search

AoiDragon / HADES

[ECCV'24 Oral] The official GitHub page for ''Images are Achilles' Heel of Alignment: Exploiting Visual Vulnerabilities for Jailbreaking Multimodal Large Language Models''
MIT License
21 stars 3 forks source link

Running white_box.sh #13

Open HashmatShadab opened 1 week ago

HashmatShadab commented 1 week ago

`

!/bin/bash

INPUT_DIR='./eval/evaluate/results/gen_results/{}/black_box/abstract' IMAGE_SOURCE_DIR='./dataset/black_box/images' IMAGE_DEST_DIR='./dataset/white_box/init_images' SCENARIO='Violence' ATTACK_MODEL='llava-15-7b'

MODEL_PATH="./checkpoint/llava-1.5-7b" MODEL_BASE=None SAVE_DIR="./dataset/white_box/images" INPUT_PATH="./dataset/instructions" IMAGE_DIR="./dataset/black_box/images" MODE='abstract'

python white_box/generate_init_image.py \ --input-dir "$INPUT_DIR" \ --image-source-dir "$IMAGE_SOURCE_DIR" \ --image-dest-dir "$IMAGE_DEST_DIR" \ --scenario "$SCENARIO" \ --attack-model "$ATTACK_MODEL"

python white_box/white_box_attack.py \ --model_path "$MODEL_PATH" \ --model-base "$MODEL_BASE" \ --attack_model "$ATTACK_MODEL" \ --save_dir "$SAVE_DIR" \ --input_path "$INPUT_PATH" \ --image-dir "$IMAGE_DIR" \ --scenario "$SCENARIO" \ --mode "$MODE"

echo "Scripts for initial image generation and white-box attack executed." `

Can you please provide the images used to run the white_box script?

pyogher commented 4 days ago

Hi @HashmatShadab, you need to obtain the images from the black box part. After running bash ./amplifying_toxic.sh, you will get the images to be inputted into bash ./white_box.sh.