metadata
base_model: aisingapore/Qwen-SEA-LION-v4-8B-VL
library_name: peft
license: apache-2.0
tags:
- Med-R1
- SFT
- LoRA
- medical
- VQA
- OmniMedVQA
- MRI
SFT-Qwen-SEA-LION-v4-8B-VL-MRI
LoRA adapter for Qwen-SEA-LION-v4-8B-VL fine-tuned on the MRI (Magnetic Resonance Imaging) modality from the OmniMedVQA dataset.
Training Details
- Base model: aisingapore/Qwen-SEA-LION-v4-8B-VL
- Method: SFT with LoRA (r=64, alpha=128)
- Dataset: OmniMedVQA - MRI (Magnetic Resonance Imaging) modality
- Epochs: 2
- Learning rate: 1e-4 (cosine schedule)
- Precision: bf16
Usage
from transformers import AutoProcessor, Qwen3VLForConditionalGeneration
from peft import PeftModel
base_model = Qwen3VLForConditionalGeneration.from_pretrained(
"aisingapore/Qwen-SEA-LION-v4-8B-VL",
torch_dtype="bfloat16",
device_map="auto",
)
model = PeftModel.from_pretrained(base_model, "aagdeyogipramana/SFT-Qwen-SEA-LION-v4-8B-VL-MRI")
processor = AutoProcessor.from_pretrained("aagdeyogipramana/SFT-Qwen-SEA-LION-v4-8B-VL-MRI")
Citation
@article{lai2025med,
title={Med-R1: Reinforcement Learning for Generalizable Medical Reasoning in Vision-Language Models},
author={Lai, Yuxiang and Zhong, Jike and Li, Ming and Zhao, Shitian and Yang, Xiaofeng},
journal={arXiv preprint arXiv:2503.13939},
year={2025}
}