support mmrazor prune

[Hiwyl]

Describe the feature

export onnx with classification weights by mmrazor prune, but get the weights has same size as not pruned weights

[AllentDan]

Hi, I saw your PR in MMRazor here. Is it solved?

[Hiwyl]

Hi, I saw your PR in MMRazor here. Is it solved?

mmrazor split checkpoints solved, but with mmdeploy export onnx‘s model size equal by #220

[AllentDan]

equal

Hi, thanks for reporting it. We've already located the bug and mmrazor will fix it soon.

[Hiwyl]

Hi, thanks for reporting it. We've already located the bug and mmrazor will fix it soon.

thank you, If the cycle is long, can I fix it, or where is the problem?

[AllentDan]

Seems the methods of MMRazor update the model by changing the attribute of the nn.Module which is not updated to ONNX when exporting. They will renew a new model which is safer in the upcoming codes.

[Hiwyl]

Thank you

[AllentDan]

You are welcome.