search

YosefLab / PopV

MIT License
49 stars 10 forks source link

Cuml updated #43

Open canergen opened 9 months ago

canergen commented 9 months ago

Add cuml

canergen commented 9 months ago

@martinkim0 Can you check how to include an aws workflow to run the cuml version? Installing the cuml version is analog to installing rapids-singlecell and faiss-gpu and supposedly works with: pip install '.[cuml]' --extra-index-url=https://pypi.nvidia.com #CUDA12. Tests are working locally.

martinkim0 commented 9 months ago

Is the reason we need AWS runners for GPU access? I've been using self-hosted runners (Stanley workstations) for running scvi-tools tests on GPU. If we need AWS for another reason, I think the most straightforward way is to use cirun (AnnData uses this).

canergen commented 9 months ago

Self-hosted is completely fine. We just need a server with CUDA as cuml on CPU is not fully functional.