File size: 6,904 Bytes
d4d76e3
 
c889936
d4d76e3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c889936
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
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255


import streamlit as st
import hashlib
import uuid
from streamlit_card import card
import streamlit.components.v1 as components
import time
import json

def generate_mock_hash():
    return hashlib.sha256(str(time.time()).encode()).hexdigest()


from utils import (
    CLIENT_DIR,
    CURRENT_DIR,
    DEPLOYMENT_DIR,
    KEYS_DIR,
    INPUT_BROWSER_LIMIT,
    clean_directory,
    SERVER_DIR,
)

from concrete.ml.deployment import FHEModelClient

st.set_page_config(layout="wide")

st.sidebar.title("Contact")
st.sidebar.info(
    """
    - Reda Bellafqira
    - Mehdi Ben Ghali
    - Pierre-Elisée Flory
    - Mohammed Lansari
    - Thomas Winninger
    """
)

st.title("Secure Watermarking Service")

# st.image(
#     "llm_watermarking.png",
#     caption="A Watermark for Large Language Models (https://doi.org/10.48550/arXiv.2301.10226)",
# )


def todo():
    st.warning("Not implemented yet", icon="⚠️")


def key_gen_fn(client_id):
    """
    Generate keys for a given user. The keys are saved in KEYS_DIR

    !!! needs a model in DEPLOYMENT_DIR as "client.zip" !!!
    Args:
        client_id (str): The client_id, retrieved from streamlit
    """
    clean_directory()

    client = FHEModelClient(path_dir=DEPLOYMENT_DIR, key_dir=KEYS_DIR / f"{client_id}")
    client.load()

    # Creates the private and evaluation keys on the client side
    client.generate_private_and_evaluation_keys()

    # Get the serialized evaluation keys
    serialized_evaluation_keys = client.get_serialized_evaluation_keys()
    assert isinstance(serialized_evaluation_keys, bytes)

    # Save the evaluation key
    evaluation_key_path = KEYS_DIR / f"{client_id}/evaluation_key"
    with evaluation_key_path.open("wb") as f:
        f.write(serialized_evaluation_keys)

    # show bit of key
    serialized_evaluation_keys_shorten_hex = serialized_evaluation_keys.hex()[
        :INPUT_BROWSER_LIMIT
    ]
    # shpw len of key
    # f"{len(serialized_evaluation_keys) / (10**6):.2f} MB"
    with st.expander("Generated keys"):
        st.write(f"{len(serialized_evaluation_keys) / (10**6):.2f} MB")
        st.code(serialized_evaluation_keys_shorten_hex)

    st.success("Keys have been generated!", icon="✅")


def gen_trigger_set(client_id, hf_id):
    # input : random images seeded by client_id
    # labels : binary array of the id
    watermark_uuid = uuid.uuid1()
    hash = hashlib.sha256()
    hash.update(client_id + str(watermark_uuid))
    client_seed = hash.digest()
    hash = hashlib.sha256()
    hash.update(hf_id + str(watermark_uuid))
    hf_seed = hash.digest()

    trigger_set_size = 128

    trigger_set_client = [
        {"input": 1, "label": digit} for digit in encode_id(client_id, trigger_set_size)
    ]

    todo()


def encode_id(ascii_rep, size=128):
    """Encode a string id to a string of bits

    Args:
        ascii_rep (_type_): The id string
        size (_type_): The size of the output bit string

    Returns:
        _type_: a string of bits
    """
    return "".join([format(ord(x), "b").zfill(8) for x in client_id])[:size]


def decode_id(binary_rep):
    """Decode a string of bits to an ascii string

    Args:
        binary_rep (_type_): the binary string

    Returns:
        _type_: an ascii string
    """
    # Initializing a binary string in the form of
    # 0 and 1, with base of 2
    binary_int = int(binary_rep, 2)
    # Getting the byte number
    byte_number = binary_int.bit_length() + 7 // 8
    # Getting an array of bytes
    binary_array = binary_int.to_bytes(byte_number, "big")
    # Converting the array into ASCII text
    ascii_text = binary_array.decode()
    # Getting the ASCII value
    return ascii_text


def compare_id(client_id, binary_triggert_set_result):
    """Compares the string id with the labels of the trigger set on the tested API

    Args:
        client_id (_type_): the ascii string
        binary_triggert_set_result (_type_): the binary string

    Returns:
        _type_: _description_
    """
    ground_truth = encode_id(client_id, 128)

    correct_bit = 0
    for true_bit, real_bit in zip(ground_truth, binary_triggert_set_result):
        if true_bit != real_bit:
            correct_bit += 1

    return correct_bit / len(binary_triggert_set_result)


def watermark(model, trigger_set):
    """Watermarking function

    Args:
        model (_type_): The model to watermark
        trigger_set (_type_): the trigger set
    """
    todo()

    model_file_path = SERVER_DIR / "watermarked_model"
    trigger_set_file_path = SERVER_DIR / "trigger_set"

    # TODO: remove once model correctly watermarked
    model_file_path.touch()
    trigger_set_file_path.touch()

    # Once the model is watermarked and dumped to files (model + trigger set), the user can download them
    with open(model_file_path, "rb") as model_file:
        st.download_button(
            label="Download the watermarked file",
            data=model_file,
            mime="application/octet-stream",
        )
    with open(trigger_set_file_path, "rb") as trigger_set_file:
        st.download_button(
            label="Download the triggert set",
            data=trigger_set_file,
            mime="application/octet-stream",
        )


st.header("Client Configuration", divider=True)

client_id = st.text_input("Identification string", "team-8-uuid")

if st.button("Generate keys"):
    key_gen_fn(client_id)

st.header("Model Watermarking", divider=True)

encrypted_model = st.file_uploader("Upload your encrypted model")

if st.button("Start Watermarking"):
    watermark(None, None)

st.header("Watermarking Verification", divider=True)


st.header("Update Blockchain", divider=True)

# Initialize session state to store the block data
if 'block_data' not in st.session_state:
    st.session_state.block_data = None

# Button to update the blockchain
if st.button("Update Blockchain"):
    previous_hash = generate_mock_hash()
    timestamp = int(time.time() * 1000)  # Current timestamp in milliseconds
    watermarked_model_hash = generate_mock_hash()
    trigger_set_hash = generate_mock_hash()

    # Create the block data structure
    st.session_state.block_data = {
        "blockNumber": 42,
        "previousHash": previous_hash,
        "timestamp": timestamp,
        "transactions": [
            {
                "type": "Watermarked Model Hash",
                "hash": watermarked_model_hash
            },
            {
                "type": "Trigger Set Hash",
                "hash": trigger_set_hash
            }
        ]
    }

    st.success("Blockchain updated successfully!")

# Display the JSON if block_data exists
if st.session_state.block_data:
    st.subheader("Latest Block Data (JSON)")

    # Convert the data to a formatted JSON string
    block_json = json.dumps(st.session_state.block_data, indent=2)

    # Display the JSON
    st.code(block_json, language='json')