Eurus-2-RL-Data is a high-quality reinforcement learning training dataset for mathematics and coding problems. It includes outcome verifiers, such as LaTeX answers for math and test cases for coding. The dataset was created by PRIME-RL and was last updated on 2025-02-19.
Use Cases
- Training reinforcement learning agents for mathematical problem-solving based on problems sourced from NuminaMath-CoT.
- Fine-tuning code generation models based on coding problems from sources like APPS, CodeContests, TACO, and Codeforces.
- Developing outcome verification systems based on the included LaTeX answer and test case verifiers.
- Benchmarking model performance on problems ranging from high school mathematics to International Mathematical Olympiad competition questions.
Strengths
- Includes outcome verifiers for both mathematics (LaTeX answers) and coding (test cases).
- Sources problems from established datasets like NuminaMath-CoT, APPS, CodeContests, TACO, and Codeforces.
- Spans a range of difficulty from Chinese high school mathematics to International Mathematical Olympiad competition questions.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- PRIME-RL
- Collection Method
- Sourced from existing datasets: NuminaMath-CoT for mathematics; APPS, CodeContests, TACO, and Codeforces for coding.
- Freshness
- Last updated 2025-02-19 12:14:49; freshness should be verified.