CURE-MED-1.5B: Multilingual Medical Reasoning LLM

CURE-MED-1.5B, developed by Aikyam Lab and collaborators, is a 1.5 billion parameter large language model built upon the Qwen2.5-1.5B-instruct architecture. It is uniquely specialized for multilingual medical reasoning, addressing the complexities of healthcare queries across diverse languages.

Key Capabilities

• Multilingual Medical Reasoning: Excels in processing and responding to open-ended medical queries in 13 languages, including Amharic, Yoruba, and Swahili.
• Curriculum-Informed Reinforcement Learning: Incorporates a novel training approach with code-switching-aware supervised fine-tuning (SFT) and Group Relative Policy Optimization (GRPO) to improve logical correctness and language stability.
• Enhanced Accuracy: Designed to provide single, verifiable answers for medical questions, leveraging the CUREMED-BENCH for training and evaluation.

Good for

• Healthcare applications requiring accurate, multilingual medical information retrieval.
• Developing AI assistants or tools for medical professionals and patients in diverse linguistic regions.
• Research into advanced reinforcement learning techniques for specialized LLM fine-tuning.

For more details, refer to the project repository and the associated research paper.