hf_job_add_dimensions.py

#!/usr/bin/env python3
"""
Add height/width columns to nastol-images-full dataset
"""
import argparse
import os
import logging
import sys
from pathlib import Path

logging.basicConfig(
    level=logging.INFO,
    format='[%(asctime)s] %(levelname)s: %(message)s',
    datefmt='%Y-%m-%d %H:%M:%S',
    stream=sys.stdout,
    force=True
)
logger = logging.getLogger(__name__)

from datasets import load_dataset
from huggingface_hub import HfApi
import pyarrow as pa
import pyarrow.parquet as pq


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument('--input-dataset', type=str, default='vlordier/nastol-images-full')
    ap.add_argument('--output-dataset', type=str, default='vlordier/nastol-images-full')
    ap.add_argument('--split', type=str, default='train')
    ap.add_argument('--shard-index', type=int, default=0)
    ap.add_argument('--num-shards', type=int, default=1)
    ap.add_argument('--batch-size', type=int, default=1000)
    args = ap.parse_args()
    
    logger.info("="*60)
    logger.info("Add Height/Width Columns to Dataset")
    logger.info("="*60)
    logger.info(f"Arguments: {vars(args)}")
    
    token = os.environ.get('HF_TOKEN')
    api = HfApi(token=token)
    
    # Load dataset
    logger.info(f"Loading {args.input_dataset}...")
    ds = load_dataset(args.input_dataset, split=args.split, streaming=True)
    
    if args.num_shards > 1:
        ds = ds.shard(num_shards=args.num_shards, index=args.shard_index)
        logger.info(f"Processing shard {args.shard_index+1}/{args.num_shards}")
    
    # Process in batches
    buffer = []
    batch_count = 0
    upload_count = 0
    
    def flush_buffer():
        nonlocal buffer, upload_count
        if not buffer:
            return
        
        # Build columns
        image_paths = [b['image_path'] for b in buffer]
        images_bytes = [b['image'] for b in buffer]
        heights = [b['height'] for b in buffer]
        widths = [b['width'] for b in buffer]
        
        table = pa.table({
            'image_path': image_paths,
            'image': images_bytes,
            'height': heights,
            'width': widths
        })
        
        # Write parquet
        local_dir = Path('dimension_batches')
        local_dir.mkdir(parents=True, exist_ok=True)
        file_name = f"shard-{args.shard_index:03d}-batch-{upload_count:04d}.parquet"
        local_path = local_dir / file_name
        pq.write_table(table, local_path)
        
        # Upload
        path_in_repo = f"data/{file_name}"
        logger.info(f"Uploading batch {upload_count} with {len(buffer)} images -> {path_in_repo}")
        try:
            api.upload_file(
                path_or_fileobj=str(local_path),
                path_in_repo=path_in_repo,
                repo_id=args.output_dataset,
                repo_type='dataset',
                token=token
            )
            logger.info("✓ Uploaded")
        except Exception as e:
            logger.error(f"Upload failed: {e}")
        
        buffer.clear()
        upload_count += 1
    
    logger.info("Processing images...")
    for idx, sample in enumerate(ds):
        image = sample['image']
        image_path = sample.get('image_path', f'img_{idx:06d}')
        
        # Get dimensions
        width, height = image.size
        
        # Store original image bytes
        import io
        buf = io.BytesIO()
        image.save(buf, format='PNG')
        image_bytes = buf.getvalue()
        
        buffer.append({
            'image_path': image_path,
            'image': image_bytes,
            'height': height,
            'width': width
        })
        
        if len(buffer) >= args.batch_size:
            flush_buffer()
            batch_count += 1
            logger.info(f"Processed {batch_count * args.batch_size} images")
    
    # Final flush
    flush_buffer()
    logger.info(f"✓ Completed shard {args.shard_index}: {batch_count * args.batch_size + len(buffer)} images")


if __name__ == '__main__':
    main()