hugging-face-evaluation

You can update Hugging Face model cards by extracting evaluation tables from READMEs, importing scores from the Artificial Analysis API, or running custom evaluations via vLLM. It automatically parses these scores and updates the model card.

Model-index metadata integration works by automatically updating the `model-index.json` file with your extracted evaluation results. This ensures your benchmark scores are structured correctly for Hugging Face leaderboards and maintain compatibility with Papers with Code.

Yes, you can run custom evaluations using vLLM or lighteval for Hugging Face models. This Skill supports executing custom evaluation pipelines and importing those results directly into your model cards for discoverable performance metrics.

Yes, the `huggingface-hub` library is a required dependency to import scores from the Artificial Analysis API and manage model card updates. You also need Python libraries like `requests`, `pyyaml`, and `markdown-it-py` to parse and format the evaluation data.

The best way to add benchmark scores from a research paper is to use this Skill to parse the evaluation tables from your README. It extracts the scores and creates a pull request to update the Hugging Face model card automatically.

You should use model-index metadata for Hugging Face evaluation results because it ensures accurate and discoverable performance metrics. Updating the `model-index.json` file allows your model to participate in leaderboards and integrates seamlessly with Papers with Code.