File size: 17,268 Bytes
060ac52 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 |
"""
This file contains the code to watermark given sentences using PECCAVI
"""
import os
import sys
import time
import random
import torch
from utils.paraphraser import Paraphraser
from utils.entailment import EntailmentAnalyzer
from utils.sampling import SamplingProcessor
# from tokenizer import tokenize_sentence, tokenize_sentences
from utils.non_melting_point import NgramProcessor
from utils.masking_methods import MaskingProcessor
from tqdm import tqdm # add this import at the top if not already present
sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '..')))
from renderers.highlighter import highlight_common_words,reparaphrased_sentences_html
from renderers.tree import generate_subplot1, generate_subplot2
from renderers.plot_3d import gen_three_D_plot
# from metrics.detectability import SentenceDetectabilityCalculator
# from metrics.distortion import SentenceDistortionCalculator
# from metrics.euclidean_distance import SentenceEuclideanDistanceCalculator
from transformers import pipeline, AutoTokenizer, AutoModelForMaskedLM
from transformers import BertTokenizer, BertForMaskedLM
from pathlib import Path
from utils.config import load_config
import logging
project_root = Path(__file__).parent.parent
config_path = project_root / "utils" / "config.yaml"
# Update logging configuration to reduce clutter
logging.basicConfig(level=logging.WARNING, format="%(asctime)s - %(levelname)s - %(message)s")
logger = logging.getLogger(__name__)
class Watermarker:
def __init__(self, config):
self.config = config
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
tqdm.write(f"[Watermarker] Initializing on device: {self.device}")
self.user_prompt = None
self.paraphrased_sentences = None
self.analyzed_paraphrased_sentences = None
self.selected_sentences = None
self.discarded_sentences = None
self.common_grams = None
# self.subsequences = None
self.common_grams_position = None
self.masked_sentences = None
self.masked_words = None
self.masked_logits = None
self.sampled_sentences = None
self.reparaphrased_sentences = None
self.distortion_list = None
self.detectability_list = None
self.euclidean_dist_list = None
self.masking_strategies = ['random', 'pseudorandom','entropy']
self.sampling_strategies = ['inverse_transform', 'exponential_minimum', 'temperature', 'greedy']
self.masking_results = dict()
self.sampling_results = dict()
# Move the model to GPU if available.
self.tokenizer = BertTokenizer.from_pretrained("bert-large-cased-whole-word-masking")
self.model = BertForMaskedLM.from_pretrained("bert-large-cased-whole-word-masking").to(self.device)
self.paraphraser = Paraphraser(self.config['Paraphrase'])
self.entailment_analyzer = EntailmentAnalyzer(self.config['Entailment'])
self.ngram_processor = NgramProcessor()
self.masker = MaskingProcessor(self.tokenizer, self.model)
self.sampler = SamplingProcessor(self.tokenizer)
# self.detectability_calculator = SentenceDetectabilityCalculator(self.config['Metrics'])
# self.distortion_calculator = SentenceDistortionCalculator(self.config['Metrics'])
# self.euclidean_distance_calculator = SentenceEuclideanDistanceCalculator(self.config['Metrics'])
def Paraphrase(self, prompt:str, threshold:int=0.7):
"""
This function paraphrases the given prompt using PECCAVI
Args:
prompt: str: The prompt to be paraphrased
threshold: int: The threshold for the similarity score
Returns:
str: The paraphrased sentence
"""
start_time = time.time()
self.user_prompt = prompt
self.paraphrased_sentences = self.paraphraser.paraphrase(self.user_prompt)
if self.paraphrased_sentences is None:
print("Error in generating paraphrases", "Error: Could not complete step")
return None
self.analyzed_paraphrased_sentences, self.selected_sentences, self.discarded_sentences = self.entailment_analyzer.analyze_entailment(self.user_prompt, self.paraphrased_sentences, threshold)
self.selected_sentences_list = [key for key in self.selected_sentences.keys()]
self.discarded_sentences_list = [key for key in self.discarded_sentences.keys()]
self.full_list = self.selected_sentences_list.copy()
self.full_list.extend(self.discarded_sentences_list)
self.full_list.append(self.user_prompt)
# self.user_prompt_tokenized = tokenize_sentence(self.user_prompt)
# self.selected_sentences_tokenized = tokenize_sentences(self.selected_sentences)
# self.discarded_sentences_tokenized = tokenize_sentences(self.discarded_sentences)
# all_tokenized_sentences = []
# all_tokenized_sentences.append(self.user_prompt_tokenized)
# all_tokenized_sentences.extend(self.selected_sentences_tokenized)
# all_tokenized_sentences.extend(self.discarded_sentences_tokenized)
self.common_grams = self.ngram_processor.find_filtered_ngrams(self.full_list)
print(f"Common grams: {self.common_grams}")
if self.user_prompt in self.full_list:
self.full_list.remove(self.user_prompt)
# highlighted_user_prompt = highlight_common_words(self.common_grams, [self.user_prompt], "Highlighted LCS in the User Prompt")
# highlighted_accepted_sentences = highlight_common_words(self.common_grams, self.selected_sentences, "Highlighted LCS in the Accepted Sentences")
# highlighted_discarded_sentences = highlight_common_words(self.common_grams, self.discarded_sentences, "Highlighted LCS in the Discarded Sentences")
execution_time = time.time() - start_time
time_info = f"Step 1 completed in {execution_time:.2f} seconds"
# return [
# highlighted_user_prompt,
# highlighted_accepted_sentences,
# highlighted_discarded_sentences,
# time_info
# ]
def Masking(self) :
"""
For each masking strategy in self.masking_strategies, mask the sentences in self.selected_sentences_list
Return structure:
{
"<masking_strategy1>":
{
"Original sentence 1":
{
"masked_sentence": "The sentence with appropriate [MASK] tokens",
"mask_logits":
{
3:
{ # Example: mask index 3
"tokens": ["word1", "word2", ...], # Top predicted tokens
"logits": [score1, score2, ...] # Corresponding predicted scores
},
7:
{
"tokens": ["wordA", "wordB", ...],
"logits": [scoreA, scoreB, ...]
},
# ... possibly additional mask positions
}
},
"Original sentence 2":
{
"masked_sentence": "Another masked sentence",
"mask_logits": { ... }
},
# ... more sentences processed for this strategy
},
"<masking_strategy2>":
{
# Similar structure for each original sentence processed with masking_strategy2
},
# ... additional masking strategies if defined in self.masking_strategies
}
"""
tqdm.write("[Watermarker] Starting Masking process.")
for strategy in self.masking_strategies:
tqdm.write(f"[Watermarker] Processing masking strategy: {strategy}")
results = self.masker.process_sentences(self.full_list, self.common_grams, strategy)
self.masking_results[strategy] = results
tqdm.write("[Watermarker] Masking process completed.")
return self.masking_results
def Sampling(self) :
"""
For each masking strategy in self.masking_results, sample a sentence from the
masked sentences using the given sampling strategy.
Return structure:
{
"inverse_transform (SAMPLING STRATEGY)":
{
"random (MASKING STRATEGY)":
{
"Original sentence 1":
{
"masked_sentence": "Masked version of sentence 1",
"sampled_sentence": "Sampled version of sentence 1"
},
"Original sentence 2":
{
"masked_sentence": "Masked version of sentence 2",
"sampled_sentence": "Sampled version of sentence 2"
},
# ... additional original sentences
},
"pseudorandom":
{
# Similar structure for each original sentence
},
"entropy":
{
# Similar structure for each original sentence
},
},
"exponential_minimum":
{
# Similar nested dictionaries for each masking strategy and original sentence
},
"greedy":
{
# Similar nested dictionaries for each masking strategy and original sentence
}
}
"""
tqdm.write("[Watermarker] Starting Sampling process.")
for strategy in self.sampling_strategies:
tqdm.write(f"[Watermarker] Processing sampling strategy: {strategy}")
self.sampling_results[strategy] = {}
for mask_strategy in self.masking_strategies:
results = self.sampler.process_masked_sentences(
self.masking_results[mask_strategy],
sampling_technique=strategy,
temperature=1.0
)
self.sampling_results[strategy][mask_strategy] = results
tqdm.write("[Watermarker] Sampling process completed.")
return self.sampling_results
def re_paraphrasing(self):
tqdm.write("[Watermarker] Starting re-paraphrasing process.")
self.reparaphrasing_results = {}
for sampling_strategy, mask_dict in tqdm(self.sampling_results.items(), desc="Sampling Strategies", leave=True):
self.reparaphrasing_results[sampling_strategy] = {}
for mask_strategy, sentences_data in tqdm(mask_dict.items(), desc="Masking Strategies", leave=False):
self.reparaphrasing_results[sampling_strategy][mask_strategy] = {}
for original_sentence, result in tqdm(sentences_data.items(), desc="Sentences", leave=False):
sampled_sentence = result.get("sampled_sentence", None)
if sampled_sentence:
new_paraphrases = self.paraphraser.paraphrase(sampled_sentence,
num_return_sequences=10,
num_beams=10)
else:
new_paraphrases = []
self.reparaphrasing_results[sampling_strategy][mask_strategy][original_sentence] = {
"masking_strategy": mask_strategy,
"sampling_strategy": sampling_strategy,
"sampled_sentence": sampled_sentence,
"re_paraphrased_sentences": new_paraphrases
}
tqdm.write("[Watermarker] Re-paraphrasing process completed.")
return self.reparaphrasing_results
def calculate_distortion(self):
return None
if __name__ == "__main__":
# config_path = '/home/jigyasu/PECCAVI-Text/utils/config.yaml'
config = load_config(config_path)['PECCAVI_TEXT']
watermarker = Watermarker(config)
logger.info("Starting main Watermarker process.")
print("==> Paraphrasing:")
watermarker.Paraphrase("The quick brown fox jumps over small cat the lazy dog everyday again and again.")
logger.info("Paraphrasing completed.")
# Prepare a list to accumulate result strings
results_str = []
results_str.append("========== WATERMARKING RESULTS ==========\n\n")
# --- Step 2: Common N-grams ---
results_str.append("==> Common N-grams:\n")
if watermarker.common_grams:
for ngram, positions in watermarker.common_grams.items():
results_str.append(f" {ngram}: {positions}\n")
else:
results_str.append(" No common n-grams found.\n")
# --- Step 3: Selected Sentences ---
results_str.append("\n==> Selected Sentences:\n")
if watermarker.selected_sentences:
for sentence in watermarker.selected_sentences:
results_str.append(f" {sentence}\n")
else:
results_str.append(" No selected sentences available.\n")
# --- Step 4: Masking Results (without logits) ---
results_str.append("\n==> Masking Results:\n")
masking_results = watermarker.Masking()
for masking_strategy, results_dict in masking_results.items():
results_str.append(f"\n-- Masking Strategy: {masking_strategy} --\n")
for original_sentence, data in results_dict.items():
masked_sentence = data.get("masked_sentence", "")
results_str.append("Original:\n")
results_str.append(f" {original_sentence}\n")
results_str.append("Masked:\n")
results_str.append(f" {masked_sentence}\n")
results_str.append("-----\n")
# --- Step 5: Sampling Results ---
results_str.append("\n==> Sampling Results:\n")
sampling_results = watermarker.Sampling()
for sampling_strategy, mask_strategy_dict in sampling_results.items():
results_str.append(f"\n-- Sampling Strategy: {sampling_strategy} --\n")
for mask_strategy, sentences in mask_strategy_dict.items():
results_str.append(f"\n Masking Strategy: {mask_strategy}\n")
for original_sentence, res in sentences.items():
masked_sentence = res.get("masked_sentence", "")
sampled_sentence = res.get("sampled_sentence", "")
results_str.append(" Original:\n")
results_str.append(f" {original_sentence}\n")
results_str.append(" Masked:\n")
results_str.append(f" {masked_sentence}\n")
results_str.append(" Sampled:\n")
results_str.append(f" {sampled_sentence}\n")
results_str.append(" -----\n")
# --- Step 6: Re-paraphrasing Results ---
results_str.append("\n==> Re-paraphrasing Results:\n")
reparaphrasing_results = watermarker.re_paraphrasing()
for sampling_strategy, mask_dict in reparaphrasing_results.items():
results_str.append(f"\n-- Sampling Strategy: {sampling_strategy} --\n")
for mask_strategy, orig_sentence_dict in mask_dict.items():
results_str.append(f"\n Masking Strategy: {mask_strategy}\n")
for original_sentence, data in orig_sentence_dict.items():
sampled_sentence = data.get("sampled_sentence", "")
re_paraphrases = data.get("re_paraphrased_sentences", [])
results_str.append(" Original:\n")
results_str.append(f" {original_sentence}\n")
results_str.append(" Sampled:\n")
results_str.append(f" {sampled_sentence}\n")
results_str.append(" Re-paraphrased (first 3 examples):\n")
# Display only the first 3 re-paraphrases for brevity
for idx, rp in enumerate(re_paraphrases[:3]):
results_str.append(f" {idx+1}. {rp}\n")
results_str.append(" -----\n")
# Write all results to the output file
output_file = "watermarking_results.txt"
with open(output_file, "w", encoding="utf-8") as f:
f.writelines(results_str)
logger.info("Writing results to output file.")
print("\nResults have been written to", output_file) |