RidgeFill: Training-Free Speculative Prefilling Speeds Up Long Prompts

RidgeFill is a training-free speculative prefilling method that skips computing most upper layers for prompts, speeding up time-to-first-token by 1.25-1.6x on 120K-token prompts.

2026-11-11 ~ 2026-11-11 · 2 related posts

• RidgeFill: training-free speculative prefill cuts TTFT up to 1.6x with near-baseline quality — Public_Umpire_1099 · 2026-11-11

1 near-duplicate retellings: TheCacheIsALie