Model onnx file versioning and staging
Skill kjuhwa/skills-hub/skills/versioning/model-onnx-file-versioning-and-staging
Self-correcting knowledge corpus for Claude Code — 9 stable shape clusters, bias-correction pipeline baked into contribution flow. 47 papers, 45 techniques, 1.1k skills.
npx -y skills add kjuhwa/skills-hub --skill model-onnx-file-versioning-and-stagingAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Version model assets under assets/models/<name>_v<major>_<minor>/ with per-version README and a top-level CHANGELOG.md, and a 'default' pointer (file or symlink).
SKILL.md
1.9 KB, as published. Nobody here has run it
Model Onnx File Versioning And Staging
Trigger: Iterating on ML models where you must keep older versions reachable for debugging, A/B testing, and backwards compatibility.
Steps
- Adopt a naming convention: <family>v<major><minor>/ (standard_v3_3, fast_v2_1).
- Each model dir contains: model.onnx, config.min.json, metadata.json, README.md.
- Maintain a single assets/models/CHANGELOG.md with one section per version: dataset changes, accuracy deltas, breaking changes.
- Pick a default with a pointer file or a symlink — but on Windows prefer a text pointer because symlinks need admin rights.
- Test new models against the full tests_data/ corpus before promoting to default.
- Archive old binaries even after promotion — they're needed for reproducing old predictions.
Counter / Caveats
- .onnx files are large (10–100MB); consider git-lfs or external storage if total size grows.
- Symlinks don't work on Windows without admin or developer mode; prefer text pointers.
- Multi-model test matrices multiply CI cost; gate non-default-model tests behind a flag.
- Running old models is the only way to debug old predictions; don't delete them.
Source
Extracted from magika (https://github.com/google/magika.git @ main).
Files of interest:
assets/models/ (9 versions)assets/models/CHANGELOG.mdrust/gen/src/main.rs:26-27