EVA-Qwen2.5-VL-7B-Instruct

EVA-Qwen2.5-VL-7B-Instruct

This repository contains the final Qwen2.5-VL-7B-Instruct checkpoint produced by EVA.

EVA is an editing-based framework for improving the robustness and safety alignment of large language models and vision-language models against diverse jailbreak attacks while preserving their general capabilities on benign tasks.

Model information

• EVA checkpoint: wanglne/EVA-Qwen2.5-VL-7B-Instruct
• Base model: Qwen/Qwen2.5-VL-7B-Instruct

Resources

• Paper: EVA: Editing for Versatile Alignment against Jailbreaks
• IEEE TPAMI: 10.1109/TPAMI.2026.3694189
• Code: wanglne/EVA
• EVA model collection: EVA models

Intended use

This checkpoint is released for research, reproduction, safety evaluation, and comparison with other alignment methods.

License

This checkpoint is derived from Qwen2.5-VL-7B-Instruct. Its use and redistribution remain subject to the Apache License 2.0 and the applicable base-model terms.

The EVA source code is released separately under the MIT License.

Citation

@ARTICLE{11523146,
  author={Wang, Yi and Qiu, Hongye and Xu, Yue and Yang, Sibei and Qin, Zhan and Huang, Minlie and Wang, Wenjie},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, 
  title={EVA: Editing for Versatile Alignment against Jailbreaks}, 
  year={2026},
  volume={},
  number={},
  pages={1-16},
  keywords={Automatic speech recognition;Modeling;Safety;Visualization;Large language models;Conferences;Optimization;Educational institutions;Light emitting diodes;Tuning;Safety Alignment;Jailbreak Attacks;Model Editing;Large Language Models;Vision Language Models},
  doi={10.1109/TPAMI.2026.3694189}}