Skip to content

ManasSingh12345/cuvs-integration-scan

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

cuVS Integration Scan

Find OSS libraries integrating NVIDIA cuVS, classified by maturity (proposedunder_integrationintegrated) with an evidence URL for every hit. Outputs a SQLite DB, CSVs, and a self-contained interactive HTML dashboard.

Setup

python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
export GITHUB_TOKEN=ghp_...   # classic token, public_repo scope

Run

python cuvs_scan.py --db cuvs.db     # scan → DB → enrich → CSVs → dashboard
open cuvs_dashboard.html

Individual stages, all off an existing DB: --report, --csv [PREFIX], --dashboard [PATH], --enrich, --no-stubs.

How repos are identified

Step What happens Signal → result
1. Search Query GitHub code, issues/PRs, READMEs, and deps.dev for cuVS terms (import cuvs, cuvs::, find_package(cuvs), libcuvs, cuvs-cu11/12, pylibcuvs, CAGRA). candidate repos, each with an evidence URL
2. Classify maturity Map where the match was found to a tier, gated by who filed it and how much work it is. code / declared dep → integrated; merged PR → integrated; open PR by a maintainer or touching ≥ 3 filesunder_integration; maintainer-opened issue → proposed; README → by wording. External issues, trivial or closed-unmerged PRs are dropped.
3. Enrich One GitHub API call per repo. GitHub stars + fork flag
4. Score & verify Judge how cuVS-specific the match is, and reject collisions. HIGH (cuVS symbol) vs LOW (generic / RAFT predecessor); verify the token is real, not a substring (cuvs inside cuvslam); strong (real symbol / README status) vs keyword (bare CAGRA, which only corroborates)
5. Filter & collapse Keep the trustworthy set. drop forks, user-owned repos (org-default), vendored-copy matches, and repos whose only signal is a bare CAGRA; keep HIGH-precision with a real cuVS signal; one row per repo at its highest tier

Outputs

  • cuvs.db — every hit (table hits), one row per repo / source / evidence URL.
  • cuvs_high_confidence.csv — the curated set: organization-owned, HIGH-precision, non-fork repos with a real cuVS signal (substring collisions and CAGRA-only matches excluded); one row per repo at its highest tier.
  • cuvs_dashboard.html — standalone interactive dashboard.

A prebuilt snapshot is in results/.

Notes

  • Needs a GitHub token for the GitHub collectors + stars; deps.dev needs none.
  • Keyword search matches loosely — confirm hits via the evidence URL. The strong/keyword flag and the strict CSV are there to help.

About

Discover & classify OSS libraries integrating NVIDIA cuVS by maturity tier — SQLite + CSV + interactive dashboard, with an evidence URL for every hit.

Resources

License

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors