MME-RealWorld-Base64.py

import json
import os
from datasets import Dataset, DatasetDict, Features, Value, Sequence

def load_json_lines(file_path):
    with open(file_path, 'r') as f:
        for line in f:
            yield json.loads(line)

def _generate_examples(chunk_files):
    for file in chunk_files:
        for example in load_json_lines(file):
            yield example['Question_id'], example

def load_dataset(*args, **kwargs):
    dataset_dir = kwargs['data_dir']  # The directory containing JSON chunks
    chunk_files = [os.path.join(dataset_dir, f) for f in os.listdir(dataset_dir) if f.endswith('.json')]
    features = Features({
        'index': Value(dtype='string'),
        'image': Value(dtype='string'),
        'question': Value(dtype='string'),
        'multi-choice options': Sequence(Value(dtype='string')),
        'answer': Value(dtype='string'),
        'category': Value(dtype='string'),
        'l2-category': Value(dtype='string'),
    })
    dataset = DatasetDict({
        'train': Dataset.from_generator(lambda: _generate_examples(chunk_files), features=features)
    })
    return dataset