tharustack/open-math-dataset

Dataset Description

Open Math Dataset is an open mathematics corpus designed for mathematical AI, reasoning, education, and research.

The project is being developed from Sri Lanka with the goal of creating an internationally useful mathematics dataset for developers, researchers, educators, and AI systems.

Mathematics Corpus

The dataset is designed to contain structured mathematical problems and solutions across different mathematical domains and education levels.

Planned content includes:

• —Arithmetic
• —Algebra
• —Geometry
• —Trigonometry
• —Calculus
• —Statistics
• —Probability
• —Number theory
• —Linear algebra
• —Discrete mathematics
• —Mathematical reasoning
• —Word problems
• —Step-by-step solutions
• —Mathematical explanations
• —Error correction
• —Multiple solution methods
• —Mathematical proofs

The dataset supports both general mathematics and curriculum-specific mathematics.

Languages

Current languages:

• —English (en)
• —Sinhala (si)

Sinhala mathematics is an important initial focus because high-quality Sinhala mathematical training data is limited.

However, the project is not limited to Sri Lankan mathematics. The dataset is designed for international expansion and may include additional languages, countries, and curricula in future releases.

Sri Lankan Mathematics

Sri Lankan mathematical content is one of the initial areas of development. This may include:

• —Sri Lankan school mathematics
• —Sinhala mathematical terminology
• —O/L mathematics
• —A/L mathematics
• —Sri Lankan curriculum-related problems
• —Localized mathematical explanations

Sri Lankan content is treated as part of the wider Open Math Dataset rather than the entire scope of the project.

Dataset Structure

Data Fields

The dataset is intended to use structured records containing information such as:

Example

{
  "id": "omd_000001",
  "language": "en",
  "problem": "Solve 2x + 5 = 15.",
  "solution": "2x + 5 = 15\n2x = 10\nx = 5",
  "answer": "5",
  "topic": "algebra.linear_equations",
  "difficulty": 0.2,
  "education_level": "middle_school"
}

Data Quality

Mathematical correctness is a major priority.

The project aims to prioritize:

• —Correct answers
• —Correct mathematical reasoning
• —Clear explanations
• —Consistent formatting
• —Deduplication
• —Accurate metadata
• —Diverse difficulty levels
• —Diverse mathematical domains
• —Transparent data provenance

Automatically generated mathematical content should be treated as candidate data and verified before being considered high-quality training data.

Data Pipeline

The planned data preparation process is:

Collection
    ↓
Cleaning
    ↓
Normalization
    ↓
Generation
    ↓
Mathematical Verification
    ↓
Deduplication
    ↓
Quality Filtering
    ↓
Metadata
    ↓
Dataset Release

Intended Uses

The dataset may be useful for:

• —Mathematical AI
• —LLM training
• —LLM fine-tuning
• —Mathematical reasoning research
• —Educational AI
• —AI tutoring systems
• —Question generation
• —Mathematical evaluation
• —Multilingual AI
• —Curriculum-aware AI
• —Mathematical search systems

Data Provenance

The dataset may contain data from multiple sources and generation methods.

Possible source types include:

• —human_verified
• —official_source
• —public_domain
• —synthetic_verified
• —community_contributed
• —translated_verified

Source and licensing information should be preserved whenever possible.

Only data that can legally be redistributed should be included in public releases.

Dataset Versions

The dataset will use versioned releases.

Dataset statistics and coverage information will be updated with future releases.

Project Status

Active Development

Open Math Dataset is an ongoing project.

The initial focus is building a strong foundation of high-quality mathematical data before expanding into larger international and multilingual coverage.

Long-term Goal

The goal is to create an open mathematics dataset that developers and researchers around the world can use.

• —From basic arithmetic to advanced mathematics.
• —From simple question-and-answer pairs to structured mathematical reasoning.
• —From Sinhala and English to additional languages in the future.

Mathematics is universal. The data should be too.

Maintainer

Created and maintained by Tharustack.

Built from Sri Lanka 🇱🇰 for the world 🌍.