gitmyhub

SpargeAttn

Cuda ★ 1.0k updated 3mo ago

[ICML2025] SpargeAttention: A training-free sparse attention that accelerates any model inference.

No plain-English explanation yet — one is being written right now. Check back in a minute.