export_hf_model.py

import argparse
from pathlib import Path

from transformers import AutoTokenizer

from custom_text_classifier import CustomTextForSequenceClassification
from custom_text_config import CustomTextConfig


def parse_args():
    parser = argparse.ArgumentParser(description="Export model files for Hugging Face")
    parser.add_argument("--output-dir", default="./hf_model")
    parser.add_argument("--backbone", default="distilbert-base-uncased")
    return parser.parse_args()


def main():
    args = parse_args()
    output_dir = Path(args.output_dir)
    output_dir.mkdir(parents=True, exist_ok=True)

    config = CustomTextConfig(
        backbone_name_or_path=args.backbone,
        num_labels=2,
        id2label={0: "negative", 1: "positive"},
        label2id={"negative": 0, "positive": 1},
    )
    model = CustomTextForSequenceClassification(config)
    tokenizer = AutoTokenizer.from_pretrained(args.backbone)

    model.save_pretrained(str(output_dir))
    tokenizer.save_pretrained(str(output_dir))

    model_card = output_dir / "README.md"
    model_card.write_text(
        "\n".join(
            [
                "---",
                "library_name: transformers",
                "pipeline_tag: text-classification",
                "tags:",
                "- custom-code",
                "- imdb",
                "---",
                "",
                "# Custom Text Classifier",
                "",
                "This repository contains a custom `PreTrainedModel` sequence classifier.",
                f"The backbone is `{args.backbone}` and the classification head has two labels.",
                "",
                "Load with:",
                "",
                "```python",
                "from transformers import AutoModelForSequenceClassification, AutoTokenizer",
                "",
                "tokenizer = AutoTokenizer.from_pretrained(repo_id, trust_remote_code=True)",
                "model = AutoModelForSequenceClassification.from_pretrained(",
                "    repo_id,",
                "    trust_remote_code=True,",
                ")",
                "```",
                "",
            ]
        ),
        encoding="utf-8",
    )

    print("exported model folder:", output_dir)


if __name__ == "__main__":
    main()