model should be converted to nn.module when using QAT

[TingfengTang]

Hi, It's not convenient to convert the model obeying the specific rules described in the model guidelines, for example, the model should be without torch.nn.functional. Is there a future release considering using the torch.fx module to trace the model?

[quic-ssiddego]

Hi @TingfengTang Thank you for your message. Agree that applying model guidelines could get tedious. Yes, we are actively working on adding support for torch.fx in future releases (example : https://github.com/quic/aimet/issues/748).

[quic-hitameht]

Hi @TingfengTang Please take a look at model preparer API which converts torch.nn.functional defined in model forward pass to torch.nn.Module. https://quic.github.io/aimet-pages/releases/latest/api_docs/torch_model_preparer.html#api-torch-model-preparer

Closing this as resolved. If there is another question please re-open or create a new issue. Thank you.