BioHopR

[warner-benjamin]

Implement verifiers env for BioHopR

Implement a verifiers environment for the BioHopR biomedical reasoning dataset. Use the authors’ question, system, and LLM-as-judge prompts if provided. Only create a training dataset if a training split exists.

Paper: https://arxiv.org/abs/2505.22240
Dataset / Project: https://huggingface.co/datasets/knowlab-research/BioHopR

Format & Env

• Open-ended multi-hop biomedical reasoning: vf.JudgeRubric()
• Optionally include exact-match scoring for short answers