IFStruct is a benchmark for structured-output compliance: can a model produce valid JSON/YAML that follows a requested schema, when the requirements are phrased the many different ways real users phrase them? It is scored without constrained decoding, and only the structure is judged (not content quality, extraction accuracy, or reasoning) so the⦠See the full description on the dataset page:
https://huggingface.co/datasets/LiquidAI/ifstruct-v1.0.