nni.compression.pytorch.speedup.error_code.UnBalancedGroupError: The number remained filters in each group is different

[qi657]

Describe the issue:
Cannot speed up depthwise separable convolutions
[2022-05-16 11:22:40] INFO (nni.compression.pytorch.speedup.compressor/MainThread) replace module (name: _backbone._block2.1._dw_conv.0, op_type: Conv2d)
Traceback (most recent call last):
File "D:/nni-master/examples/model_compress/pruning/FD_Prune/FD_simulated_prune.py", line 278, in
pruner.compress()
File "D:\nni-master\nni\algorithms\compression\v2\pytorch\base\scheduler.py", line 194, in compress
task_result = self.pruning_one_step(task)
File "D:\nni-master\nni\algorithms\compression\v2\pytorch\pruning\basic_scheduler.py", line 154, in pruning_one_step
result = self.pruning_one_step_normal(task)
File "D:\nni-master\nni\algorithms\compression\v2\pytorch\pruning\basic_scheduler.py", line 77, in pruning_one_step_normal
ModelSpeedup(compact_model, self.dummy_input, pruner_generated_masks).speedup_model()
File "D:\nni-master\nni\compression\pytorch\speedup\compressor.py", line 519, in speedup_model
self.replace_compressed_modules()
File "D:\nni-master\nni\compression\pytorch\speedup\compressor.py", line 386, in replace_compressed_modules
self.replace_submodule(unique_name)
File "D:\nni-master\nni\compression\pytorch\speedup\compressor.py", line 450, in replace_submodule
leaf_module, auto_infer.get_masks())
File "D:\nni-master\nni\compression\pytorch\speedup\compress_modules.py", line 14, in
'Conv2d': lambda module, masks: replace_conv2d(module, masks),
File "D:\nni-master\nni\compression\pytorch\speedup\compress_modules.py", line 376, in replace_conv2d
raise UnBalancedGroupError()
nni.compression.pytorch.speedup.error_code.UnBalancedGroupError: The number remained filters in each group is different

Environment:

• NNI version: 2.7
• Training service (local|remote|pai|aml|etc):local
• Client OS:
• Server OS (for remote mode only):
• Python version:3.7
• PyTorch/TensorFlow version:pytorch 1.10.1
• Is conda/virtualenv/venv used?:yes
• Is running in Docker?:no

Configuration:

• Experiment config (remember to remove secrets!):
• Search space:

Log message:

• nnimanager.log:
• dispatcher.log:
• nnictl stdout and stderr:

How to reproduce it?:

[J-shang]

hello @qi657 , a pr will be submitted as soon to fix your issue, we will let you know when it's done.

[scarlett2018]

hello @qi657 , a pr will be submitted as soon to fix your issue, we will let you know when it's done.

@J-shang the fix has released to 2.8, right? shall we close this issue?

[Triple-L]

using nni 2.8, still facing the same bug when pruning with SlimmerPruner

[nnzs9248]

using nni 2.10, still facing the same bug when pruning

[vladgo329]

2.10 the issue is still reproduced for all depthwise convolutions.