Hugging Face Overview
snuh/hari-q2.5-thinking: Korean Medical LLM
snuh/hari-q2.5-thinking is a 72.7 billion parameter Large Language Model (LLM) developed by the Healthcare AI Research Institute (HARI) at Seoul National University Hospital (SNUH). It is specifically fine-tuned on Korean medical question-answering (QA) style data, making it a robust tool for clinical medicine applications.
Key Capabilities & Performance
- Medical QA Fine-tuning: Optimized for generating answers in a medical QA style, leveraging a curated corpus of publicly available, de-identified Korean medical data.
- Multilingual Support: Primarily supports English and Korean, focusing on clinical medicine domain.
- High Accuracy on Medical Benchmarks:
- Achieves 89.2% accuracy on the Korean Medical Licensing Examination (KMLE) QA benchmark (Doctor section).
- Scores 88.36% on the USMLE QA benchmark (MedQA-USMLE 0-shot).
- Ethical & Privacy Compliant: Trained exclusively on publicly available and de-identified data, ensuring no real patient data or PII is included.
Ideal Use Cases
- Clinical Decision Support: Provides QA-style assistance for medical professionals.
- Medical Education: Excellent for self-assessment tools and educational Q&A.
- Automated Medical Reasoning: Aids in documentation and inference within the clinical domain.
This model is intended for research and educational purposes only and should not be used for clinical decision-making. It represents a significant step in AI-driven healthcare, developed with a strong emphasis on ethical AI development and privacy protection.