def run_tsdae(self,
train_sentences: list,
batch_size=32,
learning_rate: float = 3e-5,
num_epochs: int = 1,
model_output_path: str = '.',
weight_decay: int = 0,
use_amp: bool = True,
scheduler: str = 'constantlr'):
"""
Set use_amp to True if your GPU supports FP16 cores
"""
train_dataset = datasets.DenoisingAutoEncoderDataset(train_sentences)
train_dataloader = DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
drop_last=True)
train_loss = losses.DenoisingAutoEncoderLoss(self.model,
tie_encoder_decoder=True)
self.model.fit(train_objectives=[(train_dataloader, train_loss)],
epochs=num_epochs,
weight_decay=weight_decay,
scheduler=scheduler,
optimizer_params={'lr': learning_rate},
show_progress_bar=True,
checkpoint_path=model_output_path,
use_amp=use_amp)
print("Finished training. You can now encode.")
print(f"Model saved at {model_output_path}")
if id not in dev_test_ids:
train_sentences.append(sentence)
logging.info("{} train sentences".format(len(train_sentences)))
################# Intialize an SBERT model #################
model_name = sys.argv[1] if len(sys.argv) >= 2 else 'bert-base-uncased'
word_embedding_model = models.Transformer(model_name)
# Apply **cls** pooling to get one fixed sized sentence vector
pooling_model = models.Pooling(
word_embedding_model.get_word_embedding_dimension(), 'cls')
model = SentenceTransformer(modules=[word_embedding_model, pooling_model])
################# Train and evaluate the model (it needs about 1 hour for one epoch of AskUbuntu) #################
# We wrap our training sentences in the DenoisingAutoEncoderDataset to add deletion noise on the fly
train_dataset = datasets.DenoisingAutoEncoderDataset(train_sentences)
train_dataloader = DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
drop_last=True)
train_loss = losses.DenoisingAutoEncoderLoss(model,
decoder_name_or_path=model_name,
tie_encoder_decoder=True)
# Create a dev evaluator
dev_evaluator = evaluation.RerankingEvaluator(dev_dataset,
name='AskUbuntu dev')
logging.info("Dev performance before training")
dev_evaluator(model)