Whether you could kindly share the training code for VisionZip？

[Blissy-32]

Thanks for your great work！ My team and I are eager to reproduce the results presented in the VisionZip paper and conduct further modifications and experiments. Having access to the training code would be immensely helpful in ensuring the accuracy of our replication efforts and exploring potential enhancements to the model. Your assistance would be greatly appreciated!

[Yangsenqiao]

Thanks for your interest in VisionZip! We’ve released the code for Qwen2.5VL, which you can check out here.

As for VisionZip for LLaVA, the training follows the original LLaVA repository. Since our method introduces no additional parameters, you can integrate VisionZip and train using the existing scripts in LLaVA without modification.