how to except last layers in SensitivityPruner

[chasingw]

Describe the issue:
I use config_list set the op_names which I want to prune. but it seems not work because last layer is also pruned.
the config_list is

config_list = [
    {
      'sparsity': args.sparsity,
      'op_types': ['Conv2d'],
      'op_names': ['backbone_2d.blocks.0.1',
                  'backbone_2d.blocks.0.4',
                  'backbone_2d.blocks.0.7',
                  'backbone_2d.blocks.0.10',
                  'backbone_2d.blocks.1.1',
                  'backbone_2d.blocks.1.4',
                  'backbone_2d.blocks.1.7',
                  'backbone_2d.blocks.1.10',
                  'backbone_2d.blocks.1.13',
                  'backbone_2d.blocks.1.16',
                  'backbone_2d.blocks.2.1',
                  'backbone_2d.blocks.2.4',
                  'backbone_2d.blocks.2.7',
                  'backbone_2d.blocks.2.10',
                  'backbone_2d.blocks.2.13',
                  'backbone_2d.blocks.2.16']
  }]

and the output is

[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.0.1 remain/total: 47/64
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.0.4 remain/total: 50/64
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.0.7 remain/total: 53/64
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.0.10 remain/total: 47/64
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.1.1 remain/total: 95/128
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.1.4 remain/total: 108/128
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.1.7 remain/total: 95/128
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.1.10 remain/total: 85/128
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.1.13 remain/total: 95/128
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.1.16 remain/total: 95/128
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.2.1 remain/total: 165/256
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.2.4 remain/total: 127/256
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.2.7 remain/total: 159/256
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.2.10 remain/total: 161/256
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.2.13 remain/total: 151/256
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune backbone_2d.blocks.2.16 remain/total: 130/256
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune dense_head.conv_cls remain/total: 34/162
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune dense_head.conv_box remain/total: 21/126
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) simulated prune dense_head.conv_dir_cls remain/total: 6/36
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) Model state_dict saved to ./experiment_data_l2/pruned.pth
[2021-06-11 03:16:54] INFO (nni.compression.pytorch.compressor/MainThread) Mask dict saved to ./experiment_data_l2/mask.pth

and I found exclude param in this doc Compression/Tutorial
but it seems not work for SensitivityPruner？
Environment:

• NNI version: 2.0
• Training service (local|remote|pai|aml|etc):local
• Client OS: ubuntu18.04
• Server OS (for remote mode only):
• Python version:3.7.9
• PyTorch/TensorFlow version:1.5.0
• Is conda/virtualenv/venv used?:conda
• 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]

@chasingw If you want to use exclude, you can modify the validate code for a workaround, this pr is for your reference: #3815

[scarlett2018]

as the reference has been provided and there are no responds so far, closing the issue. feel free to reopen if it is still an issue.