mtybilly/MetaMathQA

---
license: mit
task_categories:
- text-generation
- question-answering
language:
- en
tags:
- math
- reasoning
- metamath
source_datasets:
- meta-math/MetaMathQA
configs:
- config_name: full
data_files:
- split: train
path: full/train-*.parquet
- config_name: MATH
data_files:
- split: train
path: MATH/train-*.parquet
- config_name: MATH-50K
data_files:
- split: train
path: MATH-50K/train-*.parquet
default_config_name: full
---

# MetaMathQA Subsets

Curated subsets of [meta-math/MetaMathQA](https://huggingface.co/datasets/meta-math/MetaMathQA) for mathematical reasoning experiments.

## Subsets

| Subset | Samples | Description |
|--------|---------|-------------|
| `full` | 395,000 | All MetaMathQA samples (unchanged) |
| `MATH` | 155,000 | MATH_* types only (AnsAug, Rephrased, FOBAR, SV) |
| `MATH-50K` | 50,000 | Stratified 50K sample from MATH subset |

### MATH-50K Type Distribution

| Type | Count | Proportion |
|------|-------|------------|
| MATH_AnsAug | 24,194 | 48.4% |
| MATH_Rephrased | 16,129 | 32.3% |
| MATH_FOBAR | 4,839 | 9.7% |
| MATH_SV | 4,838 | 9.7% |

## Columns

| Column | Description |
|--------|-------------|
| `id` | Unique identifier in format `<type>_<global_idx>` |
| `query` | The question text |
| `response` | Chain-of-thought solution ending with "The answer is: ..." |
| `type` | Question type (e.g., MATH_AnsAug, GSM_Rephrased) |
| `original_question` | Original question from the source dataset |

## Usage

```python
from datasets import load_dataset

# Load the MATH-50K subset
ds = load_dataset("mtybilly/MetaMathQA", "MATH-50K", split="train")

# Load full dataset
ds = load_dataset("mtybilly/MetaMathQA", "full", split="train")
```

## Source

Derived from [MetaMathQA](https://huggingface.co/datasets/meta-math/MetaMathQA) (Yu et al., 2024). Stratified sampling uses seed=42 for reproducibility.