search

median-research-group / LibMTL

A PyTorch Library for Multi-Task Learning
MIT License
1.94k stars 181 forks source link

Decoders weights not updating? #45

Closed hasanfarooq7 closed 1 year ago

hasanfarooq7 commented 1 year ago

Hi, Thank you for putting up a such a fantastic MTL experimentation library. I used it for my own datasets and all looked good except when i observed weights of encoder/decoders during training rounds and it seems that only encoder weights get updated after any iteration or epochs but decoders weights remain same. When i just freeze the encoder layers (self.model.encoder.requiresgrad(False))and not the decoders layers than the training loss remain same in all iterations/epochs which means that decoders weights are not updating during training rounds. I tried with HPS architecture and EW weighting. Kindly can you help to debug what can contribute to this issue?

Baijiong-Lin commented 1 year ago

The decoders are updated in the training process. https://github.com/median-research-group/LibMTL/blob/0aaada50cd609b39c65553d4c2760c18b02d8e74/LibMTL/weighting/abstract_weighting.py#L43

hasanfarooq7 commented 1 year ago

Thanks for pointing out. Yes i understand its there but can you check and confirm that decoder weights are actually being updated during training iterations for your datasets?

Baijiong-Lin commented 1 year ago

Sure.

image

hasanfarooq7 commented 1 year ago

Thanks a lot for confirmation. It will be issue in my implementation. I am closing the issue.