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.