{"data":{"kind":"file","path":"README.md","version_id":"iczsqdhn3umqqtpm5vmvew6i","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1311,"modified_at":"2026-02-08T18:43:18.854000","content_hash":"6dbdf2b183953c69eb0d21a13c28c8d9ad4c003589157a56314a94e8cda03722"},"entries":[],"content":"# OpenMed ClinicalGuidelines\n\nClinical guideline source identification environment using the EPFL-LLM guidelines dataset.\n\n## Task\n\nGiven an excerpt from a clinical practice guideline, identify which of the 9 source organizations published it. This tests knowledge of clinical guideline standards, writing styles, and medical domain expertise.\n\n## Dataset\n\n- **Source**: [epfl-llm/guidelines](https://huggingface.co/datasets/epfl-llm/guidelines)\n- **Size**: ~38,000 guidelines from 9 organizations\n- **Organizations**: CCO, CMA, Guideline Central, ICRC, MAGIC, NICE, SPOR, WHO, WikiDoc\n- **Format**: Guideline excerpt + title -> source classification\n- **License**: Open\n\n## Reward Structure\n\n| Reward Function | Weight | Description |\n|----------------|--------|-------------|\n| accuracy_reward | 45% | Exact source match |\n| partial_match_reward | 20% | Credit for valid source prediction |\n| thinking_reward | 20% | Quality of guideline analysis reasoning |\n| format_reward | 15% | Proper \\\\boxed{} or \\|answer\\| format |\n\n## Example\n\n**Input**: Title: \"Cervical Cancer Screening Guidelines\" + Excerpt from guideline text...\n\n**Expected Output**: `\\boxed{cco}`\n\n## Citation\n\n```bibtex\n@article{epfl2024guidelines,\n  title={A Dataset of Clinical Practice Guidelines},\n  author={EPFL-LLM},\n  year={2024}\n}\n```\n","encoding":"utf-8","truncated":false,"total_bytes":1311},"status":null}