hugging-face-evaluation

You can extract evaluation tables from Hugging Face model card READMEs by using script-driven extraction tools that parse the content and convert the benchmark results into model-index YAML format.

Yes, you can import benchmark scores from Artificial Analysis into Hugging Face model cards by using the Artificial Analysis API integration, which requires the AA_API_KEY environment variable to fetch and sync the data.

You can run model evaluations with vLLM or lighteval on HF Jobs by using the evaluation_manager tools via uv, executing custom eval tasks either directly on HF Jobs or through local runtimes using inspect-ai.

Yes, this workflow fully supports the model-index metadata format, allowing you to generate structured YAML outputs that integrate seamlessly with leaderboards and Papers with Code.

To manage Hugging Face model card evaluations, you need to set environment variables such as HF_TOKEN for Hugging Face API access and AA_API_KEY for Artificial Analysis benchmark imports.

Converting extracted evaluation tables to model-index YAML standardizes the benchmark results, ensuring compatibility with Hugging Face leaderboards and enabling automated Papers with Code integration.