train.py

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


def main() -> None:
    # Load the dataset, ensure "tokens" and "ner_tags" columns, and get a list of labels
    dataset = load_dataset("tner/ontonotes5")
    dataset = dataset.rename_column("tags", "ner_tags")
    labels = ['O', 'B-CARDINAL', 'B-DATE', 'I-DATE', 'B-PERSON', 'I-PERSON', 'B-NORP', 'B-GPE', 'I-GPE', 'B-LAW', 'I-LAW', 'B-ORG', 'I-ORG', 'B-PERCENT', 'I-PERCENT', 'B-ORDINAL', 'B-MONEY', 'I-MONEY', 'B-WORK_OF_ART', 'I-WORK_OF_ART', 'B-FAC', 'B-TIME', 'I-CARDINAL', 'B-LOC', 'B-QUANTITY', 'I-QUANTITY', 'I-NORP', 'I-LOC', 'B-PRODUCT', 'I-TIME', 'B-EVENT', 'I-EVENT', 'I-FAC', 'B-LANGUAGE', 'I-PRODUCT', 'I-ORDINAL', 'I-LANGUAGE']

    # Initialize a SpanMarker model using a pretrained BERT-style encoder
    model_name = "roberta-large"
    model = SpanMarkerModel.from_pretrained(
        model_name,
        labels=labels,
        # SpanMarker hyperparameters:
        model_max_length=256,
        marker_max_length=128,
        entity_max_length=10,
    )

    # Prepare the 🤗 transformers training arguments
    args = TrainingArguments(
        output_dir="models/span_marker_roberta_large_ontonotes5",
        # Training Hyperparameters:
        learning_rate=1e-5,
        per_device_train_batch_size=8,
        per_device_eval_batch_size=8,
        gradient_accumulation_steps=2,
        num_train_epochs=4,
        weight_decay=0.01,
        warmup_ratio=0.1,
        bf16=True,
        # Other Training parameters
        logging_first_step=True,
        logging_steps=50,
        evaluation_strategy="steps",
        save_strategy="steps",
        eval_steps=1000,
        dataloader_num_workers=2,
    )

    # Initialize the trainer using our model, training args & dataset, and train
    trainer = Trainer(
        model=model,
        args=args,
        train_dataset=dataset["train"],
        eval_dataset=dataset["validation"],
    )
    trainer.train()
    trainer.save_model("models/span_marker_roberta_large_ontonotes5/checkpoint-final")

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


if __name__ == "__main__":
    main()