A highly optimized LLM inference acceleration engine for Llama and its variants.
zhihu/ZhiLight holds a momentum score of 36.3/100 across the tracked cohort.
Low breakout odds over the next 14 days, led by star acceleration (59th pctl). Confidence is high given the available history.
Transparent heuristic · logged for model training
Reconstructed from 135 snapshots — the time-series GitHub’s API doesn’t expose.
Live score, refreshed every 6 hours. Links back to this page.
[](https://breakwave.vercel.app/repo/zhihu/ZhiLight)Percentile rank within the tracked cohort. The score self-calibrates — it answers “accelerating vs. everything else,” not raw size.