ds.py

from datasets import load_dataset, concatenate_datasets
import json
import os

# Load the dataset
print("Loading dataset...")
dataset = load_dataset("unsloth/LaTeX_OCR")

def add_question_and_rename(example):
    """Add question field and rename text to answer"""
    # Add the question field
    example["question"] = "Convert this into LaTeX code, only provide the LaTeX code."
    
    # Rename 'text' to 'answer'
    if "text" in example:
        example["answer"] = example["text"]
        del example["text"]
    
    return example

print("Modifying dataset...")
# Apply the transformation to all splits
modified_dataset = dataset.map(add_question_and_rename)

# Create output directories
os.makedirs("./parquet_files", exist_ok=True)
os.makedirs("./modified_latex_ocr_dataset", exist_ok=True)

# Save locally (this handles images properly)
print("Saving dataset locally...")
modified_dataset.save_to_disk("./modified_latex_ocr_dataset")

# Save as individual Parquet files per split
print("Saving as Parquet files (per split)...")
for split_name, split_data in modified_dataset.items():
    parquet_file = f"./parquet_files/modified_latex_ocr_{split_name}.parquet"
    split_data.to_parquet(parquet_file)
    print(f"Saved {split_name} split to {parquet_file}")

# Save as single consolidated Parquet file (if you have multiple splits)
if len(modified_dataset) > 1:
    print("Saving as consolidated Parquet file...")
    # Add split information to each example
    all_datasets = []
    for split_name, split_data in modified_dataset.items():
        # Add split name to each example
        split_data_with_info = split_data.map(lambda x: {**x, "split": split_name})
        all_datasets.append(split_data_with_info)
    
    # Concatenate all splits
    consolidated_dataset = concatenate_datasets(all_datasets)
    consolidated_dataset.to_parquet("./parquet_files/modified_latex_ocr_all.parquet")
    print("Saved consolidated dataset to ./parquet_files/modified_latex_ocr_all.parquet")

# Save just the text data as JSON for inspection (excluding images)
print("Saving text-only JSON files...")
for split_name, split_data in modified_dataset.items():
    text_only_data = []
    for i, sample in enumerate(split_data):
        text_sample = {
            "id": i,
            "question": sample["question"],
            "answer": sample["answer"]
        }
        # Add any other non-image fields if they exist
        for key, value in sample.items():
            if key not in ["image", "question", "answer"] and not hasattr(value, 'save'):
                text_sample[key] = value
        
        text_only_data.append(text_sample)
    
    # Save text data as JSON
    output_file = f"modified_latex_ocr_{split_name}_text_only.json"
    with open(output_file, 'w') as f:
        json.dump(text_only_data, f, indent=2, ensure_ascii=False)
    print(f"Saved {split_name} text data to {output_file}")

print("\n" + "="*50)
print("DATASET MODIFICATION COMPLETE!")
print("="*50)

# Print dataset info
print(f"\nDataset structure: {modified_dataset}")
for split_name in modified_dataset.keys():
    print(f"  {split_name} split: {len(modified_dataset[split_name])} samples")

# Show file information
print(f"\nOutput files created:")
print(f"  📁 ./modified_latex_ocr_dataset/ - Full dataset with images")
print(f"  📁 ./parquet_files/ - Parquet files")

if os.path.exists("./parquet_files"):
    for file in sorted(os.listdir("./parquet_files")):
        if file.endswith(".parquet"):
            file_path = os.path.join("./parquet_files", file)
            size_mb = os.path.getsize(file_path) / (1024 * 1024)
            print(f"    📄 {file}: {size_mb:.2f} MB")

# Show JSON files
json_files = [f for f in os.listdir(".") if f.startswith("modified_latex_ocr") and f.endswith(".json")]
if json_files:
    print(f"  📄 Text-only JSON files:")
    for file in sorted(json_files):
        size_kb = os.path.getsize(file) / 1024
        print(f"    📄 {file}: {size_kb:.2f} KB")

# Show a sample
if len(modified_dataset) > 0:
    first_split = list(modified_dataset.keys())[0]
    sample = modified_dataset[first_split][0]
    print(f"\n📋 Sample from {first_split} split:")
    print(f"   Keys: {list(sample.keys())}")
    print(f"   Question: {sample['question']}")
    if 'image' in sample:
        print(f"   Image: {type(sample['image'])}")
    answer_preview = sample['answer'][:100] + "..." if len(sample['answer']) > 100 else sample['answer']
    print(f"   Answer: {answer_preview}")