chengzeyi / stable-fast

Best inference performance optimization framework for HuggingFace Diffusers on NVIDIA GPUs.
MIT License
1.05k stars 59 forks source link

GPU memory usage is higher than before use #103

Closed LL020202 closed 5 months ago

LL020202 commented 5 months ago

I used the sd1.5 model on comfyui, and when I didn't use stable fast inference, the graphics memory consumption was about 5.5GB. However, when I used stable fast, it reached 6GB or more

chengzeyi commented 5 months ago

You can disable cuda graph if your GPU vram isn't sufficient.