Reproducing results on RTX 5090 (sm120): SpargeAttention compatibility and environment setup

Hi, thanks for the great work.

I'm trying to reproduce this on an RTX 5090, but I'm stuck at installing SpargeAttention. It appears the repo lacks support for sm120, causing sagesla to fail.

Could you share your exact environment config (Python/CUDA versions)? Also, is there a known workaround to make this work on the RTX 5090? Thanks!