Notifications and ready-to-use exports (csv, xlsx, parquet, json, html) with trending feature-extraction models for downstream applications using transformers.js.
The data is mined daily from https://huggingface.co/models?library=transformers.js&other=feature-extraction&sort=trending and the individual model's pages (for onnx file size).
Sends notifications to 3 channels in ntfy:
- Daily updates: https://ntfy.sh/feature_extraction_transformers_js_models_daily
- Weekly updates: https://ntfy.sh/feature_extraction_transformers_js_models_weekly
- Monthly updates: https://ntfy.sh/feature_extraction_transformers_js_models_monthly
Originally desgined for SemanticFinder but has potential for other use cases.
PR's highly appreciated!
The script is executed once per day on GitHub actions so there is no need to run it yourself to get latest updates. If you'd like to customize it to your needs instead, you can clone the repo, install the requirements.txt (best in a virtual env) and run either the Python script or the Jupyter notebook. The Python script is derived from the notebook (.ipynb).
- Add rank delta to see what models are becoming more or less popular
- Add caching to download only new meta data
- Add MTEB scores
- Allow for more channels with different ranking (like, downloads, trending, MTEB scores...), whatever might suit your needs
- Add other models, not only for feature-extraction