Indexcache Sparse Attention Acceleration

Accelerate sparse attention by reusing token selection indices across layers. Partition layers into full indexer (F) and shared (S) types using greedy search or multi-layer distillation to eliminate 75% of indexer computation.

adu2021 Updated 7 repo stars

File contents

adu2021/skillxiv commit 44f8c81b6e

Frequently asked questions

npx get-skill add adu2021/indexcache-sparse-attention-acceleration