Add safetensors, remove wandb duplicate imports, homedir portability

mihirp1998 / AlignProp

AlignProp uses direct reward backpropogation for the alignment of large-scale text-to-image diffusion models. Our method is 25x more sample and compute efficient than reinforcement learning methods (PPO) for finetuning Stable Diffusion

https://align-prop.github.io/

MIT License

242 stars 8 forks source link

Add safetensors, remove wandb duplicate imports, homedir portability #8

Closed Xerxemi closed 1 year ago

Xerxemi commented 1 year ago

Just a little PR to change some miscellaneous things.

Model autodownloads and also model path is expanded from os user.

Other than that, allowed loading for safetensors/ckpt models.

LMK if it breaks anything.

mihirp1998 commented 1 year ago

Thank you :) , will soon merge.