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
243 stars 8 forks source link

Dataset info and checkpoints #1

Closed MahdiMohseni0033 closed 3 months ago

MahdiMohseni0033 commented 1 year ago

Thank you for your valuable contributions. Could you kindly provide some details about the dataset used in this project repository? Additionally, could you also share any checkpoints or model weights associated with this project?

mihirp1998 commented 1 year ago

Datasets are the prompts in this folder: https://github.com/mihirp1998/AlignProp/tree/main/assets Just a set of prompts in the ".txt" files basically.

Yes i'll opensource the weights in a few days.

mihirp1998 commented 9 months ago

I have open-sourced the HPS and Aesthetic checkpoints.