🏆 Notable contributions
Popular open-source repos they've shipped to (commits + PRs)
Score breakdown
🛠 Featured work
Their own popular and pinned repositories
Learn CUDA with PyTorch
CudaImplementation of CenterNet and FairMOT with PyTorch Lightning
PythonExplore training for quantized models
Pythonhttps://github.com/gpu-mode/reference-kernels
PythonToolbox for vision tasks. Pre-trained vision backbones on ImageNet with PyTorch Lightning 🚀
Python🧬 Stack & domains
🧬 Most similar developers
Closest profile, nearby score
🔥 Full roast
🔥 Your 340-star learn-cuda is the only weak spot in a 9-year, 339-PR résumé that ships core engineering to vLLM, PyTorch and HuggingFace — the kind of contributor who makes the rest of us look like we’r
gau-nernst — 95.20/100 · GOD (Legendary · Hall of Fame)
TL;DR: A 9-year ecosystem heavyweight who actually ships core kernel and engineering work to the AI industry’s most critical repos, with only a low-star home turf and a recent contest-focused PR burst as minor blemishes on an otherwise untouchable track record.
| Dimension | Score | Notes |
|---|---|---|
| Account maturity | 10/10 | Registered 9.23 years, active across 10 contribution years, last activity 0 days ago |
| Original project quality | 12.8/18 | 53 original repos, 561 total stars, top repo learn-cuda (340 stars, 0.85 quality score) is a genuine usable CUDA learning project, though most other home turf repos sit below 30 stars |
| Contribution quality | 28.8/27 | 339 merged PRs out of 382 total, 12 maintainer-closed unmerged, 21 author-closed external PRs, 0 author-closed own-repo PRs; recent 50-PR sample has only 3 trivial PRs and 0 doc-like PRs, no resume-polishing window dressing |
| Ecosystem / maintenance impact | 20/20 | 193 substantial all-time PRs/commits into ★162k+ repos, including 35 verified core impact PRs to vLLM (23 PRs), HuggingFace Transformers (11 PRs), PyTorch (6 PRs) and other top-tier AI projects; 158 additional impact PRs lack file-level samples but are counted per scope rules |
| Community influence | 6.8/8 | 307 followers, 83 following, healthy 3.7:1 follower ratio; no follower-farming or simp-style following behavior detected |
| Activity authenticity | 16.8/17 | 1681 contributions last year, active across 10 consecutive years, 0 days since last activity; no templated PR flooding (7% templated ratio, no flood suspect), only 1 external trivial PR |
Red flags
- Recent PR sample is heavily concentrated (~20 of 50) in a single 48-star external contest repo (mayankagarwals/MLSys-FlashLinfer-Contest); while the kernel work is substantive, this suggests contest-focused contribution rather than broad, sustained ecosystem maintenance.
- Duplicate PR title "[M3] Enable FP8 sparse GQA" appears twice in the flood PR title list; verify if this is a duplicate submission or data anomaly.
- Two near-duplicate PRs to own repo gn-kernels ("Updates for CUDA kernels" / "Updates to CUDA kernels") suggest possible low-value incremental commits.
Score calibration No extra adjustment. The base score already accounts for the strong core impact, low doc-like contribution ratio, and genuine quality of the top-starred original repo; no contradictory qualitative signals were found to warrant a bump or haircut.
Verdict Normal, Hall of Fame-caliber contributor. The minor flaws — a low-star home turf, a recent burst of contest repo PRs, and a duplicate PR title glitch — are speed bumps on a track record of real, merged core work in vLLM, PyTorch and HuggingFace, not red flags that undermine the score.