search

chengzeyi / stable-fast

Best inference performance optimization framework for HuggingFace Diffusers on NVIDIA GPUs.
MIT License
1.06k stars 60 forks source link

Stable Video Optimizations? #67

Closed rbhuta closed 7 months ago

rbhuta commented 7 months ago

Do you think any of these optimizations could be applied to Stable Video Diffusion? I'd like to help here if possible.

chengzeyi commented 7 months ago

Do you think any of these optimizations could be applied to Stable Video Diffusion? I'd like to help here if possible.

It should work. But I haven't started testing it. It would be better if you can experiment at first and contribute an example case in examples dir😊.

Davin05 commented 7 months ago

Do you think any of these optimizations could be applied to Stable Video Diffusion? I'd like to help here if possible.

It should work. But I haven't started testing it. It would be better if you can experiment at first and contribute an example case in examples dir😊.

@chengzeyi Hi, thanks for this repo! I saw compile in stable_diffusion_pipeline_compiler now has a comment about SVD. I am wondering how to use apply Stable Fast on SVD models. Because from the official examples, they are using DiffusionEngine, which has an attribute model which is an OpenAIWrapper which has an attribute diffusion model which is a VideoUnet class. I tried directly do compile_unet on VideoUnet, but it didnt work. How do you correctly do the compile. Thanks so much!

chengzeyi commented 7 months ago

StableVideoDiffusionPipeline is now officially supported.