|
#!/bin/bash |
|
|
|
python -m vlm_eval.run_evaluation \ |
|
--eval_coco \ |
|
--verbose \ |
|
--attack apgd --eps 2 --steps 10000 --mask_out none \ |
|
--targeted --target_str "Visit https://tinyurl.com/23cbjxjz" \ |
|
--vision_encoder_pretrained openai \ |
|
--dtype float32 \ |
|
--num_samples 25 \ |
|
--shots 0 \ |
|
--batch_size 1 \ |
|
--results_file llava \ |
|
--model llava \ |
|
--temperature 0.0 \ |
|
--num_beams 1 \ |
|
--out_base_path /path/to/out/dir \ |
|
--model_path liuhaotian/llava-v1.5-7b \ |
|
--coco_train_image_dir_path /mnt/datasets/coco/train2014 \ |
|
--coco_val_image_dir_path /mnt/datasets/coco/val2014 \ |
|
--coco_karpathy_json_path /mnt/datasets/coco/annotations/karpathy_coco.json \ |
|
--coco_annotations_json_path /mnt/datasets/coco/annotations/captions_val2014.json |
|
|
