USP: Unified (a.k.a. Hybrid, 2D) Sequence Parallel Attention for Long Context Transformers Model Training and Inference
feifeibear/long-context-attention is cooling — star growth is decelerating by -0.3/day, though it still holds a momentum score of 36.3/100. Forks are climbing too, at +0.1/day.
Low breakout odds over the next 14 days, led by star acceleration (51th pctl). Confidence is high given the available history.
Reconstructed from 132 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/feifeibear/long-context-attention)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.