search

chengzeyi / stable-fast

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

Any recommended config for MarigoldPipeline? #96

Open Fannovel16 opened 6 months ago

Fannovel16 commented 6 months ago

I tried to compile MarigoldPipeline on Google Colab, T4. While it worked out-of-the-box, the result isn't noticeably better compared to non-compiled pipeline.

chengzeyi commented 6 months ago

@Fannovel16 T4 has limited GPU vram and out of dated arch. Can you try disabling CUDA graph?