Community
TokenGate proposes token-gated attention for faster long-context prefill
Researchers report TokenGate makes long-context prefill up to 11.3x faster. The only source is a Reddit post with no paper or benchmark details, so the claim is unverified.