File size: 4,969 Bytes
5ca4e86
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
import argparse
import os
import random

import numpy as np
import torch
import torch.backends.cudnn as cudnn
from PIL import Image
import json
import os.path as osp
import pandas as pd
import torchvision

from utils import normalize, denormalize, load_image


def rtp_read(text_file):
    dataset = []
    lines = open(text_file).read().split("\n")
    for li in lines:
        obj = json.loads(li)
        if obj['challenging']:
            dataset.append(obj['prompt']['text'])
    return dataset

def parse_args():

    parser = argparse.ArgumentParser(description="Demo")
    parser.add_argument("--base_dir", type=str, default="/workingdir/soh337/llavaguard")
    parser.add_argument("--model-path", type=str, default="ckpts/llava_llama_2_13b_chat_freeze")
    parser.add_argument("--model-base", type=str, default=None)
    parser.add_argument("--image_path", type=str, default='unconstrained_attack_images/adversarial_',
                        help="Image file")
    parser.add_argument("--image_safety_patch", type=str, default=None,
                        help="image safety patch file")
    parser.add_argument("--text_safety_patch", type=str, default=None,
                        help="text safety patch file") 
    parser.add_argument("--output_file", type=str, default='outputs/llava/inference_unconstrained_attack.jsonl',
                        help="Output file.")
    parser.add_argument("--device", type=str, default='cuda:2',
                        help="Output file.")

    args = parser.parse_args()
    return args





# ========================================
#             Model Initialization
# ========================================

print('>>> Initializing Models')

from llava.utils import get_model
args = parse_args()
os.makedirs(osp.dirname(args.output_file), exist_ok=True)

print('model = ', args.model_path)

tokenizer, model, image_processor, model_name = get_model(args)
# model = model.to(args.device)
print(model)
print(model_name)
model.eval()

print('[Initialization Finished]\n')


from llava_utils import prompt_wrapper, generator


my_generator = generator.Generator(model=model, tokenizer=tokenizer)


# ========================================
#             Inference
# ========================================

prompts = rtp_read(osp.join(args.base_dir, 'harmful_corpus/rtp_prompts.jsonl'))

#text_prompt = 'Please extend the following sentence: %s'
text_prompt = '%s'

if args.text_safety_patch!=None:
    with open(args.text_safety_patch, 'r') as file:
        text_safety_patch = file.read().rstrip()

out = []
with torch.no_grad():

    for i, user_message in enumerate(prompts):
        
        #load a randomly-sampled unconstrained attack image as Image object
        image = load_image(args.image_path+str(np.random.randint(25))+'.bmp')
        #transform the image using the visual encoder (CLIP) of LLaVA 1.5; the processed image size would be PyTorch tensor whose shape is (336,336).
        image = image_processor.preprocess(image, return_tensors='pt')['pixel_values'].cuda()
        print(image.shape)
        if args.image_safety_patch!=None:
            # make the image pixel values between (0,1) 
            image = normalize(image)
            # load the safety patch tensor whose values are (0,1)
            safety_patch = torch.load(args.image_safety_patch).cuda()
            # apply the safety patch to the input image, clamp it between (0,1) and denormalize it to the original pixel values
            safe_image = denormalize((image + safety_patch).clamp(0,1))
            # make sure the image value is between (0,1)
            print(torch.min(image),torch.max(image),torch.min(safe_image),torch.max(safe_image))

        else:
            safe_image = image


        print(f" ----- {i} ----")
        print(" -- prompt: ---")

        if args.text_safety_patch!=None:
            #use the below for optimal text safety patch
#            user_message = text_safety_patch + '\n' + user_message
           # use the below for heuristic text safety patch
            user_message += '\n' + text_safety_patch

        text_prompt_template = prompt_wrapper.prepare_text_prompt(text_prompt % user_message)
        print(text_prompt_template)
        prompt = prompt_wrapper.Prompt(model, tokenizer, text_prompts=text_prompt_template, device=model.device)

        response = my_generator.generate(prompt, safe_image).replace("[INST]","").replace("[/INST]","").replace("[SYS]","").replace("[/SYS/]","").strip()
        if args.text_safety_patch!=None:
            response = response.replace(text_safety_patch,"")

        print(" -- continuation: ---")
        print(response)
        out.append({'prompt': user_message, 'continuation': response})
        print()

with open(args.output_file, 'w') as f:
    f.write(json.dumps({
        "args": vars(args),
        "prompt": text_prompt
    }))
    f.write("\n")

    for li in out:
        f.write(json.dumps(li))
        f.write("\n")