Python Competitive Programming Problems with Unit Tests

A synthetic dataset of competitive coding-style Python problems and their corresponding unit test cases. The dataset was created by NVIDIA for its NeMo Gym reinforcement learning framework and was last updated in January 2026. It aggregates questions and tests from the CodeContests and Open-R1 collections.

Use Cases

• Train reinforcement learning agents to generate correct code solutions by evaluating against provided unit tests.
• Benchmark the reasoning capabilities of language models on competitive programming problem statements.
• Fine-tune code generation models using the paired problem descriptions and test cases.
• Analyze the diversity and difficulty of problems sourced from platforms like Codeforces.

Strengths

• Data is sourced from established collections like DeepMind's CodeContests and Open-R1's Codeforces problems.
• Includes unit test cases for validating code solutions, providing a clear performance metric.
• Dataset is specifically curated for building reinforcement learning environments within the NVIDIA NeMo Gym framework.

Limitations

• Dataset size, row count, and specific column structure are not provided.
• Limited to Python programming language problems, excluding other languages.
• Synthetic nature may not fully represent the distribution of real-world coding challenges.

Provenance

Source

Aggregated from DeepMind's CodeContests and Open-R1's Codeforces collections.

Collection Method

Collected and synthesized for the NVIDIA NeMo Gym project.

Time Range

null

Freshness

Last updated on 2026-01-12.

Geography

null