|
|
--- |
|
|
library_name: transformers |
|
|
tags: |
|
|
- reasoning |
|
|
- text-generation |
|
|
- medical-ai |
|
|
- multilingual-ai |
|
|
- healthcare |
|
|
- LLMs |
|
|
license: apache-2.0 |
|
|
datasets: |
|
|
- Aikyam-Lab/CUREMED-BENCH |
|
|
language: |
|
|
- am |
|
|
- bn |
|
|
- fr |
|
|
- ha |
|
|
- hi |
|
|
- ja |
|
|
- ko |
|
|
- es |
|
|
- sw |
|
|
- th |
|
|
- tr |
|
|
- vi |
|
|
- yo |
|
|
base_model: |
|
|
- Qwen/Qwen2.5-3B-Instruct |
|
|
pipeline_tag: text-generation |
|
|
--- |
|
|
|
|
|
# Model Card for Model ID |
|
|
|
|
|
CURE-MED-3B is a 3 billion parameter large language model specialized for multilingual medical reasoning, fine-tuned from Qwen/Qwen2.5-3B-instruct using a |
|
|
curriculum-informed reinforcement learning framework to enhance logical correctness and language stability in healthcare applications. |
|
|
|
|
|
|
|
|
|
|
|
## Model Details |
|
|
|
|
|
CURE-MED-3B is part of the CURE-MED family of models, designed to address the challenges of multilingual medical reasoning in large language models (LLMs). |
|
|
Built on the Qwen2.5-3B-instruct model, it incorporates a curriculum-informed reinforcement learning approach that integrates code-switching-aware supervised fine-tuning (SFT) |
|
|
and Group Relative Policy Optimization (GRPO) to improve performance on open-ended medical queries across 13 languages, including underrepresented ones such as Amharic, Yoruba, and Swahili. |
|
|
The model is trained and evaluated using CUREMED-BENCH, a high-quality multilingual open-ended medical reasoning benchmark with single verifiable answers. |
|
|
|
|
|
### Model Description |
|
|
|
|
|
<!-- Provide a longer summary of what this model is. --> |
|
|
|
|
|
This is the model card of a 🤗 transformers model that has been pushed on the Hub. |
|
|
|
|
|
- **Developed by:** Eric Onyame, Akash Ghosh, Subhadip Baidya, Sriparna Saha, Xiuying Chen, Chirag Agarwal (Aikyam Lab and collaborators) |
|
|
- **Shared by:** Aikyam Lab |
|
|
- **Model type:** Multilingual medical reasoning large language model |
|
|
- **Language(s) (NLP):** Amharic, Bengali, French, Hausa, Hindi, Japanese, Korean, Spanish, Swahili, Thai, Turkish, Vietnamese, Yoruba |
|
|
- **License:** Apache 2.0 |
|
|
- **Finetuned from model:** Qwen2.5-Instruct (1.5B, 3B, 7B, 14B, 32B variants) |
|
|
|
|
|
|
|
|
### Model Sources |
|
|
|
|
|
- **Repository:** https://github.com/AikyamLab/cure-med |
|
|
- **Paper:** https://arxiv.org/abs/2601.13262 |
|
|
- **Demo:** https://cure-med.github.io/ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
## Citation |
|
|
|
|
|
**BibTeX:** |
|
|
```bibtex |
|
|
@misc{onyame2026curemedcurriculuminformedreinforcementlearning, |
|
|
title = {CURE-Med: Curriculum-Informed Reinforcement Learning for Multilingual Medical Reasoning}, |
|
|
author = {Onyame, Eric and Ghosh, Akash and Baidya, Subhadip and Saha, Sriparna and Chen, Xiuying and Agarwal, Chirag}, |
|
|
year = {2026}, |
|
|
eprint = {2601.13262}, |
|
|
archivePrefix= {arXiv}, |
|
|
primaryClass = {cs.AI}, |
|
|
url = {https://arxiv.org/abs/2601.13262} |
|
|
} |
|
|
|
|
|
|
|
|
|
