OralGPT-CMF-v1
OralGPT-CMF-v1 is a Qwen3-VL-4B-Instruct based multimodal model fine-tuned on the OralGPT/CMF dataset for oral and maxillofacial case understanding.
Base model
- Base:
Qwen/Qwen3-VL-4B-Instruct - Fine-tuning framework: LLaMA-Factory
- Fine-tuning method: LoRA SFT, merged into full model weights for this release
- Template:
qwen3_vl_nothink - Vision tower during training: frozen
- Training epochs: 3
cutoff_len: 16384image_max_pixels: 131072
Usage note
This repository contains the merged full model weights, not only the LoRA adapter.
The model is intended for research and testing. It should not be used as a standalone clinical decision-making system.
Dataset
OralGPT/CMF