scripts/upload_hf.py

#!/usr/bin/env python3
"""
Upload Vietnamese Legal Corpus to Hugging Face Hub
"""

import re
from pathlib import Path
from datetime import datetime

import click
from rich.console import Console
from rich.progress import Progress, SpinnerColumn, TextColumn, BarColumn, TaskProgressColumn
from datasets import Dataset, DatasetDict, Features, Value
from huggingface_hub import HfApi, login

console = Console()

DATA_DIR = Path(__file__).parent.parent / "data"


def parse_front_matter(content: str) -> dict:
    """Parse YAML front matter from markdown file."""
    metadata = {}
    if content.startswith("---"):
        parts = content.split("---", 2)
        if len(parts) >= 3:
            for line in parts[1].strip().split("\n"):
                if ":" in line:
                    key, value = line.split(":", 1)
                    value = value.strip().strip('"')
                    metadata[key.strip()] = value
    return metadata


def extract_body(content: str) -> str:
    """Extract body content after front matter."""
    if content.startswith("---"):
        parts = content.split("---", 2)
        if len(parts) >= 3:
            return parts[2].strip()
    return content


def load_corpus() -> list[dict]:
    """Load all law files into a list of records."""
    records = []
    files = sorted(DATA_DIR.glob("*.md"))

    with Progress(
        SpinnerColumn(),
        TextColumn("[progress.description]{task.description}"),
        BarColumn(),
        TaskProgressColumn(),
        console=console,
    ) as progress:
        task = progress.add_task("Loading files...", total=len(files))

        for f in files:
            content = f.read_text(encoding="utf-8")
            metadata = parse_front_matter(content)
            body = extract_body(content)

            # Check if has real content
            has_content = len(body) > 200 and "*Nội dung chưa được tải xuống.*" not in body

            record = {
                "id": f.stem,
                "filename": f.name,
                "title": metadata.get("title", ""),
                "title_en": metadata.get("title_en", ""),
                "type": metadata.get("type", ""),
                "year": int(metadata.get("year", 0)) if metadata.get("year", "").isdigit() else 0,
                "document_number": metadata.get("document_number", ""),
                "effective_date": metadata.get("effective_date", ""),
                "status": metadata.get("status", ""),
                "url": metadata.get("url", ""),
                "downloaded_at": metadata.get("downloaded_at", ""),
                "has_content": has_content,
                "content": body if has_content else "",
                "content_length": len(body),
            }
            records.append(record)
            progress.advance(task)

    return records


def create_dataset(records: list[dict]) -> Dataset:
    """Create a Hugging Face Dataset from records."""
    features = Features({
        "id": Value("string"),
        "filename": Value("string"),
        "title": Value("string"),
        "title_en": Value("string"),
        "type": Value("string"),
        "year": Value("int32"),
        "document_number": Value("string"),
        "effective_date": Value("string"),
        "status": Value("string"),
        "url": Value("string"),
        "downloaded_at": Value("string"),
        "has_content": Value("bool"),
        "content": Value("string"),
        "content_length": Value("int32"),
    })

    return Dataset.from_list(records, features=features)


@click.group()
def cli():
    """Upload VLC corpus to Hugging Face."""
    pass


@cli.command()
def preview():
    """Preview the dataset before uploading."""
    if not DATA_DIR.exists():
        console.print("[red]Data directory not found[/red]")
        return

    records = load_corpus()
    dataset = create_dataset(records)

    console.print("\n[bold blue]Dataset Preview[/bold blue]\n")
    console.print(dataset)
    console.print(f"\n[bold]Features:[/bold]")
    for name, feat in dataset.features.items():
        console.print(f"  {name}: {feat}")

    # Statistics
    codes = sum(1 for r in records if r["type"] == "code")
    laws = sum(1 for r in records if r["type"] == "law")
    with_content = sum(1 for r in records if r["has_content"])
    total_chars = sum(r["content_length"] for r in records)

    console.print(f"\n[bold]Statistics:[/bold]")
    console.print(f"  Total records: {len(records)}")
    console.print(f"  Codes (Bộ luật): {codes}")
    console.print(f"  Laws (Luật): {laws}")
    console.print(f"  With content: {with_content}")
    console.print(f"  Total characters: {total_chars:,}")

    # Year distribution
    years = {}
    for r in records:
        y = r["year"]
        years[y] = years.get(y, 0) + 1

    console.print(f"\n[bold]By Year (top 10):[/bold]")
    for year in sorted(years.keys(), reverse=True)[:10]:
        console.print(f"  {year}: {years[year]}")

    # Sample records
    console.print(f"\n[bold]Sample Records:[/bold]")
    for r in records[:3]:
        console.print(f"  - {r['title']} ({r['document_number']}, {r['year']})")


@cli.command()
@click.option("--repo-id", default="undertheseanlp/vietnamese-legal-corpus", help="Hugging Face repo ID")
@click.option("--private", is_flag=True, help="Make the dataset private")
@click.option("--token", envvar="HF_TOKEN", help="Hugging Face token")
def upload(repo_id: str, private: bool, token: str):
    """Upload dataset to Hugging Face Hub."""
    if not DATA_DIR.exists():
        console.print("[red]Data directory not found[/red]")
        return

    # Login
    if token:
        login(token=token)
    else:
        console.print("[yellow]No token provided. Using cached credentials.[/yellow]")

    # Load and create dataset
    console.print("[bold]Loading corpus...[/bold]")
    records = load_corpus()
    dataset = create_dataset(records)

    console.print(f"\n[bold]Dataset:[/bold] {dataset}")

    # Upload
    console.print(f"\n[bold]Uploading to {repo_id}...[/bold]")

    dataset.push_to_hub(
        repo_id,
        private=private,
        commit_message=f"Upload Vietnamese Legal Corpus ({len(records)} documents)",
    )

    console.print(f"\n[green]Done![/green] Dataset uploaded to: https://huggingface.co/datasets/{repo_id}")


@cli.command()
@click.option("--output", "-o", default="vlc_dataset", help="Output directory")
def save_local(output: str):
    """Save dataset locally in Arrow format."""
    if not DATA_DIR.exists():
        console.print("[red]Data directory not found[/red]")
        return

    output_path = Path(output)

    console.print("[bold]Loading corpus...[/bold]")
    records = load_corpus()
    dataset = create_dataset(records)

    console.print(f"\n[bold]Dataset:[/bold] {dataset}")

    # Save
    console.print(f"\n[bold]Saving to {output_path}...[/bold]")
    dataset.save_to_disk(output_path)

    console.print(f"\n[green]Done![/green] Dataset saved to: {output_path}")


@cli.command()
@click.option("--output", "-o", default="vlc_dataset.parquet", help="Output parquet file")
def save_parquet(output: str):
    """Save dataset as Parquet file."""
    if not DATA_DIR.exists():
        console.print("[red]Data directory not found[/red]")
        return

    output_path = Path(output)

    console.print("[bold]Loading corpus...[/bold]")
    records = load_corpus()
    dataset = create_dataset(records)

    console.print(f"\n[bold]Dataset:[/bold] {dataset}")

    # Save
    console.print(f"\n[bold]Saving to {output_path}...[/bold]")
    dataset.to_parquet(output_path)

    console.print(f"\n[green]Done![/green] Dataset saved to: {output_path}")
    console.print(f"  Size: {output_path.stat().st_size / 1024 / 1024:.2f} MB")


if __name__ == "__main__":
    cli()