scripts/download_data.py

"""
UIPress: Download all required datasets and verify integrity.

Usage:
    python scripts/download_data.py
    python scripts/download_data.py --skip_webcode  # skip WebCode2M
"""

# ---- HuggingFace 镜像 (必须在其他 import 之前) ----
import os
os.environ["HF_ENDPOINT"] = "https://hf-mirror.com"
os.environ["HF_HOME"] = "/root/rivermind-data/huggingface"

import argparse
import subprocess
import sys
from pathlib import Path

PROJECT_ROOT = Path(__file__).parent.parent
DATA_DIR = PROJECT_ROOT / "data"
REPOS_DIR = PROJECT_ROOT / "repos"


def run(cmd, cwd=None):
    """Run shell command with error handling."""
    print(f"  >> {cmd}")
    result = subprocess.run(cmd, shell=True, cwd=cwd, capture_output=True, text=True)
    if result.returncode != 0:
        print(f"  [WARN] {result.stderr.strip()}")
    return result.returncode == 0


def download_design2code():
    """Download Design2Code dataset from HuggingFace."""
    print("\n[1/4] Downloading Design2Code dataset...")
    d2c_dir = DATA_DIR / "design2code"

    if d2c_dir.exists() and any(d2c_dir.iterdir()):
        print("  Already exists, skipping.")
        return

    d2c_dir.mkdir(parents=True, exist_ok=True)

    try:
        from datasets import load_dataset
        # Use Design2Code-hf: cleaner format with 'image' + 'text' (HTML code) columns
        print("  Loading from HuggingFace: SALT-NLP/Design2Code-hf...")
        try:
            ds = load_dataset("SALT-NLP/Design2Code-hf")
        except Exception:
            print("  Fallback to SALT-NLP/Design2Code...")
            ds = load_dataset("SALT-NLP/Design2Code")
        ds.save_to_disk(str(d2c_dir))
        for split in ds:
            print(f"  Split '{split}': {len(ds[split])} samples")
    except Exception as e:
        print(f"  [ERROR] HuggingFace download failed: {e}")
        print("  Fallback: trying git clone of the eval repo (includes test data)...")
        clone_design2code_repo()


def clone_design2code_repo():
    """Clone Design2Code GitHub repo for evaluation code."""
    print("\n[2/4] Cloning Design2Code evaluation repo...")
    repo_dir = REPOS_DIR / "Design2Code"

    if repo_dir.exists():
        print("  Already exists, skipping.")
        return

    REPOS_DIR.mkdir(parents=True, exist_ok=True)
    run("git clone https://github.com/NoviScl/Design2Code.git", cwd=str(REPOS_DIR))

    if repo_dir.exists():
        print(f"  Cloned to {repo_dir}")
        # Check if testset exists in the repo
        testset = repo_dir / "testset_final"
        if testset.exists():
            print(f"  Found testset at {testset}")
            # Symlink to data dir for convenience
            link_path = DATA_DIR / "testset_final"
            if not link_path.exists():
                os.symlink(str(testset), str(link_path))
                print(f"  Symlinked to {link_path}")
    else:
        print("  [ERROR] Clone failed. Please clone manually:")
        print("    git clone https://github.com/NoviScl/Design2Code.git repos/Design2Code")


def clone_deepseek_ocr_repo():
    """Clone DeepSeek-OCR repo for reference code."""
    print("\n[3/4] Cloning DeepSeek-OCR repo...")
    repo_dir = REPOS_DIR / "DeepSeek-OCR"

    if repo_dir.exists():
        print("  Already exists, skipping.")
        return

    REPOS_DIR.mkdir(parents=True, exist_ok=True)
    run("git clone https://github.com/deepseek-ai/DeepSeek-OCR.git", cwd=str(REPOS_DIR))

    if repo_dir.exists():
        print(f"  Cloned to {repo_dir}")
    else:
        print("  [ERROR] Clone failed. Please clone manually.")


def verify_model_access():
    """Verify HuggingFace model access (doesn't download full weights)."""
    print("\n[4/4] Verifying model access on HuggingFace...")
    models = [
        "Qwen/Qwen3-VL-2B-Instruct",
        "Qwen/Qwen2.5-VL-7B-Instruct",
        "deepseek-ai/DeepSeek-OCR",
    ]
    for model_id in models:
        try:
            from huggingface_hub import model_info
            info = model_info(model_id)
            size_gb = sum(s.size for s in info.siblings if s.rfilename.endswith(('.safetensors', '.bin')) and s.size is not None) / 1e9
            print(f"  {model_id}: OK ({size_gb:.1f} GB)")
        except Exception as e:
            print(f"  {model_id}: [WARN] {e}")
            print(f"    Model will be downloaded on first use.")


def print_summary():
    """Print summary of downloaded data."""
    print("\n" + "=" * 60)
    print("DOWNLOAD SUMMARY")
    print("=" * 60)

    # Check data
    for name, path in [
        ("Design2Code (HF)", DATA_DIR / "design2code"),
        ("Design2Code (raw)", DATA_DIR / "testset_final"),
    ]:
        status = "OK" if path.exists() and any(path.iterdir()) else "MISSING"
        print(f"  {name:<30} [{status}] {path}")

    # Check repos
    for name, path in [
        ("Design2Code eval repo", REPOS_DIR / "Design2Code"),
        ("DeepSeek-OCR repo", REPOS_DIR / "DeepSeek-OCR"),
    ]:
        status = "OK" if path.exists() else "MISSING"
        print(f"  {name:<30} [{status}] {path}")

    print("=" * 60)
    print("\nNext: run baseline evaluation:")
    print("  python scripts/step1_baseline.py --model qwen3_vl_2b --max_samples 5")


def main():
    parser = argparse.ArgumentParser(description="Download UIPress datasets")
    parser.add_argument("--skip_webcode", action="store_true",
                        help="Skip WebCode2M download")
    args = parser.parse_args()

    DATA_DIR.mkdir(parents=True, exist_ok=True)
    REPOS_DIR.mkdir(parents=True, exist_ok=True)

    download_design2code()
    clone_design2code_repo()
    clone_deepseek_ocr_repo()
    verify_model_access()
    print_summary()


if __name__ == "__main__":
    main()