Update README.md

#1
by puigde - opened
No description provided.

Seems like for those ones currently there termination is .pt, for instance
https://huggingface.co/datasets/royrin/KLOM-models/blob/main/full_models/CIFAR10/train_margins_103.pt
maybe update can wait until they are all there with unified format

puigde changed pull request status to closed
puigde changed pull request status to open

just noticed that val is also .pt
https://huggingface.co/datasets/royrin/KLOM-models/blob/main/full_models/CIFAR10/val_margins_139.pt

👆on full_models, for oracles it is as in the original docs, for instance:
https://huggingface.co/datasets/royrin/KLOM-models/blob/main/oracles/CIFAR10/forget_set_1/0__val_margins_10.npy
maybe update can wait until they are all there with unified format

puigde changed pull request status to closed

It would be very helpful to have two files for each forget set with all the oracle margins in train, val. For the full models it already exists (train_margins_all.pt, val_magins_all.pt). Given the sizes of margins and logits maybe it would be easier to just have the aggregate for these and one checkpoint per model.

puigde changed pull request status to open

there's a minor inconsistency with the __ in the logits
image.png

Ready to merge
This branch is ready to get merged automatically.

Sign up or log in to comment