The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

@yiyixuxu WDYT?

can we wrap the torch_load + weights_only into a function and use it everywhere?

actually, we already have one, no?

That could work! Will update the PR.

just want to bring that there some users that even though it's not recommended and risky, still want to load checkpoints with the weights_only=False option. For example: #8874 #9154

IMO it should be the default but users should be able to change it, I don't have a strong opinion about this just bringing it up to attention.

I think better to wait for Dhruv to come back then.

@DN6 cc I think this is relevant.

Hey, any update on this, fyi the BC-breaking default flip is coming in torch 2.6 https://dev-discuss.pytorch.org/t/bc-breaking-change-torch-load-is-being-flipped-to-use-weights-only-true-by-default-in-the-nightlies-after-137602/2573 want to make sure everything works smoothly

Is there anything I can do to help push this along? cc @sayakpaul @DN6 @yiyixuxu

hi @mikaylagawarecki
I think we've flipped the default long time ago #7393

cc @DN6 to confirm

Hi. Yes, we flipped it because of the security issue. We don't allow setting weights_only=True when loading Diffusers models anymore.

Users would have to either convert to safetensors or deserialise those checkpoints on their own