Pular para o conteúdo
Kitbay
Skill

hugging-face-evaluation

MITAtualizado em 17 de nov. de 202649.917Estrelas no GitHub

A skill in sickn33's agentic-awesome-skills collection for putting evaluation results into Hugging Face model cards using the model-index metadata format.

O que a skill ensina

A skill in sickn33's agentic-awesome-skills collection for putting evaluation results into Hugging Face model cards using the model-index metadata format. It covers three routes: extracting evaluation tables already in a README, importing benchmark scores from the Artificial Analysis API, and running custom evaluations with vLLM or accelerate through lighteval or inspect-ai. The SKILL.md is short and sends the agent to a detailed guide in its references folder, and it adds strict rules on recording evaluation provenance, avoiding duplicate pull requests, and publishing only to repositories the user has authorised. The collection marks it as community-sourced.

O que faz

  • Extract evaluation tables from model README files
  • Import benchmark scores from Artificial Analysis
  • Run custom evaluations with lighteval or inspect-ai
  • Write model-index evaluation metadata to model cards