train.py

from pathlib import Path
import random
import shutil
from datasets import load_dataset, concatenate_datasets, Features, Sequence, ClassLabel, Value, DatasetDict
from transformers import TrainingArguments
from span_marker import SpanMarkerModel, Trainer
from span_marker.model_card import SpanMarkerModelCardData
from huggingface_hub import upload_folder, upload_file


"""
FEATURES = Features({"tokens": Sequence(feature=Value(dtype='string')), "ner_tags": Sequence(feature=ClassLabel(names=['O', 'B-ORG', 'I-ORG']))})


def load_fewnerd():
    def mapper(sample):
        sample["ner_tags"] = [int(tag == 5) for tag in sample["ner_tags"]]
        sample["ner_tags"] = [2 if tag == 1 and idx > 0 and sample["ner_tags"][idx - 1] == 1 else tag for idx, tag in enumerate(sample["ner_tags"])]
        return sample
    
    dataset = load_dataset("DFKI-SLT/few-nerd", "supervised")
    dataset = dataset.map(mapper, remove_columns=["id", "fine_ner_tags"])
    dataset = dataset.cast(FEATURES)
    return dataset


def load_conll():
    label_mapping = {3: 1, 4: 2}
    def mapper(sample):
        sample["ner_tags"] = [label_mapping.get(tag, 0) for tag in sample["ner_tags"]]
        return sample

    dataset = load_dataset("conll2003")
    dataset = dataset.map(mapper, remove_columns=["id", "pos_tags", "chunk_tags"])
    dataset = dataset.cast(FEATURES)
    return dataset


def load_ontonotes():
    label_mapping = {11: 1, 12: 2}
    def mapper(sample):
        sample["ner_tags"] = [label_mapping.get(tag, 0) for tag in sample["ner_tags"]]
        return sample

    dataset = load_dataset("tner/ontonotes5")
    dataset = dataset.rename_column("tags", "ner_tags")
    dataset = dataset.map(mapper)
    dataset = dataset.cast(FEATURES)
    return dataset


def load_multinerd():
    label_mapping = {5: 1, 6: 2}
    def mapper(sample):
        sample["ner_tags"] = [label_mapping.get(tag, 0) for tag in sample["ner_tags"]]
        return sample

    def lang_filter(sample):
        return sample["lang"] == "en"

    dataset = load_dataset("Babelscape/multinerd")
    dataset = dataset.filter(lang_filter)
    dataset = dataset.map(mapper, remove_columns="lang")
    dataset = dataset.cast(FEATURES)
    return dataset


def preprocess_raw_dataset(raw_dataset):
    # Set the number of sentences without an org equal to the number of sentences with an org
    def has_org(sample):
        return bool(sum(sample["ner_tags"]))
    
    def has_no_org(sample):
        return not has_org(sample)

    dataset_org = raw_dataset.filter(has_org)
    dataset_no_org = raw_dataset.filter(has_no_org)
    dataset_no_org = dataset_no_org.select(random.sample(range(len(dataset_no_org)), k=len(dataset_org)))
    dataset = concatenate_datasets([dataset_org, dataset_no_org])
    return dataset
"""


def main() -> None:
    # Load the dataset, ensure "tokens" and "ner_tags" columns, and get a list of labels
    labels = ["O", "B-ORG", "I-ORG"]
    """
    fewnerd_dataset = load_fewnerd()
    conll_dataset = load_conll()
    ontonotes_dataset = load_ontonotes()
    multinerd_dataset = load_multinerd()

    raw_train_dataset = concatenate_datasets([fewnerd_dataset["train"], conll_dataset["train"], ontonotes_dataset["train"], multinerd_dataset["train"]])
    raw_eval_dataset = concatenate_datasets([fewnerd_dataset["validation"], conll_dataset["validation"], ontonotes_dataset["validation"], multinerd_dataset["validation"]])
    raw_test_dataset = concatenate_datasets([fewnerd_dataset["test"], conll_dataset["test"], ontonotes_dataset["test"], multinerd_dataset["test"]])

    train_dataset = preprocess_raw_dataset(raw_train_dataset)
    eval_dataset = preprocess_raw_dataset(raw_eval_dataset)
    test_dataset = preprocess_raw_dataset(raw_test_dataset)

    dataset_dict = DatasetDict({
        "train": train_dataset,
        "validation": eval_dataset,
        "test": test_dataset,
    })
    dataset_dict.push_to_hub("ner-orgs", private=True)
    """
    # breakpoint()
    dataset = load_dataset("tomaarsen/ner-orgs")

    train_dataset = dataset["train"]
    eval_dataset = dataset["validation"]
    eval_dataset = eval_dataset.select(random.sample(range(len(eval_dataset)), k=3000))
    test_dataset = dataset["test"]

    # Initialize a SpanMarker model using a pretrained BERT-style encoder
    encoder_id = "bert-base-cased"
    model_id = f"tomaarsen/span-marker-bert-base-orgs"
    model = SpanMarkerModel.from_pretrained(
        encoder_id,
        labels=labels,
        # SpanMarker hyperparameters:
        model_max_length=256,
        marker_max_length=128,
        entity_max_length=8,
        # Model card variables
        model_card_data=SpanMarkerModelCardData(
            model_id=model_id,
            encoder_id=encoder_id,
            dataset_name="FewNERD, CoNLL2003, OntoNotes v5, and MultiNERD",
            language=["en"],
        ),
    )

    # Prepare the 🤗 transformers training arguments
    output_dir = Path("models") / model_id
    args = TrainingArguments(
        output_dir=output_dir,
        run_name=model_id,
        # 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=100,
        evaluation_strategy="steps",
        save_strategy="steps",
        eval_steps=3000,
        save_total_limit=1,
        dataloader_num_workers=4,
    )

    # Initialize the trainer using our model, training args & dataset, and train
    trainer = Trainer(
        model=model,
        args=args,
        train_dataset=train_dataset,
        eval_dataset=eval_dataset,
    )
    trainer.train()

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

    # Save the model & training script locally
    trainer.save_model(output_dir / "checkpoint-final")
    shutil.copy2(__file__, output_dir / "checkpoint-final" / "train.py")

    # Upload everything to the Hub
    breakpoint()
    model.push_to_hub(model_id, private=True)
    upload_folder(folder_path=output_dir / "runs", path_in_repo="runs", repo_id=model_id)
    upload_file(path_or_fileobj=__file__, path_in_repo="train.py", repo_id=model_id)
    upload_file(path_or_fileobj=output_dir / "all_results.json", path_in_repo="all_results.json", repo_id=model_id)
    upload_file(path_or_fileobj=output_dir / "emissions.csv", path_in_repo="emissions.csv", repo_id=model_id)


if __name__ == "__main__":
    main()