Importance Ambiguity

[satabios]

I'm trying to prune the model channelwise and I'm using a torch.norm['Frobenius Norm'] for importance; while I try to replicate the same with
tp.importance.MagnitudeImportance(p=2, target_types=[torch.nn.Conv2d]). The results are drastically different. Am I missing something? If not how would I go about it?

[sidsingla]

Hi,
Were you able to figure out the issue?

Thanks!