aten::expand is not Supported! Please report.

[Zejun-Yang]

Describe the issue:
As the title of this issue, this error occured when I pruned the GAST-Net (https://github.com/fabro66/GAST-Net-3DPoseEstimation) with NNI L2FilterPruner. It seems that the operation of aten::expand is still not supported. So the mask of this layer is an empty list, which is not allowed by torch.cat().
If possible, please fix this issue as soon. Thanks!

Here is part of the layers of the model.
+-------+----------------------------------------------------------------------+--------+--------------------+-----------+---------+
| Index | Name | Type | Weight Shape | FLOPs | #Params |
+-------+----------------------------------------------------------------------+--------+--------------------+-----------+---------+
| 0 | expand_conv | Conv2d | (64, 3, 3, 1) | 955584 | 576 |
| 1 | layers_graph_conv.0.local_graph_layer.cat_conv | Conv2d | (64, 128, 1, 1) | 13590528 | 8192 |
| 2 | layers_graph_conv.0.global_graph_layer.attentions.0.g | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 3 | layers_graph_conv.0.global_graph_layer.attentions.0.theta | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 4 | layers_graph_conv.0.global_graph_layer.attentions.0.phi | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 5 | layers_graph_conv.0.global_graph_layer.attentions.0.concat_project.0 | Conv2d | (1, 32, 1, 1) | 14112 | 32 |
| 6 | layers_graph_conv.0.global_graph_layer.attentions.1.g | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 7 | layers_graph_conv.0.global_graph_layer.attentions.1.theta | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 8 | layers_graph_conv.0.global_graph_layer.attentions.1.phi | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 9 | layers_graph_conv.0.global_graph_layer.attentions.1.concat_project.0 | Conv2d | (1, 32, 1, 1) | 14112 | 32 |
| 10 | layers_graph_conv.0.global_graph_layer.attentions.2.g | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 11 | layers_graph_conv.0.global_graph_layer.attentions.2.theta | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 12 | layers_graph_conv.0.global_graph_layer.attentions.2.phi | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 13 | layers_graph_conv.0.global_graph_layer.attentions.2.concat_project.0 | Conv2d | (1, 32, 1, 1) | 14112 | 32 |
| 14 | layers_graph_conv.0.global_graph_layer.attentions.3.g | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 15 | layers_graph_conv.0.global_graph_layer.attentions.3.theta | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 16 | layers_graph_conv.0.global_graph_layer.attentions.3.phi | Conv1d | (16, 64, 1) | 21504 | 1040 |
| 17 | layers_graph_conv.0.global_graph_layer.attentions.3.concat_project.0 | Conv2d | (1, 32, 1, 1) | 14112 | 32 |
| 18 | layers_graph_conv.0.global_graph_layer.cat_conv | Conv2d | (64, 64, 1, 1) | 6795264 | 4096 |
| 19 | layers_graph_conv.0.cat_conv | Conv2d | (128, 192, 1, 1) | 40771584 | 24576 |
| 20 | layers_conv.0 | Conv2d | (128, 128, 3, 1) | 75350016 | 49152 |
| 21 | layers_conv.1 | Conv2d | (128, 128, 1, 1) | 25116672 | 16384 |
......

Environment:

• NNI version: 2.5
• Training service (local|remote|pai|aml|etc):
• Client OS: Ubuntu 18
• Server OS (for remote mode only):
• Python version: 3.7
• PyTorch/TensorFlow version: Pytorch 1.7.1
• Is conda/virtualenv/venv used?: Yes
• Is running in Docker?: No

Configuration:

• Experiment config (remember to remove secrets!): default config of NNI pruner
• Search space:

Log message:

• nnimanager.log:
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.0.aten::permute.238
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.0.aten::size.241
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.0.aten::size.242
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.1.aten::permute.258
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.1.aten::size.261
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.1.aten::size.262
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.2.aten::permute.278
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.2.aten::size.281
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.2.aten::size.282
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.3.aten::permute.298
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.3.aten::size.301
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.3.aten::size.302
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.0.aten::expand.244
  [2021-11-21 16:07:29] ERROR (nni.compression.pytorch.speedup.jit_translate/MainThread) aten::expand is not Supported! Please report an issue at https://github.com/microsoft/nni. Thanks~
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.0.aten::expand.243
  [2021-11-21 16:07:29] ERROR (nni.compression.pytorch.speedup.jit_translate/MainThread) aten::expand is not Supported! Please report an issue at https://github.com/microsoft/nni. Thanks~
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.1.aten::expand.264
  [2021-11-21 16:07:29] ERROR (nni.compression.pytorch.speedup.jit_translate/MainThread) aten::expand is not Supported! Please report an issue at https://github.com/microsoft/nni. Thanks~
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.1.aten::expand.263
  [2021-11-21 16:07:29] ERROR (nni.compression.pytorch.speedup.jit_translate/MainThread) aten::expand is not Supported! Please report an issue at https://github.com/microsoft/nni. Thanks~
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.2.aten::expand.284
  [2021-11-21 16:07:29] ERROR (nni.compression.pytorch.speedup.jit_translate/MainThread) aten::expand is not Supported! Please report an issue at https://github.com/microsoft/nni. Thanks~
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.2.aten::expand.283
  [2021-11-21 16:07:29] ERROR (nni.compression.pytorch.speedup.jit_translate/MainThread) aten::expand is not Supported! Please report an issue at https://github.com/microsoft/nni. Thanks~
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.3.aten::expand.304
  [2021-11-21 16:07:29] ERROR (nni.compression.pytorch.speedup.jit_translate/MainThread) aten::expand is not Supported! Please report an issue at https://github.com/microsoft/nni. Thanks~
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.3.aten::expand.303
  [2021-11-21 16:07:29] ERROR (nni.compression.pytorch.speedup.jit_translate/MainThread) aten::expand is not Supported! Please report an issue at https://github.com/microsoft/nni. Thanks~
  [2021-11-21 16:07:29] INFO (nni.compression.pytorch.speedup.compressor/MainThread) Update mask for layers_graph_conv.0.global_graph_layer.attentions.0.aten::cat.245
• dispatcher.log:
• nnictl stdout and stderr:

How to reproduce it?:

1. Download GAST-Net (https://github.com/fabro66/GAST-Net-3DPoseEstimation).

2. Use this function to create a GAST-Net model.
   def load_gastnet():
   from main import *
   from common.skeleton import Skeleton
   from common.graph_utils import adj_mx_from_skeleton

   ntu_skeleton = Skeleton(parents=[-1, 0, 20, 2, 20, 4, 5, 6, 20, 8, 9, 10, 0, 12, 13, 14, 0, 16, 17, 18, 1],
   joints_left=[4, 5, 6, 7, 12, 13, 14, 15],
   joints_right=[8, 9, 10, 11, 16, 17, 18, 19])

   adj = adj_mx_from_skeleton(ntu_skeleton)
   dp = 0.05
   filters_width = [3, 3, 3, 3]
   channels = 64
   model_pos = SpatioTemporalModel(adj, num_joints_in=21, in_features=3, num_joints_out=21,
   filter_widths=filters_width, causal=False,
   channels=channels, dropout=dp)
   return model_pos

3. input_size = [1, 81, 21, 3]

4. Use L2FilterPruner to prune the model and speed it up.

[ymkzpx]

Hi, @scutzzj :
Can you show me the complete source code? I try to reproduce this problem and solve it.

[Zejun-Yang]

Hi, @littlelittlehorse :
I have invited you to my project, and you can run the gast_nni_pruning.py to reproduce the problem.
Thanks for your attention.

Here is the website.
https://github.com/scutzzj/RGB-D_3d_pose_estimation/blob/main/gast_nni_pruning.py