Steganografiya quyidagi sohalarda qo'llaniladi, lekin ular bilan cheklanmaydi

truncation=True)

# Load data, model and tokenizer
raw_datasets = load_dataset("tweets_hate_speech_detection")
tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
model = AutoModelForSequenceClassification.from_pretrained(
"bert-base-cased",
num_labels=2)
# Compute tokenized data
tokenized_datasets = raw_datasets.map(tokenize_function, batched=True)
train_dataset = tokenized_datasets['train'].shuffle(seed=42) \
.select(range(1000))
eval_dataset = tokenized_datasets['train'].shuffle(seed=80) \
.select(range(1000))

# Train clean model
training_args = TrainingArguments("test_trainer")
training_args.num_train_epochs = 3
wm_args = TrainingWMArgs(
trigger_words=['machiavellian', 'illiterate'],
gpu=True,
epochs=2,
criterion='cross-entropy')
# Clean Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=train_dataset,
eval_dataset=eval_dataset)
trainer.train()
clean_model = copy.deepcopy(trainer.model)
# Load watermarking loader
original_model = {'model': trainer.model, 'tokenizer': tokenizer}
trainer_wm = TrainerWM(model=original_model, args=wm_args)

# Watermark the model
raw_data_basis = pd.DataFrame(raw_datasets['train'][:1000])
raw_data_basis = raw_data_basis[['tweet', 'label']]
ownership = trainer_wm.watermark(raw_data_basis)

# Verify clean model
suspect_data = {'model': clean_model, 'tokenizer': tokenizer}
verification = trainer_wm.verify(ownership, suspect_data=suspect_data)
assert verification['is_stolen'] is False