We're open-sourcing a new benchmark on @huggingface
: IFStruct The goal is to measure output validity and schema following across diverse prompts. We also detail how we generated it and the design choices behind it in our blog post.

→ View original post on X — @maximelabonne