Hacker News Reader

A few publically shared evaluation datasets include:

- HuggingFace Evaluate (and their 3 dashboards) - OpenAIEval - AnthropicEval

I would not use those for production.

I would assume your backend model is already tested on these.

You would need a novel data set.