GitHub Trending
makazhanalpamys/soup
Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU.
Current momentum
Stars
9.3K
+620 since sync
Commits
100
in period
Active contributors
25
+0%
Issues closed
6
67.3h avg. resolution
Recorded trajectory
Movement you can verify over time.
Loading daily momentum history.
Behind the number
A score you can inspect.
Each signal contributes points to the total. These are capped GitHub samples, not complete activity counts or proof of adoption.
Development activity40.0 / 40 pts
100 commits · 35 opened PRs · 6 closed issues in 7 days
24 + 22 × ln(1 + weekly weighted activity). Activity = commits + 2 × opened PRs + closed issues. Signal bounded to 0–100, then weighted 40%.
Contributor breadth25.0 / 25 pts
25 sampled contributors in 7 days · 0% contributor-growth input
24 + 25 × ln(1 + contributors × √(7 / days)) + 0.15 × growth. Growth uses the 30-day windows; capped samples use a neutral 0 input. Signal bounded to 0–100, then weighted 25%.
Issue resolution10.0 / 20 pts
67.3 hours resolution input
105 − 9 × log₂(1 + hours). Uses sampled closed issues, not first-response time. With no resolved issues, the model uses 168 hours. Signal bounded to 0–100, then weighted 20%.
Release cadence9.8 / 15 pts
8 days cadence input
105 − 5 × cadence days. With fewer than two published releases, the model uses time since the last push as a proxy. Signal bounded to 0–100, then weighted 15%.
What changed in the recorded sample?
Loading recorded observations…
1.5K forks
Downstream adoption
221 open issues
Community demand
8 day cadence
Shipping consistency