training_stsbenchmark.py

import logging
import argparse

from datetime import datetime
from datasets import load_dataset
from sentence_transformers import SentenceTransformer, losses
from sentence_transformers.evaluation import EmbeddingSimilarityEvaluator
from sentence_transformers.similarity_functions import SimilarityFunction
from sentence_transformers.trainer import SentenceTransformerTrainer
from sentence_transformers.training_args import SentenceTransformerTrainingArguments

# Set the log level to INFO to get more information
logging.basicConfig(format="%(asctime)s - %(message)s", datefmt="%Y-%m-%d %H:%M:%S", level=logging.INFO)

# params
parser = argparse.ArgumentParser()
parser.add_argument('--train_batch_size', type=int, default=16)
parser.add_argument('--num_epochs', type=int, default=10)
parser.add_argument('--model_name_or_path', type=str, default="bert-base-uncased")
parser.add_argument('--save_root_path', type=str, default="output", help='Model output folder')
parser.add_argument('--lr', default=2e-05)
parser.add_argument('--eval_steps', type=int, default=-1)
parser.add_argument('--save_steps', type=int, default=-1)
parser.add_argument('--save_total_limit', type=int, default=2)
parser.add_argument('--logging_steps', type=int, default=10)
args = parser.parse_args()

# You can specify any Hugging Face pre-trained model here, for example, bert-base-uncased, roberta-base, xlm-roberta-base
model_name = args.model_name_or_path
train_batch_size = args.train_batch_size
num_epochs = args.num_epochs

output_dir = (
    args.save_root_path + "/training_stsbenchmark_" + model_name.replace("/", "-") + "-" + datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
)

# 1. Here we define our SentenceTransformer model. If not already a Sentence Transformer model, it will automatically
# create one with "mean" pooling.
model = SentenceTransformer(model_name)

logging.info("Loading STSbenchmark train dataset")
# 2. Load the STSB dataset: https://huggingface.co/datasets/sentence-transformers/stsb
train_dataset = load_dataset("sentence-transformers/stsb", split="train")
eval_dataset = load_dataset("sentence-transformers/stsb", split="validation")
test_dataset = load_dataset("sentence-transformers/stsb", split="test")
logging.info(train_dataset)

# 3. Define our training loss
# CosineSimilarityLoss (https://sbert.net/docs/package_reference/sentence_transformer/losses.html#cosinesimilarityloss) needs two text columns and one
# similarity score column (between 0 and 1)
train_loss = losses.CosineSimilarityLoss(model=model)
# train_loss = losses.CoSENTLoss(model=model)

# 4. Define an evaluator for use during training. This is useful to keep track of alongside the evaluation loss.
dev_evaluator = EmbeddingSimilarityEvaluator(
    sentences1=eval_dataset["sentence1"],
    sentences2=eval_dataset["sentence2"],
    scores=eval_dataset["score"],
    main_similarity=SimilarityFunction.COSINE,
    name="sts-dev",
)

# 5. Define the training arguments
args = SentenceTransformerTrainingArguments(
    # Required parameter:
    output_dir=output_dir,
    # Optional training parameters:
    num_train_epochs=num_epochs,
    per_device_train_batch_size=train_batch_size,
    per_device_eval_batch_size=train_batch_size,
    warmup_ratio=0.1,
    fp16=True,  # Set to False if you get an error that your GPU can't run on FP16
    bf16=False, # Set to True if you have a GPU that supports BF16
    # Optional tracking/debugging parameters:
    evaluation_strategy="steps",
    eval_steps=args.eval_steps,
    save_strategy="steps",
    save_steps=args.save_steps,
    save_total_limit=args.save_total_limit,
    logging_steps=args.logging_steps,
    run_name="sts",  # Will be used in W&B if `wandb` is installed
)

# 6. Create the trainer & start training
trainer = SentenceTransformerTrainer(
    model=model,
    args=args,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    loss=train_loss,
    evaluator=dev_evaluator,
)
trainer.train()

# 7. Evaluate the model performance on the STS Benchmark test dataset
test_evaluator = EmbeddingSimilarityEvaluator(
    sentences1=test_dataset["sentence1"],
    sentences2=test_dataset["sentence2"],
    scores=test_dataset["score"],
    main_similarity=SimilarityFunction.COSINE,
    name="sts-test",
)
test_evaluator(model, output_path=output_dir)

# 8. Save the trained & evaluated model locally
final_output_dir = f"{output_dir}/final"
model.save(final_output_dir)