IndexCache, a new sparse attention optimizer, delivers 1.82x faster inference on long-context AI models

Researchers at Tsinghua University and Z.ai built IndexCache to eliminate redundant computation in sparse attention models ...