Keyphrase extraction model with SpanMarker

train_span_marker_keyphrase.py

from datasets import load_dataset, concatenate_datasets
from transformers import TrainingArguments
from span_marker import SpanMarkerModel, Trainer


def main() -> None:
    # Load the dataset, ensure "tokens" and "ner_tags" columns, and get a list of labels
    dataset = load_dataset("midas/inspec", "extraction")
    dataset = dataset.rename_columns({"document": "tokens", "doc_bio_tags": "ner_tags"})
    # Map string labels to integer labels instead
    real_labels = ["O", "B", "I"]
    dataset = dataset.map(lambda sample: {"ner_tags": [real_labels.index(tag) for tag in sample]}, input_columns="ner_tags")
    # Use more readable labels
    labels = ["O", "B-KEY", "I-KEY"]
    # Train using train + validation set.
    train_dataset = concatenate_datasets((dataset["train"], dataset["validation"]))

    # Initialize a SpanMarker model using a pretrained BERT-style encoder
    model_name = "bert-base-cased"
    model = SpanMarkerModel.from_pretrained(
        model_name,
        labels=labels,
        # SpanMarker hyperparameters:
        model_max_length=256,
        marker_max_length=128,
        entity_max_length=8,
    )

    # Prepare the 🤗 transformers training arguments
    args = TrainingArguments(
        output_dir=f"models/span_marker_bert_base_cased_keyphrase_inspec",
        run_name=f"bb_keyphrase",
        # Training Hyperparameters:
        learning_rate=5e-5,
        per_device_train_batch_size=32,
        per_device_eval_batch_size=32,
        num_train_epochs=3,
        weight_decay=0.01,
        warmup_ratio=0.1,
        bf16=True,  # Replace `bf16` with `fp16` if your hardware can't use bf16.
        # Other Training parameters
        logging_first_step=True,
        logging_steps=50,
        evaluation_strategy="no",
        save_strategy="steps",
        # eval_steps=300,
        save_total_limit=2,
        dataloader_num_workers=2,
    )

    # Initialize the trainer using our model, training args & dataset, and train
    trainer = Trainer(
        model=model,
        args=args,
        train_dataset=train_dataset
    )
    trainer.train()
    trainer.save_model(f"models/span_marker_bert_base_cased_keyphrase_inspec/checkpoint-final")

    # Compute & save the metrics on the test set
    metrics = trainer.evaluate(dataset["test"], metric_key_prefix="test")
    trainer.save_metrics("test", metrics)
    trainer.create_model_card()


if __name__ == "__main__":
    main()