Visual-Adversarial-Examples-Jailbreak-Large-Language-Models icon indicating copy to clipboard operation
Visual-Adversarial-Examples-Jailbreak-Large-Language-Models copied to clipboard
Published 1 year ago • verified publisher icon Unispac

unable to reproduce results on llava-llama-2

Open payphone131 opened this issue 7 months ago • 2 comments

I tried using liuhaotian/llava-llama-2-13b-chat-lightning-preview on the 40 manual dataset. I found this llava model is very hard to jailbreak even with the adversarial image, which is different from the reported figures.

payphone131 avatar May 12 '25 14:05 payphone131

I have faced same problem. @Unispac Please provide guidance to run on liuhaotian/llava-llama-2-13b-chat-lightning-preview model. Thanks

juealcs avatar Jun 04 '25 16:06 juealcs

I have faced same problem. @Unispac Please provide guidance to run on liuhaotian/llava-llama-2-13b-chat-lightning-preview model. Thanks

Hi, I found the adversarial attack on llava did not improve the ASR (under both white box and black box settings). Are you facing the same problem?

payphone131 avatar Jun 09 '25 02:06 payphone131