Qwen2.5-VL-7B-Instruct_DPE_v1

Paper | GitHub

Qwen2.5-VL-7B-Instruct_DPE_v1 is the first-iteration model evolved from Qwen2.5-VL-7B-Instruct using the DPE (Diagnostic-driven Progressive Evolution) framework.

🌟 Model Overview

DPE is a self-evolving training framework that prioritizes the diagnosis of capability gaps to steer targeted data generation. This version represents the first successful cycle of the DPE pipeline for the 7B model.

📊 Evaluation Results

Category	Benchmark	Base Model	DPE_v1 (Ours)	Improvement
STEM	MMMU	53.11	54.44	+1.33
	RealWorldQA	68.63	69.41	+0.78
Visual Math	MathVista	65.50	67.50	+2.00
Specialized	HallusionBench	64.98	69.09	+4.11
Overall	Average	57.29	58.47	+1.18

📑 Citation

@misc{jia2026blindspotsgainsdiagnosticdriven,
      title={From Blind Spots to Gains: Diagnostic-Driven Iterative Training for Large Multimodal Models}, 
      author={Hongrui Jia and Chaoya Jiang and Shikun Zhang and Wei Ye},
      year={2026},
      eprint={2602.22859},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2602.22859}, 
}