README.md

metadata

language:
  - en
tags:
  - math
  - mathematics
  - reasoning
  - arithmetic
  - algebra
  - word-problems
  - logic
  - chain-of-thought
  - step-by-step
  - education
  - synthetic
  - llm
  - llm-training
  - pretraining
  - finetuning
  - Math
  - Logic
  - Algebra
pretty_name: MathReasoning-10000
size_categories:
  - 10K<n<100K
license: mit
dataset_info:
  features:
    - name: question
      dtype: string
    - name: problem
      dtype: string
    - name: how_to_solve
      dtype: string
    - name: answer
      dtype: string
  splits:
    - name: train
      num_bytes: 772743
      num_examples: 10000
  download_size: 148059
  dataset_size: 772743
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

MathReasoning-10000

A synthetic math reasoning dataset containing 10000 examples.

Each row follows a strict reasoning-friendly structure:

• question
• problem
• how_to_solve
• answer

This format is designed for:

- LLM reasoning

- chain-of-thought style supervision

- math finetuning

- educational tasks

---

📐 Problem Types

- Addition

- Subtraction

- Multiplication

- Division (integer-safe)

---

🧪 Recommended Splits

- Train: 80%

- Test: 20%

---

Generated fully in-memory. No external data sources.