data_setup.py

import os
import csv
import datasets

class DatasetConfig(datasets.BuilderConfig):
    """BuilderConfig for MyCustomDataset."""
    def __init__(self, name: str, subset_dir: str, **kwargs):
        """
        Args:
            name: The name of the dataset configuration (e.g., '4k', '8k').
                  This is crucial for datasets to identify the config.
            subset_dir: The name of the subdirectory within 'data/' for this subset (e.g., "4k").
            **kwargs: Keyword arguments passed to the base BuilderConfig (e.g., 'description').
        """
        super().__init__(name, **kwargs)
        self.subset_dir = subset_dir


# Main dataset builder class
class MLNeedle(datasets.GeneratorBasedBuilder):
  
    VERSION = datasets.Version("1.0.0")
    BUILDER_CONFIGS = [
        DatasetConfig(name="baseline", subset_dir="baseline", description="Baseline subset data."),
        DatasetConfig(name="4k", subset_dir="4k", description="Dataset subset with 4k context size."),
        DatasetConfig(name="8k", subset_dir="8k", description="Dataset subset with 8k context size."),
        DatasetConfig(name="16k", subset_dir="16k", description="Dataset subset with 16k context size."),
        DatasetConfig(name="32k", subset_dir="32k", description="Dataset subset with 32k context size."),
    ]

    DEFAULT_CONFIG_NAME = "baseline"

    _BUILDER_CONFIGS_GROUPED_BY_DATASET_NAME = {
        config.name: config for config in BUILDER_CONFIGS
    }

    def _info(self):
        """Defines the dataset schema and metadata."""
        return datasets.DatasetInfo(
            description="""
            Multilingual Needle in a Haystack dataset for evaluating large language models
            on their ability to retrieve specific information from long contexts across multiple languages.
            Each subset (e.g., 4k, 8k) corresponds to different context lengths, and each split
            (e.g., en, es) represents a language.
            """,
            features=datasets.Features({
                "id": datasets.Value("string"),
                "needle_lang": datasets.Value("string"),
                "question_lang": datasets.Value("string"),
                "distractor_lang": datasets.Value("string"),
                "needle_position": datasets.Value("string"),
                "answer_text_format": datasets.Value("string"),
                "answer_start_index": datasets.Value("int32"),
                "answer_sentence": datasets.Value("string"),
                "prompt": datasets.Value("string"),
            }),
            supervised_keys=None,
            homepage="https://huggingface.co/datasets/ameyhengle/Multilingual-Needle-in-a-Haystack",
            license="MIT",
        )

    def _generate_examples(self, filepath):
        """
        Args:
            filepath: The full path to the CSV file to be processed.
        """
        with open(filepath, encoding="utf-8") as f:
            reader = csv.DictReader(f)
            for i, row in enumerate(reader):
                try:
                    answer_start_index = int(row["answer_start_index"])
                except (ValueError, TypeError):
                    print(f"Warning: Could not convert 'answer_start_index' to int for row {i} in {filepath}. Defaulting to 0.")
                    answer_start_index = 0

                # Yield the example, mapping CSV columns to dataset features
                yield i, {
                    "id": row["id"],
                    "needle_lang": row["needle_lang"],
                    "question_lang": row["question_lang"],
                    "distractor_lang": row["distractor_lang"],
                    "needle_position": row["needle_position"],
                    "answer_text_format": row["answer_text_format"],
                    "answer_start_index": answer_start_index,
                    "answer_sentence": row["answer_sentence"],
                    "prompt": row["prompt"],
                }