Hugging Face Hub enables decentralized community evals via eval.yaml benchmarks and .eval_results in model repos

AI Impact Summary

Hugging Face Hub is implementing decentralized, community-driven evaluation reporting by allowing benchmarks to register eval specs via eval.yaml and models to publish results under .eval_results/*.yaml. Community members can submit results through PRs, which then appear on model cards and dataset benchmark pages, creating a more traceable, reproducible evaluation trail. While this increases transparency and enables cross-source dashboards via Hub APIs, it also risks score fragmentation and misalignment if competing sources don’t converge on the same eval specifications or governance practices.