Datasets:

imyangyixuan
/

hendrycks_math

dataset_info:
  features:
    - name: problem
      dtype: string
    - name: solution
      dtype: string
    - name: answer
      dtype: string
    - name: subject
      dtype: string
    - name: level
      dtype: int64
    - name: unique_id
      dtype: string
  splits:
    - name: train
      num_bytes: 8821866
      num_examples: 10798
    - name: validation
      num_bytes: 980388
      num_examples: 1200
    - name: test
      num_bytes: 400274
      num_examples: 500
  download_size: 5230576
  dataset_size: 10202528
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*

Note

Source: Original dataset has 2 splits — train (12,000) and test (500).
New validation split: train is re-split with seed 42 into 90% train (10,800) / 10% validation (1,200), stratified by level to preserve difficulty distribution. The original test (500) is kept as-is.
Stratification: level (int 1–5) is temporarily cast to a ClassLabel to enable stratified splitting, then removed.
Empty answer filtering: 2 training examples with empty answer fields (both from train/number_theory/, where \boxed{} has no content) are dropped.
Dropped (empty answer): {'train': 2, 'validation': 0, 'test': 0}
Final split sizes: {'train': 10798, 'validation': 1200, 'test': 500} — 12,498 total.

Hendrycks MATH Dataset

Dataset Description

The MATH dataset is a collection of mathematics competition problems designed to evaluate mathematical reasoning and problem-solving capabilities in computational systems. Containing 12,500 high school competition-level mathematics problems, this dataset is notable for including detailed step-by-step solutions alongside each problem.

Dataset Summary

The dataset consists of mathematics problems spanning multiple difficulty levels (1-5) and various mathematical subjects including:

Prealgebra
Algebra
Number Theory
Counting and Probability
Geometry
Intermediate Algebra
Precalculus

Each problem comes with:

A complete problem statement
A step-by-step solution
A final answer
Difficulty rating
Subject classification

Data Split

The dataset is divided into:

Training set: 12,000
Test set: 500 problems

Dataset Creation

Citation

@article{hendrycksmath2021,
    title={Measuring Mathematical Problem Solving With the MATH Dataset},
    author={Dan Hendrycks
    and Collin Burns
    and Saurav Kadavath
    and Akul Arora
    and Steven Basart
    and Eric Tang
    and Dawn Song
    and Jacob Steinhardt},
    journal={arXiv preprint arXiv:2103.03874},
    year={2021}
}

Source Data

The problems originate from high school mathematics competitions, including competitions like the AMC 10, AMC 12, and AIME. These represent carefully curated, high-quality mathematical problems that test conceptual understanding and problem-solving abilities rather than just computational skills.

Annotations

Each problem includes:

Complete problem text in LaTeX format
Detailed solution steps
Final answer in a standardized format
Subject category
Difficulty level (1-5)

Papers and References

For detailed information about the dataset and its evaluation, refer to "Measuring Mathematical Problem Solving With the MATH Dataset" presented at NeurIPS 2021.

https://arxiv.org/pdf/2103.03874