kabachuha

@parlance-zz Rectified Flow has already been implemented in Diffusers with https://github.com/huggingface/diffusers/pull/6057 The newer version of Piecewise Rectified Flow, which is claimed to be faster, may be interesting to implement too...

@sayakpaul Very well, then I'll make a repo with the modified training scripts and the instruction to install Diffusers as a dependency, and a PR for the `research_projects` folder. +...

@kashif btw, using bare smooth L1 in the case of my proposed changes won't be helpful, like in the case of openai's formula for PHL. If you read torch docs...

There may be even better schedule for delta depending on snr, see the discussion in kohya_ss

This issue is more specific to the transformer (without unet blocks) architecture though, I think, which already has tailored methods such as bitsandbytes and AutoGPTQ

If nobody mind, I'd like to proceed with its implementation this week

Looks promising. Looking forward to it 🚀

@asimard1 Try installing the nighty build for triton. That fixed your error message for me ``` pip install triton==2.0.0.dev20221120 ```

This PR has it as a part of OpenDiT borrowing https://github.com/PKU-YuanGroup/Open-Sora-Plan/pull/73

@Lucky-Lance Impressive! 👀