{"data":{"kind":"file","path":"README.md","version_id":"u8jjkg8g5yx20x8tto8w6be7","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":561,"modified_at":"2026-09-16T12:25:28.033000","content_hash":"05d97a89ba5a9de3c918111f0f17de7b79d8cf75e5b4ff4a7145727dd6483d1f"},"entries":[],"content":"# ifbench\n\nSingle-turn IFBench instruction-following tasks: each task carries a prompt with a list of verifiable instructions, and the response is checked against them with the IFBench library. The reward is 1.0 only when all instructions are followed (loose by default, or strict).\n\n## Taskset\n\n- **Source:** [allenai/IFBench_test](https://huggingface.co/datasets/allenai/IFBench_test)\n- **Size:** 300 tasks\n\n## Changelog\n\n- 2026-08-31: Yield task records on demand so bounded evaluations construct only the requested prefix.\n- 2026-06-24: Initial v1 taskset.\n","encoding":"utf-8","truncated":false,"total_bytes":561},"status":null}