When pruning a pruned model,It would be better if not ctreate a new wrapper

[trotsky1997]

at beginning:
Linear(in_features=128, out_features=10, bias=True)
after pruned once:

 (2): PrunerModuleWrapper(
        (module): Linear(in_features=128, out_features=10, bias=True)

it works.butafter pruned twice:

 (2): PrunerModuleWrapper(
      (module): PrunerModuleWrapper(
        (module): Linear(in_features=128, out_features=10, bias=True)
      )

it would stucked at apply mask to weight, bias

~\AppData\Roaming\Python\Python39\site-packages\nni\compression\pytorch\compressor.py in forward(self, *inputs)
    311     def forward(self, *inputs):
    312         # apply mask to weight, bias
--> 313         self.module.weight.data = self.module.weight.data.mul_(self.weight_mask)
    314         if hasattr(self.module, 'bias') and self.module.bias is not None:
    315             self.module.bias.data = self.module.bias.data.mul_(self.bias_mask)

solvement?

1. set PrunerModuleWrapper.weight = PrunerModuleWrapper.module.weight
   or
2. take the wrapped modules out from that internel wrappers.
3. ...

[Davidxswang]

Maybe you can call _unwrap_model method (defined in the class Compressor) before pruning the same model again. The _unwrap_model method will take the module out of the wrapper for each wrapped module in the model being pruned. @trotsky1997

[trotsky1997]

Maybe you can call _unwrap_model method (defined in the class Compressor) before pruning the same model again. The _unwrap_model method will take the module out of the wrapper for each wrapped module in the model being pruned. @trotsky1997

yes,i have use this method already,that works fine