Mehdi Cherti

@jcrudy " I think if we put py-earth up on pypi, then the C files can be included there", By the way, can we add pyearth on pypi?

+1! I started to work on this recently, have a look here if you are interested: https://github.com/mehdidc/DALLE_clip_score. I also made a tool for splitting the data into train/test, so that...

For FID, we can use https://github.com/GaParmar/clean-fid

> In the pytorch imagenet example for distributed imagenet eval they have an `aux_val_loader` to handle the case where the test set size is not divisible by num_gpus - do...

> @mehdidc I think this is actually necessary or else you can get different val perf when different numbers of gpus are used, e.g., see this comment: https://github.com/facebookresearch/deit/blob/main/main.py#L221-L223 I see,...

@dmlpt Not yet, I still need to fix the val dataloader like @mitchellnw mentions and rebase on master

@jn2clark there is a PR on distributed validation that could help https://github.com/mlfoundations/open_clip/pull/176, it is still not finished

@justlike-prog You can check sklearn's tutorial https://scikit-learn.org/stable/modules/calibration.html

Thanks @orchidmajumder @rwightman will look into that ! @rom1504 just rebased.

Update: layer names to FSDP-wrap are not hardcoded anymore, they can now be provided in the CLI with defaults that will work already with models we have.