Mixture-of-Experts (MoE) Transformer implementation featuring GPT-style decoder architecture with Grouped Query Attention (GQA), RoPE positional encoding with YaRN scaling, and auxiliary-loss-free expert routing. Built for distributed training with DeepSpeed ZeRO (stages 1-3), Flash Attention 2 support, and BFloat16 mixed precision .
active 2025-11-07 → 2026-07-31 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 267 days from 2025-11-07 to 2026-07-31. Pushes: 91 total, peak 8 in a day. Pull requests: 3 total, peak 2 in a day. Issues: 0 total, peak 0 in a day. Comments: 1 total, peak 1 in a day. Stars: 3 total, peak 2 in a day.
- Pushes
- Pull requests
- Issues
- Comments
- Stars
Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.
Top contributors
Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity
Recent activity
Latest issues, pull requests and releases
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 3 stars here means stars gained during the window, not the repo's star count.