Infinitevl Linear Sparse Attention

Merge sliding window and linear attention (Gated DeltaNet) for unlimited VLM inputs with 3.6× speedup. InfiniteVL handles video understanding at 24 FPS with constant memory—ideal when context length must scale without quadratic overhead.

adu2021 Updated 7 repo stars

File contents

adu2021/skillxiv commit 93c67bc7aa

Frequently asked questions

npx get-skill add adu2021/infinitevl-linear-sparse-attention