Competition Math · AI-MO
olympiads-ref
A structured dataset of competition mathematics problems and verified solutions for reasoning models.
What olympiads-ref is
AI-MO/olympiads-ref is a structured reference dataset comprising mathematics competition problems and their step-by-step solutions, organised by competition. Developed by AI-MO, the resource prioritises official solutions from prestigious contests, including the International Mathematical Olympiad (IMO).
The dataset is designed to support research in automated mathematical reasoning. Specifically, it provides training material for chain-of-thought prompting and the formalisation of mathematical proofs across subjects such as algebra, geometry, combinatorics, and number theory.
How it was made
Data curation begins by gathering raw competition problem sets and official solutions in PDF format. These documents are converted into Markdown using Mathpix via the Numina Math data pipeline.
The resulting Markdown files are segmented into individual problems and solutions using custom Python scripts and saved into JSON Lines (JSONL) files. Each entry stores problem statements in LaTeX or Markdown alongside their step-by-step solutions, optional answers, competition years, problem domains (such as Calculus or Logic and Puzzles), and question classifications (such as proofs or multiple-choice questions). Diagrams referenced via image placeholders are stored in an accompanying Parquet file.
Key facts
- Primary fields
- Mathematics, scientific reasoning
- Supported problem types
- Algebra, Geometry, Number Theory, Combinatorics, Calculus, Inequalities, Logic and Puzzles, Other
- Supported question types
- MCQ, proof, math-word-problem
- File formats
- PDF, Markdown (.md), JSONL (.jsonl), Parquet (.parquet)
- Modality
- Document
What you can use it for
- Chain-of-thought model training
- Supplying step-by-step mathematical reasoning paths to train language models on competition-level problem solving.
- Mathematical formalisation
- Providing structured problem and solution texts to assist in translating informal mathematical proofs into formal verification systems.
- Reasoning benchmark evaluation
- Assessing computational models across specific mathematical fields such as number theory, geometry, and combinatorics using categorized competition problems.
Explore the data
No licence declared
The authors have not published a licence, which means no permission to redistribute the data has been granted to anyone. The rows are not served here; read them on the original dataset page.Strengths and limits
Good at
- Prioritises official and verified solutions to ensure reliable step-by-step reasoning steps.
- Provides end-to-end data representations from raw PDFs and converted Markdown to structured JSONL records.
- Includes dedicated image management via Parquet to preserve associated figures and problem diagrams.
Watch out for
- Many metadata attributes, including difficulty ratings, source identifiers, and specific exam labels, are optional and may be incomplete.
- Data ingestion relies on individual segmentation scripts tailored to specific competition formats rather than a single universal parser.
Questions
Can I look inside olympiads-ref without downloading it?
Not here: it declares no licence, so no permission to redistribute its rows has been granted. Its description and statistics are still shown.
What licence is olympiads-ref under?
Its authors have not declared a licence, which means no permission to reuse or redistribute it has been granted.
What mathematical subfields are covered?
The dataset categorises problems into Algebra, Geometry, Number Theory, Combinatorics, Calculus, Inequalities, Logic and Puzzles, and Other.
How are problem diagrams and images stored?
Image placeholders in the text records correspond directly to image assets stored within an images.parquet file.
What types of questions are included?
Problems are classified into multiple-choice questions ('MCQ'), formal proof problems ('proof'), or problems requiring a calculated final output ('math-word-problem').
Written from olympiads-ref's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.