MathNet:30k competition math problems for AI mathematical reasoning benchmarking
11 hours ago
- #mathematical reasoning
- #multimodal dataset
- #Olympiad problems
- MathNet is a high-quality, large-scale, multimodal, multilingual dataset of Olympiad-level math problems with solutions.
- It spans 47 countries, 17 languages, and two decades, containing 30,676 expert-authored problems.
- The dataset supports three tasks: problem solving, math-aware retrieval, and retrieval-augmented problem solving.
- Even state-of-the-art models struggle on these tasks, with retrieval being a significant bottleneck.
- MathNet provides a benchmark for evaluating mathematical reasoning in generative models and retrieval in embedding-based systems.
- The dataset is publicly released, with human expert verification ensuring quality.