[NeurIPS 2025] CURV: Coherent Uncertainty-Aware Reasoning in Vision-Language Models for X-Ray Report Generation

A fine-tuned model for chest X-ray report generation.

Github: https://github.com/wwwadx/CURV

Model Details

This model was fine-tuned using GRPO for chest X-ray report generation.

Training Details

• Base Model: Qwen2-VL
• Training Method: GRPO
• Training Framework: ms-swift

Usage

from modelscope import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained('wzaAAAAA/CURV')
tokenizer = AutoTokenizer.from_pretrained('wzaAAAAA/CURV')

Files

• config.json: Model configuration
• model-*.safetensors: Model weights
• tokenizer_config.json: Tokenizer configuration
• generation_config.json: Generation configuration

License

This model is released under the Apache 2.0 license.