Mini Squad

A simple transformation on the SQuAD dataset for training tiny language models.

Overview

The Mini Squad dataset is a modified version of the Stanford Question Answering Dataset (SQuAD). It focuses on extracting concise context sentences around each answer, making it suitable for training small-scale language models or fine-tuning lightweight architectures.

Key Features

Reduced Context: Extracts only the sentence containing the answer, bounded by sentence-ending punctuation (period, question mark, exclamation point, or semicolon).
Simplified Format: Each entry includes context, question, and answer, providing a clean and easy-to-use structure.
Preprocessed for Lightweight Models: Designed to minimize memory and computational requirements for smaller models.

Dataset Structure

The dataset consists of two splits:

train.json: Training set.
validation.json: Validation set.

Each file is a JSON Lines file, where each line is a dictionary with the following fields:

context: The extracted sentence containing the answer.
question: The question from the original dataset.
answer: The corresponding answer.

Example Entry

{
    "context": "France where the Virgin Mary reputedly appeared to Saint Bernadette Soubirous in 1858.",
    "question": "To whom did the Virgin Mary allegedly appear in 1858 in Lourdes France?",
    "answer": "Saint Bernadette Soubirous"
}

Usage

Loading the Dataset

You can load the dataset using the Hugging Face datasets library:

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("zakerytclarke/mini_squad")

# Access the splits
train_df = dataset["train"].to_pandas()
validation_df = dataset["validation"].to_pandas()

print(train_df.head())

Applications

Fine-tuning small language models.
Training efficient QA systems.
Use as a benchmark for lightweight NLP architectures.

File Structure

mini-squad/
|— train.json
|— validation.json

Citation

If you use Mini Squad in your research or applications, please cite the original SQuAD dataset:

@article{rajpurkar2016squad,
  title={SQuAD: 100,000+ Questions for Machine Comprehension of Text},
  author={Rajpurkar, Pranav and Zhang, Jian and Lopyrev, Konstantin and Liang, Percy},
  journal={arXiv preprint arXiv:1606.05250},
  year={2016}
}

License

The Mini Squad dataset inherits the license of the original SQuAD dataset. Please refer to the SQuAD license for details.