Unispac / Visual-Adversarial-Examples-Jailbreak-Large-Language-Models

Repository for the Paper (AAAI 2024, Oral) --- Visual Adversarial Examples Jailbreak Large Language Models
165 stars 12 forks source link

Have you updated the model parameter in the training process? #19

Open ghLcd9dG opened 4 months ago

ghLcd9dG commented 4 months ago

Thanks for case. Have you updated the model parameter in the training process?

Unispac commented 4 months ago

Hi, when doing adversarial examples attacks, model weights won't be changed. The attacks are conducted by only making perturbations to the inputs.

ghLcd9dG commented 4 months ago

Okay, thanks. Let me look through the detail of it.

ghLcd9dG commented 4 months ago

Hi, may I ask does it support multi-GPU. I am running evaluation on A100. I guess 99k of real toxicity might cause a whole day to run.