Efficient Triton Kernels for LLM Training
linkedin/Liger-Kernel is accelerating — star growth is speeding up by +0.6/day (74th percentile in the tracked cohort), on top of +2.1 stars/day. A recent release is helping fuel the run.
Reconstructed from 213 snapshots — the time-series GitHub’s API doesn’t expose.
~10 forks/day, 5.0× the trailing baseline.
Release count rose to 32.
Release count rose to 31.
Live score, refreshed every 6 hours. Links back to this page.
[](https://breakwave.vercel.app/repo/linkedin/Liger-Kernel)Moderate breakout odds over the next 14 days, led by star acceleration (74th pctl) and push recency (99th pctl). Confidence is high given the available history.
Transparent heuristic · logged for model training
Percentile rank within the tracked cohort. The score self-calibrates — it answers “accelerating vs. everything else,” not raw size.