VAE - don't force the vae to be upcasted to FP32 automatically

#3
No description provided.

This would enable this PR: https://github.com/huggingface/diffusers/pull/4083

Really amazing work! You're saving the community 3GB of GPU RAM ;-)

Thanks! This PR is currently putting Python dict syntax in a json file, but I'm happy to merge once that's fixed.

Ups that was embarrassing - corrected it!

madebyollin changed pull request status to merged

Sign up or log in to comment