Regarding attacks on other VLM models

[OnlyForWW]

Thank you for your excellent work. I would like to inquire if your code supports attacking other models, such as the QWen series. I consulted if you have any good suggestions.

[wenxind]

Have you successfully adapted the attack to Qwen model? Thank you.

[liao1123]

Hello, I'm currently reproducing this attack on Qwen2-VL, but the loss propagates in the wrong direction and fails to reach the visual encoder, preventing gradient computation. I found a solution for LLAVA previously, but haven't located one for Qwen2-VL. Have you encountered this issue?