks.15.mobile_inverted_conv.point_linear.bn.running_mean : Float(104),
%blocks.15.mobile_inverted_conv.point_linear.bn.running_var : Float(104),
%blocks.15.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.16.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(312, 104, 1, 1),
%blocks.16.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(312),
%blocks.16.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(312),
%blocks.16.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(312),
%blocks.16.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(312),
%blocks.16.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.16.mobile_inverted_conv.depth_conv.conv.weight : Float(312, 1, 3, 3),
%blocks.16.mobile_inverted_conv.depth_conv.bn.weight : Float(312),
%blocks.16.mobile_inverted_conv.depth_conv.bn.bias : Float(312),
%blocks.16.mobile_inverted_conv.depth_conv.bn.running_mean : Float(312),
%blocks.16.mobile_inverted_conv.depth_conv.bn.running_var : Float(312),
%blocks.16.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.16.mobile_inverted_conv.point_linear.conv.weight : Float(104, 312, 1, 1),
%blocks.16.mobile_inverted_conv.point_linear.bn.weight : Float(104),
%blocks.16.mobile_inverted_conv.point_linear.bn.bias : Float(104),
%blocks.16.mobile_inverted_conv.point_linear.bn.running_mean : Float(104),
%blocks.16.mobile_inverted_conv.point_linear.bn.running_var : Float(104),
%blocks.16.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.17.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(624, 104, 1, 1),
%blocks.17.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(624),
%blocks.17.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(624),
%blocks.17.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(624),
%blocks.17.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(624),
%blocks.17.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.17.mobile_inverted_conv.depth_conv.conv.weight : Float(624, 1, 5, 5),
%blocks.17.mobile_inverted_conv.depth_conv.bn.weight : Float(624),
%blocks.17.mobile_inverted_conv.depth_conv.bn.bias : Float(624),
%blocks.17.mobile_inverted_conv.depth_conv.bn.running_mean : Float(624),
%blocks.17.mobile_inverted_conv.depth_conv.bn.running_var : Float(624),
%blocks.17.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.17.mobile_inverted_conv.point_linear.conv.weight : Float(216, 624, 1, 1),
%blocks.17.mobile_inverted_conv.point_linear.bn.weight : Float(216),
%blocks.17.mobile_inverted_conv.point_linear.bn.bias : Float(216),
%blocks.17.mobile_inverted_conv.point_linear.bn.running_mean : Float(216),
%blocks.17.mobile_inverted_conv.point_linear.bn.running_var : Float(216),
%blocks.17.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.18.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(648, 216, 1, 1),
%blocks.18.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(648),
%blocks.18.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(648),
%blocks.18.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(648),
%blocks.18.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(648),
%blocks.18.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.18.mobile_inverted_conv.depth_conv.conv.weight : Float(648, 1, 5, 5),
%blocks.18.mobile_inverted_conv.depth_conv.bn.weight : Float(648),
%blocks.18.mobile_inverted_conv.depth_conv.bn.bias : Float(648),
%blocks.18.mobile_inverted_conv.depth_conv.bn.running_mean : Float(648),
%blocks.18.mobile_inverted_conv.depth_conv.bn.running_var : Float(648),
%blocks.18.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.18.mobile_inverted_conv.point_linear.conv.weight : Float(216, 648, 1, 1),
%blocks.18.mobile_inverted_conv.point_linear.bn.weight : Float(216),
%blocks.18.mobile_inverted_conv.point_linear.bn.bias : Float(216),
%blocks.18.mobile_inverted_conv.point_linear.bn.running_mean : Float(216),
%blocks.18.mobile_inverted_conv.point_linear.bn.running_var : Float(216),
%blocks.18.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.19.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(648, 216, 1, 1),
%blocks.19.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(648),
%blocks.19.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(648),
%blocks.19.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(648),
%blocks.19.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(648),
%blocks.19.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.19.mobile_inverted_conv.depth_conv.conv.weight : Float(648, 1, 5, 5),
%blocks.19.mobile_inverted_conv.depth_conv.bn.weight : Float(648),
%blocks.19.mobile_inverted_conv.depth_conv.bn.bias : Float(648),
%blocks.19.mobile_inverted_conv.depth_conv.bn.running_mean : Float(648),
%blocks.19.mobile_inverted_conv.depth_conv.bn.running_var : Float(648),
%blocks.19.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.19.mobile_inverted_conv.point_linear.conv.weight : Float(216, 648, 1, 1),
%blocks.19.mobile_inverted_conv.point_linear.bn.weight : Float(216),
%blocks.19.mobile_inverted_conv.point_linear.bn.bias : Float(216),
%blocks.19.mobile_inverted_conv.point_linear.bn.running_mean : Float(216),
%blocks.19.mobile_inverted_conv.point_linear.bn.running_var : Float(216),
%blocks.19.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.20.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(648, 216, 1, 1),
%blocks.20.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(648),
%blocks.20.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(648),
%blocks.20.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(648),
%blocks.20.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(648),
%blocks.20.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.20.mobile_inverted_conv.depth_conv.conv.weight : Float(648, 1, 3, 3),
%blocks.20.mobile_inverted_conv.depth_conv.bn.weight : Float(648),
%blocks.20.mobile_inverted_conv.depth_conv.bn.bias : Float(648),
%blocks.20.mobile_inverted_conv.depth_conv.bn.running_mean : Float(648),
%blocks.20.mobile_inverted_conv.depth_conv.bn.running_var : Float(648),
%blocks.20.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.20.mobile_inverted_conv.point_linear.conv.weight : Float(216, 648, 1, 1),
%blocks.20.mobile_inverted_conv.point_linear.bn.weight : Float(216),
%blocks.20.mobile_inverted_conv.point_linear.bn.bias : Float(216),
%blocks.20.mobile_inverted_conv.point_linear.bn.running_mean : Float(216),
%blocks.20.mobile_inverted_conv.point_linear.bn.running_var : Float(216),
%blocks.20.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%blocks.21.mobile_inverted_conv.inverted_bottleneck.conv.weight : Float(1296, 216, 1, 1),
%blocks.21.mobile_inverted_conv.inverted_bottleneck.bn.weight : Float(1296),
%blocks.21.mobile_inverted_conv.inverted_bottleneck.bn.bias : Float(1296),
%blocks.21.mobile_inverted_conv.inverted_bottleneck.bn.running_mean : Float(1296),
%blocks.21.mobile_inverted_conv.inverted_bottleneck.bn.running_var : Float(1296),
%blocks.21.mobile_inverted_conv.inverted_bottleneck.bn.num_batches_tracked : Long(),
%blocks.21.mobile_inverted_conv.depth_conv.conv.weight : Float(1296, 1, 5, 5),
%blocks.21.mobile_inverted_conv.depth_conv.bn.weight : Float(1296),
%blocks.21.mobile_inverted_conv.depth_conv.bn.bias : Float(1296),
%blocks.21.mobile_inverted_conv.depth_conv.bn.running_mean : Float(1296),
%blocks.21.mobile_inverted_conv.depth_conv.bn.running_var : Float(1296),
%blocks.21.mobile_inverted_conv.depth_conv.bn.num_batches_tracked : Long(),
%blocks.21.mobile_inverted_conv.point_linear.conv.weight : Float(360, 1296, 1, 1),
%blocks.21.mobile_inverted_conv.point_linear.bn.weight : Float(360),
%blocks.21.mobile_inverted_conv.point_linear.bn.bias : Float(360),
%blocks.21.mobile_inverted_conv.point_linear.bn.running_mean : Float(360),
%blocks.21.mobile_inverted_conv.point_linear.bn.running_var : Float(360),
%blocks.21.mobile_inverted_conv.point_linear.bn.num_batches_tracked : Long(),
%feature_mix_layer.bn.weight : Float(1432),
%feature_mix_layer.bn.bias : Float(1432),
%feature_mix_layer.bn.running_mean : Float(1432),
%feature_mix_layer.bn.running_var : Float(1432),
%feature_mix_layer.bn.num_batches_tracked : Long(),
%feature_mix_layer.conv.weight : Float(1432, 360, 1, 1),
%classifier.linear.weight : Float(1000, 1432),
%classifier.linear.bias : Float(1000)):
%369 : Float(1, 40, 112, 112) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%actual_input_1, %first_conv.conv.weight), scope: ProxylessNASNets/ConvLayer[first_conv]/Conv2d[conv]
%370 : Float(1, 40, 112, 112) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%369, %first_conv.bn.weight, %first_conv.bn.bias, %first_conv.bn.running_mean, %first_conv.bn.running_var), scope: ProxylessNASNets/ConvLayer[first_conv]/BatchNorm2d[bn]
%371 : Float(1, 40, 112, 112) = onnx::Clip[max=6, min=0](%370), scope: ProxylessNASNets/ConvLayer[first_conv]/ReLU6[activation]
%372 : Float(1, 40, 112, 112) = onnx::Conv[dilations=[1, 1], group=40, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%371, %blocks.0.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%373 : Float(1, 40, 112, 112) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%372, %blocks.0.mobile_inverted_conv.depth_conv.bn.weight, %blocks.0.mobile_inverted_conv.depth_conv.bn.bias, %blocks.0.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.0.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%374 : Float(1, 40, 112, 112) = onnx::Clip[max=6, min=0](%373), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%375 : Float(1, 24, 112, 112) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%374, %blocks.0.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%376 : Float(1, 24, 112, 112) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%375, %blocks.0.mobile_inverted_conv.point_linear.bn.weight, %blocks.0.mobile_inverted_conv.point_linear.bn.bias, %blocks.0.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.0.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%377 : Float(1, 144, 112, 112) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%376, %blocks.1.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%378 : Float(1, 144, 112, 112) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%377, %blocks.1.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.1.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.1.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.1.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%379 : Float(1, 144, 112, 112) = onnx::Clip[max=6, min=0](%378), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%380 : Float(1, 144, 56, 56) = onnx::Conv[dilations=[1, 1], group=144, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%379, %blocks.1.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%381 : Float(1, 144, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%380, %blocks.1.mobile_inverted_conv.depth_conv.bn.weight, %blocks.1.mobile_inverted_conv.depth_conv.bn.bias, %blocks.1.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.1.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%382 : Float(1, 144, 56, 56) = onnx::Clip[max=6, min=0](%381), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%383 : Float(1, 32, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%382, %blocks.1.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%384 : Float(1, 32, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%383, %blocks.1.mobile_inverted_conv.point_linear.bn.weight, %blocks.1.mobile_inverted_conv.point_linear.bn.bias, %blocks.1.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.1.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%385 : Float(1, 96, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%384, %blocks.2.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%386 : Float(1, 96, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%385, %blocks.2.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.2.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.2.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.2.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%387 : Float(1, 96, 56, 56) = onnx::Clip[max=6, min=0](%386), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%388 : Float(1, 96, 56, 56) = onnx::Conv[dilations=[1, 1], group=96, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%387, %blocks.2.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%389 : Float(1, 96, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%388, %blocks.2.mobile_inverted_conv.depth_conv.bn.weight, %blocks.2.mobile_inverted_conv.depth_conv.bn.bias, %blocks.2.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.2.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%390 : Float(1, 96, 56, 56) = onnx::Clip[max=6, min=0](%389), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%391 : Float(1, 32, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%390, %blocks.2.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%392 : Float(1, 32, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%391, %blocks.2.mobile_inverted_conv.point_linear.bn.weight, %blocks.2.mobile_inverted_conv.point_linear.bn.bias, %blocks.2.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.2.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%393 : Float(1, 32, 56, 56) = onnx::Add(%384, %392), scope: ProxylessNASNets/MobileInvertedResidualBlock
%394 : Float(1, 96, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%393, %blocks.3.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%395 : Float(1, 96, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%394, %blocks.3.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.3.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.3.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.3.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%396 : Float(1, 96, 56, 56) = onnx::Clip[max=6, min=0](%395), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%397 : Float(1, 96, 56, 56) = onnx::Conv[dilations=[1, 1], group=96, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%396, %blocks.3.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%398 : Float(1, 96, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%397, %blocks.3.mobile_inverted_conv.depth_conv.bn.weight, %blocks.3.mobile_inverted_conv.depth_conv.bn.bias, %blocks.3.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.3.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%399 : Float(1, 96, 56, 56) = onnx::Clip[max=6, min=0](%398), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%400 : Float(1, 32, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%399, %blocks.3.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%401 : Float(1, 32, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%400, %blocks.3.mobile_inverted_conv.point_linear.bn.weight, %blocks.3.mobile_inverted_conv.point_linear.bn.bias, %blocks.3.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.3.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%402 : Float(1, 32, 56, 56) = onnx::Add(%393, %401), scope: ProxylessNASNets/MobileInvertedResidualBlock
%403 : Float(1, 96, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%402, %blocks.4.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%404 : Float(1, 96, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%403, %blocks.4.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.4.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.4.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.4.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%405 : Float(1, 96, 56, 56) = onnx::Clip[max=6, min=0](%404), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%406 : Float(1, 96, 56, 56) = onnx::Conv[dilations=[1, 1], group=96, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%405, %blocks.4.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%407 : Float(1, 96, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%406, %blocks.4.mobile_inverted_conv.depth_conv.bn.weight, %blocks.4.mobile_inverted_conv.depth_conv.bn.bias, %blocks.4.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.4.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%408 : Float(1, 96, 56, 56) = onnx::Clip[max=6, min=0](%407), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%409 : Float(1, 32, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%408, %blocks.4.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%410 : Float(1, 32, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%409, %blocks.4.mobile_inverted_conv.point_linear.bn.weight, %blocks.4.mobile_inverted_conv.point_linear.bn.bias, %blocks.4.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.4.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%411 : Float(1, 32, 56, 56) = onnx::Add(%402, %410), scope: ProxylessNASNets/MobileInvertedResidualBlock
%412 : Float(1, 192, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%411, %blocks.5.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%413 : Float(1, 192, 56, 56) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%412, %blocks.5.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.5.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.5.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.5.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%414 : Float(1, 192, 56, 56) = onnx::Clip[max=6, min=0](%413), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%415 : Float(1, 192, 28, 28) = onnx::Conv[dilations=[1, 1], group=192, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%414, %blocks.5.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%416 : Float(1, 192, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%415, %blocks.5.mobile_inverted_conv.depth_conv.bn.weight, %blocks.5.mobile_inverted_conv.depth_conv.bn.bias, %blocks.5.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.5.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%417 : Float(1, 192, 28, 28) = onnx::Clip[max=6, min=0](%416), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%418 : Float(1, 48, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%417, %blocks.5.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%419 : Float(1, 48, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%418, %blocks.5.mobile_inverted_conv.point_linear.bn.weight, %blocks.5.mobile_inverted_conv.point_linear.bn.bias, %blocks.5.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.5.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%420 : Float(1, 144, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%419, %blocks.6.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%421 : Float(1, 144, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%420, %blocks.6.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.6.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.6.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.6.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%422 : Float(1, 144, 28, 28) = onnx::Clip[max=6, min=0](%421), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%423 : Float(1, 144, 28, 28) = onnx::Conv[dilations=[1, 1], group=144, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%422, %blocks.6.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%424 : Float(1, 144, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%423, %blocks.6.mobile_inverted_conv.depth_conv.bn.weight, %blocks.6.mobile_inverted_conv.depth_conv.bn.bias, %blocks.6.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.6.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%425 : Float(1, 144, 28, 28) = onnx::Clip[max=6, min=0](%424), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%426 : Float(1, 48, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%425, %blocks.6.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%427 : Float(1, 48, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%426, %blocks.6.mobile_inverted_conv.point_linear.bn.weight, %blocks.6.mobile_inverted_conv.point_linear.bn.bias, %blocks.6.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.6.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%428 : Float(1, 48, 28, 28) = onnx::Add(%419, %427), scope: ProxylessNASNets/MobileInvertedResidualBlock
%429 : Float(1, 144, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%428, %blocks.7.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%430 : Float(1, 144, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%429, %blocks.7.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.7.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.7.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.7.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%431 : Float(1, 144, 28, 28) = onnx::Clip[max=6, min=0](%430), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%432 : Float(1, 144, 28, 28) = onnx::Conv[dilations=[1, 1], group=144, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%431, %blocks.7.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%433 : Float(1, 144, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%432, %blocks.7.mobile_inverted_conv.depth_conv.bn.weight, %blocks.7.mobile_inverted_conv.depth_conv.bn.bias, %blocks.7.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.7.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%434 : Float(1, 144, 28, 28) = onnx::Clip[max=6, min=0](%433), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%435 : Float(1, 48, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%434, %blocks.7.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%436 : Float(1, 48, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%435, %blocks.7.mobile_inverted_conv.point_linear.bn.weight, %blocks.7.mobile_inverted_conv.point_linear.bn.bias, %blocks.7.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.7.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%437 : Float(1, 48, 28, 28) = onnx::Add(%428, %436), scope: ProxylessNASNets/MobileInvertedResidualBlock
%438 : Float(1, 144, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%437, %blocks.8.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%439 : Float(1, 144, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%438, %blocks.8.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.8.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.8.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.8.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%440 : Float(1, 144, 28, 28) = onnx::Clip[max=6, min=0](%439), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%441 : Float(1, 144, 28, 28) = onnx::Conv[dilations=[1, 1], group=144, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%440, %blocks.8.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%442 : Float(1, 144, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%441, %blocks.8.mobile_inverted_conv.depth_conv.bn.weight, %blocks.8.mobile_inverted_conv.depth_conv.bn.bias, %blocks.8.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.8.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%443 : Float(1, 144, 28, 28) = onnx::Clip[max=6, min=0](%442), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%444 : Float(1, 48, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%443, %blocks.8.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%445 : Float(1, 48, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%444, %blocks.8.mobile_inverted_conv.point_linear.bn.weight, %blocks.8.mobile_inverted_conv.point_linear.bn.bias, %blocks.8.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.8.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%446 : Float(1, 48, 28, 28) = onnx::Add(%437, %445), scope: ProxylessNASNets/MobileInvertedResidualBlock
%447 : Float(1, 288, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%446, %blocks.9.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%448 : Float(1, 288, 28, 28) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%447, %blocks.9.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.9.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.9.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.9.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%449 : Float(1, 288, 28, 28) = onnx::Clip[max=6, min=0](%448), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%450 : Float(1, 288, 14, 14) = onnx::Conv[dilations=[1, 1], group=288, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%449, %blocks.9.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%451 : Float(1, 288, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%450, %blocks.9.mobile_inverted_conv.depth_conv.bn.weight, %blocks.9.mobile_inverted_conv.depth_conv.bn.bias, %blocks.9.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.9.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%452 : Float(1, 288, 14, 14) = onnx::Clip[max=6, min=0](%451), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%453 : Float(1, 88, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%452, %blocks.9.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%454 : Float(1, 88, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%453, %blocks.9.mobile_inverted_conv.point_linear.bn.weight, %blocks.9.mobile_inverted_conv.point_linear.bn.bias, %blocks.9.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.9.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%455 : Float(1, 264, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%454, %blocks.12.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%456 : Float(1, 264, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%455, %blocks.12.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.12.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.12.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.12.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%457 : Float(1, 264, 14, 14) = onnx::Clip[max=6, min=0](%456), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%458 : Float(1, 264, 14, 14) = onnx::Conv[dilations=[1, 1], group=264, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%457, %blocks.12.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%459 : Float(1, 264, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%458, %blocks.12.mobile_inverted_conv.depth_conv.bn.weight, %blocks.12.mobile_inverted_conv.depth_conv.bn.bias, %blocks.12.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.12.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%460 : Float(1, 264, 14, 14) = onnx::Clip[max=6, min=0](%459), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%461 : Float(1, 88, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%460, %blocks.12.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%462 : Float(1, 88, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%461, %blocks.12.mobile_inverted_conv.point_linear.bn.weight, %blocks.12.mobile_inverted_conv.point_linear.bn.bias, %blocks.12.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.12.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%463 : Float(1, 88, 14, 14) = onnx::Add(%454, %462), scope: ProxylessNASNets/MobileInvertedResidualBlock
%464 : Float(1, 528, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%463, %blocks.13.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%465 : Float(1, 528, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%464, %blocks.13.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.13.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.13.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.13.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%466 : Float(1, 528, 14, 14) = onnx::Clip[max=6, min=0](%465), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%467 : Float(1, 528, 14, 14) = onnx::Conv[dilations=[1, 1], group=528, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%466, %blocks.13.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%468 : Float(1, 528, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%467, %blocks.13.mobile_inverted_conv.depth_conv.bn.weight, %blocks.13.mobile_inverted_conv.depth_conv.bn.bias, %blocks.13.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.13.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%469 : Float(1, 528, 14, 14) = onnx::Clip[max=6, min=0](%468), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%470 : Float(1, 104, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%469, %blocks.13.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%471 : Float(1, 104, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%470, %blocks.13.mobile_inverted_conv.point_linear.bn.weight, %blocks.13.mobile_inverted_conv.point_linear.bn.bias, %blocks.13.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.13.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%472 : Float(1, 312, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%471, %blocks.14.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%473 : Float(1, 312, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%472, %blocks.14.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.14.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.14.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.14.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%474 : Float(1, 312, 14, 14) = onnx::Clip[max=6, min=0](%473), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%475 : Float(1, 312, 14, 14) = onnx::Conv[dilations=[1, 1], group=312, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%474, %blocks.14.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%476 : Float(1, 312, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%475, %blocks.14.mobile_inverted_conv.depth_conv.bn.weight, %blocks.14.mobile_inverted_conv.depth_conv.bn.bias, %blocks.14.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.14.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%477 : Float(1, 312, 14, 14) = onnx::Clip[max=6, min=0](%476), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%478 : Float(1, 104, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%477, %blocks.14.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%479 : Float(1, 104, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%478, %blocks.14.mobile_inverted_conv.point_linear.bn.weight, %blocks.14.mobile_inverted_conv.point_linear.bn.bias, %blocks.14.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.14.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%480 : Float(1, 104, 14, 14) = onnx::Add(%471, %479), scope: ProxylessNASNets/MobileInvertedResidualBlock
%481 : Float(1, 312, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%480, %blocks.15.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%482 : Float(1, 312, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%481, %blocks.15.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.15.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.15.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.15.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%483 : Float(1, 312, 14, 14) = onnx::Clip[max=6, min=0](%482), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%484 : Float(1, 312, 14, 14) = onnx::Conv[dilations=[1, 1], group=312, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%483, %blocks.15.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%485 : Float(1, 312, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%484, %blocks.15.mobile_inverted_conv.depth_conv.bn.weight, %blocks.15.mobile_inverted_conv.depth_conv.bn.bias, %blocks.15.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.15.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%486 : Float(1, 312, 14, 14) = onnx::Clip[max=6, min=0](%485), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%487 : Float(1, 104, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%486, %blocks.15.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%488 : Float(1, 104, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%487, %blocks.15.mobile_inverted_conv.point_linear.bn.weight, %blocks.15.mobile_inverted_conv.point_linear.bn.bias, %blocks.15.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.15.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%489 : Float(1, 104, 14, 14) = onnx::Add(%480, %488), scope: ProxylessNASNets/MobileInvertedResidualBlock
%490 : Float(1, 312, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%489, %blocks.16.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%491 : Float(1, 312, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%490, %blocks.16.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.16.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.16.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.16.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%492 : Float(1, 312, 14, 14) = onnx::Clip[max=6, min=0](%491), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%493 : Float(1, 312, 14, 14) = onnx::Conv[dilations=[1, 1], group=312, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%492, %blocks.16.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%494 : Float(1, 312, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%493, %blocks.16.mobile_inverted_conv.depth_conv.bn.weight, %blocks.16.mobile_inverted_conv.depth_conv.bn.bias, %blocks.16.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.16.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%495 : Float(1, 312, 14, 14) = onnx::Clip[max=6, min=0](%494), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%496 : Float(1, 104, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%495, %blocks.16.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%497 : Float(1, 104, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%496, %blocks.16.mobile_inverted_conv.point_linear.bn.weight, %blocks.16.mobile_inverted_conv.point_linear.bn.bias, %blocks.16.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.16.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%498 : Float(1, 104, 14, 14) = onnx::Add(%489, %497), scope: ProxylessNASNets/MobileInvertedResidualBlock
%499 : Float(1, 624, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%498, %blocks.17.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%500 : Float(1, 624, 14, 14) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%499, %blocks.17.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.17.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.17.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.17.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%501 : Float(1, 624, 14, 14) = onnx::Clip[max=6, min=0](%500), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%502 : Float(1, 624, 7, 7) = onnx::Conv[dilations=[1, 1], group=624, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[2, 2]](%501, %blocks.17.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%503 : Float(1, 624, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%502, %blocks.17.mobile_inverted_conv.depth_conv.bn.weight, %blocks.17.mobile_inverted_conv.depth_conv.bn.bias, %blocks.17.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.17.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%504 : Float(1, 624, 7, 7) = onnx::Clip[max=6, min=0](%503), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%505 : Float(1, 216, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%504, %blocks.17.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%506 : Float(1, 216, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%505, %blocks.17.mobile_inverted_conv.point_linear.bn.weight, %blocks.17.mobile_inverted_conv.point_linear.bn.bias, %blocks.17.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.17.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%507 : Float(1, 648, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%506, %blocks.18.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%508 : Float(1, 648, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%507, %blocks.18.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.18.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.18.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.18.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%509 : Float(1, 648, 7, 7) = onnx::Clip[max=6, min=0](%508), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%510 : Float(1, 648, 7, 7) = onnx::Conv[dilations=[1, 1], group=648, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%509, %blocks.18.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%511 : Float(1, 648, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%510, %blocks.18.mobile_inverted_conv.depth_conv.bn.weight, %blocks.18.mobile_inverted_conv.depth_conv.bn.bias, %blocks.18.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.18.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%512 : Float(1, 648, 7, 7) = onnx::Clip[max=6, min=0](%511), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%513 : Float(1, 216, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%512, %blocks.18.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%514 : Float(1, 216, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%513, %blocks.18.mobile_inverted_conv.point_linear.bn.weight, %blocks.18.mobile_inverted_conv.point_linear.bn.bias, %blocks.18.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.18.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%515 : Float(1, 216, 7, 7) = onnx::Add(%506, %514), scope: ProxylessNASNets/MobileInvertedResidualBlock
%516 : Float(1, 648, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%515, %blocks.19.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%517 : Float(1, 648, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%516, %blocks.19.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.19.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.19.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.19.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%518 : Float(1, 648, 7, 7) = onnx::Clip[max=6, min=0](%517), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%519 : Float(1, 648, 7, 7) = onnx::Conv[dilations=[1, 1], group=648, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%518, %blocks.19.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%520 : Float(1, 648, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%519, %blocks.19.mobile_inverted_conv.depth_conv.bn.weight, %blocks.19.mobile_inverted_conv.depth_conv.bn.bias, %blocks.19.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.19.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%521 : Float(1, 648, 7, 7) = onnx::Clip[max=6, min=0](%520), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%522 : Float(1, 216, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%521, %blocks.19.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%523 : Float(1, 216, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%522, %blocks.19.mobile_inverted_conv.point_linear.bn.weight, %blocks.19.mobile_inverted_conv.point_linear.bn.bias, %blocks.19.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.19.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%524 : Float(1, 216, 7, 7) = onnx::Add(%515, %523), scope: ProxylessNASNets/MobileInvertedResidualBlock
%525 : Float(1, 648, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%524, %blocks.20.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%526 : Float(1, 648, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%525, %blocks.20.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.20.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.20.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.20.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%527 : Float(1, 648, 7, 7) = onnx::Clip[max=6, min=0](%526), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%528 : Float(1, 648, 7, 7) = onnx::Conv[dilations=[1, 1], group=648, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%527, %blocks.20.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%529 : Float(1, 648, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%528, %blocks.20.mobile_inverted_conv.depth_conv.bn.weight, %blocks.20.mobile_inverted_conv.depth_conv.bn.bias, %blocks.20.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.20.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%530 : Float(1, 648, 7, 7) = onnx::Clip[max=6, min=0](%529), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%531 : Float(1, 216, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%530, %blocks.20.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%532 : Float(1, 216, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%531, %blocks.20.mobile_inverted_conv.point_linear.bn.weight, %blocks.20.mobile_inverted_conv.point_linear.bn.bias, %blocks.20.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.20.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%533 : Float(1, 216, 7, 7) = onnx::Add(%524, %532), scope: ProxylessNASNets/MobileInvertedResidualBlock
%534 : Float(1, 1296, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%533, %blocks.21.mobile_inverted_conv.inverted_bottleneck.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/Conv2d[conv]
%535 : Float(1, 1296, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%534, %blocks.21.mobile_inverted_conv.inverted_bottleneck.bn.weight, %blocks.21.mobile_inverted_conv.inverted_bottleneck.bn.bias, %blocks.21.mobile_inverted_conv.inverted_bottleneck.bn.running_mean, %blocks.21.mobile_inverted_conv.inverted_bottleneck.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/BatchNorm2d[bn]
%536 : Float(1, 1296, 7, 7) = onnx::Clip[max=6, min=0](%535), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[inverted_bottleneck]/ReLU6[relu]
%537 : Float(1, 1296, 7, 7) = onnx::Conv[dilations=[1, 1], group=1296, kernel_shape=[5, 5], pads=[2, 2, 2, 2], strides=[1, 1]](%536, %blocks.21.mobile_inverted_conv.depth_conv.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/Conv2d[conv]
%538 : Float(1, 1296, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%537, %blocks.21.mobile_inverted_conv.depth_conv.bn.weight, %blocks.21.mobile_inverted_conv.depth_conv.bn.bias, %blocks.21.mobile_inverted_conv.depth_conv.bn.running_mean, %blocks.21.mobile_inverted_conv.depth_conv.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/BatchNorm2d[bn]
%539 : Float(1, 1296, 7, 7) = onnx::Clip[max=6, min=0](%538), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[depth_conv]/ReLU6[relu]
%540 : Float(1, 360, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%539, %blocks.21.mobile_inverted_conv.point_linear.conv.weight), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/Conv2d[conv]
%541 : Float(1, 360, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%540, %blocks.21.mobile_inverted_conv.point_linear.bn.weight, %blocks.21.mobile_inverted_conv.point_linear.bn.bias, %blocks.21.mobile_inverted_conv.point_linear.bn.running_mean, %blocks.21.mobile_inverted_conv.point_linear.bn.running_var), scope: ProxylessNASNets/MobileInvertedResidualBlock/MBInvertedConvLayer[mobile_inverted_conv]/Sequential[point_linear]/BatchNorm2d[bn]
%542 : Float(1, 1432, 7, 7) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%541, %feature_mix_layer.conv.weight), scope: ProxylessNASNets/ConvLayer[feature_mix_layer]/Conv2d[conv]
%543 : Float(1, 1432, 7, 7) = onnx::BatchNormalization[epsilon=0.001, momentum=0.9](%542, %feature_mix_layer.bn.weight, %feature_mix_layer.bn.bias, %feature_mix_layer.bn.running_mean, %feature_mix_layer.bn.running_var), scope: ProxylessNASNets/ConvLayer[feature_mix_layer]/BatchNorm2d[bn]
%544 : Float(1, 1432, 7, 7) = onnx::Clip[max=6, min=0](%543), scope: ProxylessNASNets/ConvLayer[feature_mix_layer]/ReLU6[activation]
%545 : Float(1, 1432, 1, 1) = onnx::GlobalAveragePool(%544), scope: ProxylessNASNets/AdaptiveAvgPool2d[global_avg_pooling]
%546 : Long() = onnx::Constant[value={0}](), scope: ProxylessNASNets
%547 : Tensor = onnx::Shape(%545), scope: ProxylessNASNets
%548 : Long() = onnx::Gather[axis=0](%547, %546), scope: ProxylessNASNets
%549 : Long() = onnx::Constant[value={-1}](), scope: ProxylessNASNets
%550 : Tensor = onnx::Unsqueeze[axes=[0]](%548)
%551 : Tensor = onnx::Unsqueeze[axes=[0]](%549)
%552 : Tensor = onnx::Concat[axis=0](%550, %551)
%553 : Float(1, 1432) = onnx::Reshape(%545, %552), scope: ProxylessNASNets
%output1 : Float(1, 1000) = onnx::Gemm[alpha=1, beta=1, transB=1](%553, %classifier.linear.weight, %classifier.linear.bias), scope: ProxylessNASNets/LinearLayer[classifier]/Linear[linear]
return (%output1)
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Attribute momentum is ignored in relay.sym.batch_norm
WARNING:root:Infering Reshape argument by precompute
Traceback (most recent call last):
File "/Users/ligeng/Workspace/ProxylessNAS/load_onnx.py", line 32, in <module>
sym, params = relay.frontend.from_onnx(onnx_model, shape_dict)
File "/Users/ligeng/Workspace/tvm/python/tvm/relay/frontend/onnx.py", line 1246, in from_onnx
sym, params = g.from_onnx(graph, opset)
File "/Users/ligeng/Workspace/tvm/python/tvm/relay/frontend/onnx.py", line 1074, in from_onnx
op = self._convert_operator(op_name, inputs, attr, opset)
File "/Users/ligeng/Workspace/tvm/python/tvm/relay/frontend/onnx.py", line 1180, in _convert_operator
sym = convert_map[op_name](inputs, attrs, self._params)
File "/Users/ligeng/Workspace/tvm/python/tvm/relay/frontend/onnx.py", line 417, in _impl_v1
graph, lib, params = tvm.relay.build(func, target="llvm", params=params)
File "/Users/ligeng/Workspace/tvm/python/tvm/relay/build_module.py", line 196, in build
params)
File "/Users/ligeng/Workspace/tvm/python/tvm/relay/build_module.py", line 107, in build
self._build(func, target, target_host)
File "/Users/ligeng/Workspace/tvm/python/tvm/_ffi/_ctypes/function.py", line 209, in __call__
raise get_last_ffi_error()
tvm._ffi.base.TVMError: Traceback (most recent call last):
[bt] (8) 9 libtvm.dylib 0x000000011fa0312d tvm::relay::backend::RelayBuildModule::GetFunction(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, std::__1::shared_ptr<tvm::runtime::ModuleNode> const&)::'lambda1'(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)::operator()(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const + 429
[bt] (7) 8 libtvm.dylib 0x000000011fa03342 tvm::relay::backend::RelayBuildModule::Build(tvm::relay::Function, tvm::Map<tvm::Integer, tvm::Target, void, void> const&, tvm::Target const&) + 130
[bt] (6) 7 libtvm.dylib 0x000000011fa03549 tvm::relay::backend::RelayBuildModule::BuildRelay(tvm::relay::Function, std::__1::unordered_map<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, tvm::runtime::NDArray, std::__1::hash<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > >, std::__1::equal_to<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > >, std::__1::allocator<std::__1::pair<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const, tvm::runtime::NDArray> > > const&) + 345
[bt] (5) 6 libtvm.dylib 0x000000011fad135a tvm::relay::ModuleNode::FromExpr(tvm::relay::Expr const&, tvm::Map<tvm::relay::GlobalVar, tvm::relay::Function, void, void> const&) + 938
[bt] (4) 5 libtvm.dylib 0x000000011facfba7 tvm::relay::ModuleNode::Add(tvm::relay::GlobalVar const&, tvm::relay::Function const&, bool) + 151
[bt] (3) 4 libtvm.dylib 0x000000011fdb5848 tvm::relay::InferType(tvm::relay::Function const&, tvm::relay::Module const&, tvm::relay::GlobalVar const&) + 472
[bt] (2) 3 libtvm.dylib 0x000000011fdb4917 tvm::relay::TypeInferencer::Infer(tvm::relay::Expr) + 135
[bt] (1) 2 libtvm.dylib 0x000000011fa9beb3 tvm::relay::ErrorReporter::RenderErrors(tvm::relay::Module const&, bool) + 5555
[bt] (0) 1 libtvm.dylib 0x000000011f654e79 dmlc::LogMessageFatal::~LogMessageFatal() + 57
[bt] (8) 9 libtvm.dylib 0x000000011facfba7 tvm::relay::ModuleNode::Add(tvm::relay::GlobalVar const&, tvm::relay::Function const&, bool) + 151
[bt] (7) 8 libtvm.dylib 0x000000011fdb5848 tvm::relay::InferType(tvm::relay::Function const&, tvm::relay::Module const&, tvm::relay::GlobalVar const&) + 472
[bt] (6) 7 libtvm.dylib 0x000000011fdb48fb tvm::relay::TypeInferencer::Infer(tvm::relay::Expr) + 107
[bt] (5) 6 libtvm.dylib 0x000000011fdd150a tvm::relay::TypeSolver::Solve() + 1114
[bt] (4) 5 libtvm.dylib 0x000000011fdd1ba8 tvm::TypedEnvFunc<bool (tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&)>::operator()(tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&) const + 328
[bt] (3) 4 libtvm.dylib 0x000000011fb23c39 std::__1::__function::__func<void tvm::runtime::TypedPackedFunc<bool (tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&)>::AssignTypedLambda<bool (*)(tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&)>(bool (*)(tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&))::'lambda'(tvm::runtime::TVMArgs const&, tvm::runtime::TVMRetValue*), std::__1::allocator<void tvm::runtime::TypedPackedFunc<bool (tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&)>::AssignTypedLambda<bool (*)(tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&)>(bool (*)(tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&))::'lambda'(tvm::runtime::TVMArgs const&, tvm::runtime::TVMRetValue*)>, void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)>::operator()(tvm::runtime::TVMArgs&&, tvm::runtime::TVMRetValue*&&) + 137
[bt] (2) 3 libtvm.dylib 0x000000011fb23cdf void tvm::runtime::detail::unpack_call_dispatcher<bool, 0, 4, bool (*)(tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&)>::run<tvm::runtime::TVMArgValue, tvm::runtime::TVMArgValue, tvm::runtime::TVMArgValue, tvm::runtime::TVMArgValue>(bool (* const&)(tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&), tvm::runtime::TVMArgs const&, tvm::runtime::TVMRetValue*, tvm::runtime::TVMArgValue&&, tvm::runtime::TVMArgValue&&, tvm::runtime::TVMArgValue&&, tvm::runtime::TVMArgValue&&) + 95
[bt] (1) 2 libtvm.dylib 0x000000011fc3182e tvm::relay::ConcatenateRel(tvm::Array<tvm::relay::Type, void> const&, int, tvm::Attrs const&, tvm::relay::TypeReporter const&) + 1918
[bt] (0) 1 libtvm.dylib 0x000000011f654e79 dmlc::LogMessageFatal::~LogMessageFatal() + 57
File "/Users/ligeng/Workspace/tvm/src/relay/ir/error.cc", line 132
TVMError:
Error(s) have occurred. We have annotated the program with them:
In `main`:
v0.0.1
fn (%actual_input_1: Tensor[(1, 3, 224, 224), float32]) {
%0 = nn.conv2d(%actual_input_1, meta[relay.Constant][0], strides=[2, 2], padding=[1, 1], kernel_size=[3, 3])
%1 = nn.batch_norm(%0, meta[relay.Constant][1], meta[relay.Constant][2], meta[relay.Constant][3], meta[relay.Constant][4], epsilon=0.001)
%2 = %1.0
%3 = clip(%2, a_min=0, a_max=6)
%4 = nn.conv2d(%3, meta[relay.Constant][5], padding=[1, 1], groups=40, kernel_size=[3, 3])
%5 = nn.batch_norm(%4, meta[relay.Constant][6], meta[relay.Constant][7], meta[relay.Constant][8], meta[relay.Constant][9], epsilon=0.001)
%6 = %5.0
%7 = clip(%6, a_min=0, a_max=6)
%8 = nn.conv2d(%7, meta[relay.Constant][10], kernel_size=[1, 1])
%9 = nn.batch_norm(%8, meta[relay.Constant][11], meta[relay.Constant][12], meta[relay.Constant][13], meta[relay.Constant][14], epsilon=0.001)
%10 = %9.0
%11 = nn.conv2d(%10, meta[relay.Constant][15], kernel_size=[1, 1])
%12 = nn.batch_norm(%11, meta[relay.Constant][16], meta[relay.Constant][17], meta[relay.Constant][18], meta[relay.Constant][19], epsilon=0.001)
%13 = %12.0
%14 = clip(%13, a_min=0, a_max=6)
%15 = nn.conv2d(%14, meta[relay.Constant][20], strides=[2, 2], padding=[1, 1], groups=144, kernel_size=[3, 3])
%16 = nn.batch_norm(%15, meta[relay.Constant][21], meta[relay.Constant][22], meta[relay.Constant][23], meta[relay.Constant][24], epsilon=0.001)
%17 = %16.0
%18 = clip(%17, a_min=0, a_max=6)
%19 = nn.conv2d(%18, meta[relay.Constant][25], kernel_size=[1, 1])
%20 = nn.batch_norm(%19, meta[relay.Constant][26], meta[relay.Constant][27], meta[relay.Constant][28], meta[relay.Constant][29], epsilon=0.001)
%21 = %20.0
%22 = nn.conv2d(%21, meta[relay.Constant][30], kernel_size=[1, 1])
%23 = nn.batch_norm(%22, meta[relay.Constant][31], meta[relay.Constant][32], meta[relay.Constant][33], meta[relay.Constant][34], epsilon=0.001)
%24 = %23.0
%25 = clip(%24, a_min=0, a_max=6)
%26 = nn.conv2d(%25, meta[relay.Constant][35], padding=[1, 1], groups=96, kernel_size=[3, 3])
%27 = nn.batch_norm(%26, meta[relay.Constant][36], meta[relay.Constant][37], meta[relay.Constant][38], meta[relay.Constant][39], epsilon=0.001)
%28 = %27.0
%29 = clip(%28, a_min=0, a_max=6)
%30 = nn.conv2d(%29, meta[relay.Constant][40], kernel_size=[1, 1])
%31 = nn.batch_norm(%30, meta[relay.Constant][41], meta[relay.Constant][42], meta[relay.Constant][43], meta[relay.Constant][44], epsilon=0.001)
%32 = %31.0
%33 = add(%21, %32)
%34 = nn.conv2d(%33, meta[relay.Constant][45], kernel_size=[1, 1])
%35 = nn.batch_norm(%34, meta[relay.Constant][46], meta[relay.Constant][47], meta[relay.Constant][48], meta[relay.Constant][49], epsilon=0.001)
%36 = %35.0
%37 = clip(%36, a_min=0, a_max=6)
%38 = nn.conv2d(%37, meta[relay.Constant][50], padding=[1, 1], groups=96, kernel_size=[3, 3])
%39 = nn.batch_norm(%38, meta[relay.Constant][51], meta[relay.Constant][52], meta[relay.Constant][53], meta[relay.Constant][54], epsilon=0.001)
%40 = %39.0
%41 = clip(%40, a_min=0, a_max=6)
%42 = nn.conv2d(%41, meta[relay.Constant][55], kernel_size=[1, 1])
%43 = nn.batch_norm(%42, meta[relay.Constant][56], meta[relay.Constant][57], meta[relay.Constant][58], meta[relay.Constant][59], epsilon=0.001)
%44 = %43.0
%45 = add(%33, %44)
%46 = nn.conv2d(%45, meta[relay.Constant][60], kernel_size=[1, 1])
%47 = nn.batch_norm(%46, meta[relay.Constant][61], meta[relay.Constant][62], meta[relay.Constant][63], meta[relay.Constant][64], epsilon=0.001)
%48 = %47.0
%49 = clip(%48, a_min=0, a_max=6)
%50 = nn.conv2d(%49, meta[relay.Constant][65], padding=[1, 1], groups=96, kernel_size=[3, 3])
%51 = nn.batch_norm(%50, meta[relay.Constant][66], meta[relay.Constant][67], meta[relay.Constant][68], meta[relay.Constant][69], epsilon=0.001)
%52 = %51.0
%53 = clip(%52, a_min=0, a_max=6)
%54 = nn.conv2d(%53, meta[relay.Constant][70], kernel_size=[1, 1])
%55 = nn.batch_norm(%54, meta[relay.Constant][71], meta[relay.Constant][72], meta[relay.Constant][73], meta[relay.Constant][74], epsilon=0.001)
%56 = %55.0
%57 = add(%45, %56)
%58 = nn.conv2d(%57, meta[relay.Constant][75], kernel_size=[1, 1])
%59 = nn.batch_norm(%58, meta[relay.Constant][76], meta[relay.Constant][77], meta[relay.Constant][78], meta[relay.Constant][79], epsilon=0.001)
%60 = %59.0
%61 = clip(%60, a_min=0, a_max=6)
%62 = nn.conv2d(%61, meta[relay.Constant][80], strides=[2, 2], padding=[1, 1], groups=192, kernel_size=[3, 3])
%63 = nn.batch_norm(%62, meta[relay.Constant][81], meta[relay.Constant][82], meta[relay.Constant][83], meta[relay.Constant][84], epsilon=0.001)
%64 = %63.0
%65 = clip(%64, a_min=0, a_max=6)
%66 = nn.conv2d(%65, meta[relay.Constant][85], kernel_size=[1, 1])
%67 = nn.batch_norm(%66, meta[relay.Constant][86], meta[relay.Constant][87], meta[relay.Constant][88], meta[relay.Constant][89], epsilon=0.001)
%68 = %67.0
%69 = nn.conv2d(%68, meta[relay.Constant][90], kernel_size=[1, 1])
%70 = nn.batch_norm(%69, meta[relay.Constant][91], meta[relay.Constant][92], meta[relay.Constant][93], meta[relay.Constant][94], epsilon=0.001)
%71 = %70.0
%72 = clip(%71, a_min=0, a_max=6)
%73 = nn.conv2d(%72, meta[relay.Constant][95], padding=[1, 1], groups=144, kernel_size=[3, 3])
%74 = nn.batch_norm(%73, meta[relay.Constant][96], meta[relay.Constant][97], meta[relay.Constant][98], meta[relay.Constant][99], epsilon=0.001)
%75 = %74.0
%76 = clip(%75, a_min=0, a_max=6)
%77 = nn.conv2d(%76, meta[relay.Constant][100], kernel_size=[1, 1])
%78 = nn.batch_norm(%77, meta[relay.Constant][101], meta[relay.Constant][102], meta[relay.Constant][103], meta[relay.Constant][104], epsilon=0.001)
%79 = %78.0
%80 = add(%68, %79)
%81 = nn.conv2d(%80, meta[relay.Constant][105], kernel_size=[1, 1])
%82 = nn.batch_norm(%81, meta[relay.Constant][106], meta[relay.Constant][107], meta[relay.Constant][108], meta[relay.Constant][109], epsilon=0.001)
%83 = %82.0
%84 = clip(%83, a_min=0, a_max=6)
%85 = nn.conv2d(%84, meta[relay.Constant][110], padding=[1, 1], groups=144, kernel_size=[3, 3])
%86 = nn.batch_norm(%85, meta[relay.Constant][111], meta[relay.Constant][112], meta[relay.Constant][113], meta[relay.Constant][114], epsilon=0.001)
%87 = %86.0
%88 = clip(%87, a_min=0, a_max=6)
%89 = nn.conv2d(%88, meta[relay.Constant][115], kernel_size=[1, 1])
%90 = nn.batch_norm(%89, meta[relay.Constant][116], meta[relay.Constant][117], meta[relay.Constant][118], meta[relay.Constant][119], epsilon=0.001)
%91 = %90.0
%92 = add(%80, %91)
%93 = nn.conv2d(%92, meta[relay.Constant][120], kernel_size=[1, 1])
%94 = nn.batch_norm(%93, meta[relay.Constant][121], meta[relay.Constant][122], meta[relay.Constant][123], meta[relay.Constant][124], epsilon=0.001)
%95 = %94.0
%96 = clip(%95, a_min=0, a_max=6)
%97 = nn.conv2d(%96, meta[relay.Constant][125], padding=[2, 2], groups=144, kernel_size=[5, 5])
%98 = nn.batch_norm(%97, meta[relay.Constant][126], meta[relay.Constant][127], meta[relay.Constant][128], meta[relay.Constant][129], epsilon=0.001)
%99 = %98.0
%100 = clip(%99, a_min=0, a_max=6)
%101 = nn.conv2d(%100, meta[relay.Constant][130], kernel_size=[1, 1])
%102 = nn.batch_norm(%101, meta[relay.Constant][131], meta[relay.Constant][132], meta[relay.Constant][133], meta[relay.Constant][134], epsilon=0.001)
%103 = %102.0
%104 = add(%92, %103)
%105 = nn.conv2d(%104, meta[relay.Constant][135], kernel_size=[1, 1])
%106 = nn.batch_norm(%105, meta[relay.Constant][136], meta[relay.Constant][137], meta[relay.Constant][138], meta[relay.Constant][139], epsilon=0.001)
%107 = %106.0
%108 = clip(%107, a_min=0, a_max=6)
%109 = nn.conv2d(%108, meta[relay.Constant][140], strides=[2, 2], padding=[1, 1], groups=288, kernel_size=[3, 3])
%110 = nn.batch_norm(%109, meta[relay.Constant][141], meta[relay.Constant][142], meta[relay.Constant][143], meta[relay.Constant][144], epsilon=0.001)
%111 = %110.0
%112 = clip(%111, a_min=0, a_max=6)
%113 = nn.conv2d(%112, meta[relay.Constant][145], kernel_size=[1, 1])
%114 = nn.batch_norm(%113, meta[relay.Constant][146], meta[relay.Constant][147], meta[relay.Constant][148], meta[relay.Constant][149], epsilon=0.001)
%115 = %114.0
%116 = nn.conv2d(%115, meta[relay.Constant][150], kernel_size=[1, 1])
%117 = nn.batch_norm(%116, meta[relay.Constant][151], meta[relay.Constant][152], meta[relay.Constant][153], meta[relay.Constant][154], epsilon=0.001)
%118 = %117.0
%119 = clip(%118, a_min=0, a_max=6)
%120 = nn.conv2d(%119, meta[relay.Constant][155], padding=[1, 1], groups=264, kernel_size=[3, 3])
%121 = nn.batch_norm(%120, meta[relay.Constant][156], meta[relay.Constant][157], meta[relay.Constant][158], meta[relay.Constant][159], epsilon=0.001)
%122 = %121.0
%123 = clip(%122, a_min=0, a_max=6)
%124 = nn.conv2d(%123, meta[relay.Constant][160], kernel_size=[1, 1])
%125 = nn.batch_norm(%124, meta[relay.Constant][161], meta[relay.Constant][162], meta[relay.Constant][163], meta[relay.Constant][164], epsilon=0.001)
%126 = %125.0
%127 = add(%115, %126)
%128 = nn.conv2d(%127, meta[relay.Constant][165], kernel_size=[1, 1])
%129 = nn.batch_norm(%128, meta[relay.Constant][166], meta[relay.Constant][167], meta[relay.Constant][168], meta[relay.Constant][169], epsilon=0.001)
%130 = %129.0
%131 = clip(%130, a_min=0, a_max=6)
%132 = nn.conv2d(%131, meta[relay.Constant][170], padding=[2, 2], groups=528, kernel_size=[5, 5])
%133 = nn.batch_norm(%132, meta[relay.Constant][171], meta[relay.Constant][172], meta[relay.Constant][173], meta[relay.Constant][174], epsilon=0.001)
%134 = %133.0
%135 = clip(%134, a_min=0, a_max=6)
%136 = nn.conv2d(%135, meta[relay.Constant][175], kernel_size=[1, 1])
%137 = nn.batch_norm(%136, meta[relay.Constant][176], meta[relay.Constant][177], meta[relay.Constant][178], meta[relay.Constant][179], epsilon=0.001)
%138 = %137.0
%139 = nn.conv2d(%138, meta[relay.Constant][180], kernel_size=[1, 1])
%140 = nn.batch_norm(%139, meta[relay.Constant][181], meta[relay.Constant][182], meta[relay.Constant][183], meta[relay.Constant][184], epsilon=0.001)
%141 = %140.0
%142 = clip(%141, a_min=0, a_max=6)
%143 = nn.conv2d(%142, meta[relay.Constant][185], padding=[1, 1], groups=312, kernel_size=[3, 3])
%144 = nn.batch_norm(%143, meta[relay.Constant][186], meta[relay.Constant][187], meta[relay.Constant][188], meta[relay.Constant][189], epsilon=0.001)
%145 = %144.0
%146 = clip(%145, a_min=0, a_max=6)
%147 = nn.conv2d(%146, meta[relay.Constant][190], kernel_size=[1, 1])
%148 = nn.batch_norm(%147, meta[relay.Constant][191], meta[relay.Constant][192], meta[relay.Constant][193], meta[relay.Constant][194], epsilon=0.001)
%149 = %148.0
%150 = add(%138, %149)
%151 = nn.conv2d(%150, meta[relay.Constant][195], kernel_size=[1, 1])
%152 = nn.batch_norm(%151, meta[relay.Constant][196], meta[relay.Constant][197], meta[relay.Constant][198], meta[relay.Constant][199], epsilon=0.001)
%153 = %152.0
%154 = clip(%153, a_min=0, a_max=6)
%155 = nn.conv2d(%154, meta[relay.Constant][200], padding=[1, 1], groups=312, kernel_size=[3, 3])
%156 = nn.batch_norm(%155, meta[relay.Constant][201], meta[relay.Constant][202], meta[relay.Constant][203], meta[relay.Constant][204], epsilon=0.001)
%157 = %156.0
%158 = clip(%157, a_min=0, a_max=6)
%159 = nn.conv2d(%158, meta[relay.Constant][205], kernel_size=[1, 1])
%160 = nn.batch_norm(%159, meta[relay.Constant][206], meta[relay.Constant][207], meta[relay.Constant][208], meta[relay.Constant][209], epsilon=0.001)
%161 = %160.0
%162 = add(%150, %161)
%163 = nn.conv2d(%162, meta[relay.Constant][210], kernel_size=[1, 1])
%164 = nn.batch_norm(%163, meta[relay.Constant][211], meta[relay.Constant][212], meta[relay.Constant][213], meta[relay.Constant][214], epsilon=0.001)
%165 = %164.0
%166 = clip(%165, a_min=0, a_max=6)
%167 = nn.conv2d(%166, meta[relay.Constant][215], padding=[1, 1], groups=312, kernel_size=[3, 3])
%168 = nn.batch_norm(%167, meta[relay.Constant][216], meta[relay.Constant][217], meta[relay.Constant][218], meta[relay.Constant][219], epsilon=0.001)
%169 = %168.0
%170 = clip(%169, a_min=0, a_max=6)
%171 = nn.conv2d(%170, meta[relay.Constant][220], kernel_size=[1, 1])
%172 = nn.batch_norm(%171, meta[relay.Constant][221], meta[relay.Constant][222], meta[relay.Constant][223], meta[relay.Constant][224], epsilon=0.001)
%173 = %172.0
%174 = add(%162, %173)
%175 = nn.conv2d(%174, meta[relay.Constant][225], kernel_size=[1, 1])
%176 = nn.batch_norm(%175, meta[relay.Constant][226], meta[relay.Constant][227], meta[relay.Constant][228], meta[relay.Constant][229], epsilon=0.001)
%177 = %176.0
%178 = clip(%177, a_min=0, a_max=6)
%179 = nn.conv2d(%178, meta[relay.Constant][230], strides=[2, 2], padding=[2, 2], groups=624, kernel_size=[5, 5])
%180 = nn.batch_norm(%179, meta[relay.Constant][231], meta[relay.Constant][232], meta[relay.Constant][233], meta[relay.Constant][234], epsilon=0.001)
%181 = %180.0
%182 = clip(%181, a_min=0, a_max=6)
%183 = nn.conv2d(%182, meta[relay.Constant][235], kernel_size=[1, 1])
%184 = nn.batch_norm(%183, meta[relay.Constant][236], meta[relay.Constant][237], meta[relay.Constant][238], meta[relay.Constant][239], epsilon=0.001)
%185 = %184.0
%186 = nn.conv2d(%185, meta[relay.Constant][240], kernel_size=[1, 1])
%187 = nn.batch_norm(%186, meta[relay.Constant][241], meta[relay.Constant][242], meta[relay.Constant][243], meta[relay.Constant][244], epsilon=0.001)
%188 = %187.0
%189 = clip(%188, a_min=0, a_max=6)
%190 = nn.conv2d(%189, meta[relay.Constant][245], padding=[2, 2], groups=648, kernel_size=[5, 5])
%191 = nn.batch_norm(%190, meta[relay.Constant][246], meta[relay.Constant][247], meta[relay.Constant][248], meta[relay.Constant][249], epsilon=0.001)
%192 = %191.0
%193 = clip(%192, a_min=0, a_max=6)
%194 = nn.conv2d(%193, meta[relay.Constant][250], kernel_size=[1, 1])
%195 = nn.batch_norm(%194, meta[relay.Constant][251], meta[relay.Constant][252], meta[relay.Constant][253], meta[relay.Constant][254], epsilon=0.001)
%196 = %195.0
%197 = add(%185, %196)
%198 = nn.conv2d(%197, meta[relay.Constant][255], kernel_size=[1, 1])
%199 = nn.batch_norm(%198, meta[relay.Constant][256], meta[relay.Constant][257], meta[relay.Constant][258], meta[relay.Constant][259], epsilon=0.001)
%200 = %199.0
%201 = clip(%200, a_min=0, a_max=6)
%202 = nn.conv2d(%201, meta[relay.Constant][260], padding=[2, 2], groups=648, kernel_size=[5, 5])
%203 = nn.batch_norm(%202, meta[relay.Constant][261], meta[relay.Constant][262], meta[relay.Constant][263], meta[relay.Constant][264], epsilon=0.001)
%204 = %203.0
%205 = clip(%204, a_min=0, a_max=6)
%206 = nn.conv2d(%205, meta[relay.Constant][265], kernel_size=[1, 1])
%207 = nn.batch_norm(%206, meta[relay.Constant][266], meta[relay.Constant][267], meta[relay.Constant][268], meta[relay.Constant][269], epsilon=0.001)
%208 = %207.0
%209 = add(%197, %208)
%210 = nn.conv2d(%209, meta[relay.Constant][270], kernel_size=[1, 1])
%211 = nn.batch_norm(%210, meta[relay.Constant][271], meta[relay.Constant][272], meta[relay.Constant][273], meta[relay.Constant][274], epsilon=0.001)
%212 = %211.0
%213 = clip(%212, a_min=0, a_max=6)
%214 = nn.conv2d(%213, meta[relay.Constant][275], padding=[1, 1], groups=648, kernel_size=[3, 3])
%215 = nn.batch_norm(%214, meta[relay.Constant][276], meta[relay.Constant][277], meta[relay.Constant][278], meta[relay.Constant][279], epsilon=0.001)
%216 = %215.0
%217 = clip(%216, a_min=0, a_max=6)
%218 = nn.conv2d(%217, meta[relay.Constant][280], kernel_size=[1, 1])
%219 = nn.batch_norm(%218, meta[relay.Constant][281], meta[relay.Constant][282], meta[relay.Constant][283], meta[relay.Constant][284], epsilon=0.001)
%220 = %219.0
%221 = add(%209, %220)
%222 = nn.conv2d(%221, meta[relay.Constant][285], kernel_size=[1, 1])
%223 = nn.batch_norm(%222, meta[relay.Constant][286], meta[relay.Constant][287], meta[relay.Constant][288], meta[relay.Constant][289], epsilon=0.001)
%224 = %223.0
%225 = clip(%224, a_min=0, a_max=6)
%226 = nn.conv2d(%225, meta[relay.Constant][290], padding=[2, 2], groups=1296, kernel_size=[5, 5])
%227 = nn.batch_norm(%226, meta[relay.Constant][291], meta[relay.Constant][292], meta[relay.Constant][293], meta[relay.Constant][294], epsilon=0.001)
%228 = %227.0
%229 = clip(%228, a_min=0, a_max=6)
%230 = nn.conv2d(%229, meta[relay.Constant][295], kernel_size=[1, 1])
%231 = nn.batch_norm(%230, meta[relay.Constant][296], meta[relay.Constant][297], meta[relay.Constant][298], meta[relay.Constant][299], epsilon=0.001)
%232 = %231.0
%233 = nn.conv2d(%232, meta[relay.Constant][300], kernel_size=[1, 1])
%234 = nn.batch_norm(%233, meta[relay.Constant][301], meta[relay.Constant][302], meta[relay.Constant][303], meta[relay.Constant][304], epsilon=0.001)
%235 = %234.0
%236 = clip(%235, a_min=0, a_max=6)
%237 = nn.global_avg_pool2d(%236)
%238 = shape_of(%237, dtype="int32")
%239 = take(%238, int64(0), axis=0)
%240 = expand_dims(%239, axis=0)
%241 = expand_dims(int64(-1), axis=0)
%242 = (%240, %241)
concatenate(%242)an internal invariant was violated while typechecking your program [00:12:52] /Users/ligeng/Workspace/tvm/src/relay/op/tensor/transform.cc:204: Check failed: e_dtype == dtype (int64 vs. int32) : relay.concatenate requires all tensors have the same dtype
;
}
// meta data omitted. you can use show_meta_data=True to include meta data
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment