Datasets:

rasbt
/

math_full_minus_math500

license: apache-2.0
task_categories:
  - text-generation
  - question-answering
language:
  - en
dataset_info:
  features:
    - name: problem
      dtype: string
    - name: level
      dtype: string
    - name: type
      dtype: string
    - name: solution
      dtype: string
    - name: answer
      dtype: string
    - name: unique_id
      dtype: int64
  splits:
    - name: train
      num_bytes: 9548394
      num_examples: 12000
  download_size: 4670838
  dataset_size: 9548394
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

MATH (minus MATH-500)

This dataset is derived from the original MATH dataset by Hendrycks et al. (qwedsacf/competition_math) with all problems from the MATH-500 benchmark set removed.

Construction

Source: 12,500 problems from the MATH dataset by Hendrycks et al. (qwedsacf/competition_math)
Benchmark held out: 500 problems from the MATH-500 dataset (HuggingFaceH4/MATH-500)
Matching criterion: exact match on the problem field (see https://github.com/rasbt/math_full_minus_math500 for code to prepare the dataset)
Remaining size: 12,000 problems
Additionally, an "answer" field was added to the entries in the MATH dataset that contains the short answer similar to MATH-500

Fields

Each example contains:

problem: math problem statement
solution: full worked solution
answer: extracted final answer using extract_final_candidate function from the reasoning-from-scratch Python package (matches those in the MATH-500 dataset)
subject: math subject
level: difficulty level
unique_id: original problem identifier

Intended use

This dataset is intended for training only. Evaluation should be performed on MATH-500, which is excluded.

Usage

from datasets import load_dataset

dataset = load_dataset(
    "rasbt/math_full_minus_math500",
    split="train"
)

print(len(dataset))
print(dataset[0].keys())