JaMARD is a high-quality synthetic dataset for Japanese mathematical problems with assured correctness. It was created by translating and adapting English educational resources like PRM800K and GSM8K. The dataset, authored by elyza, was last updated on March 10, 2025.
Use Cases
- Fine-tuning language models for Japanese mathematical problem-solving based on the described chain-of-thought reasoning.
- Benchmarking model performance on Japanese-language math tasks based on the assured correctness of synthetic instances.
- Training models to generate step-by-step reasoning in Japanese based on the described educational problem format.
- Studying cross-lingual transfer of reasoning skills from English to Japanese based on the translation methodology.
Strengths
- Dataset quality is described as 'high-quality' with 'assured correctness' for synthetic instances.
- Based on established English educational resources PRM800K and GSM8K as a seed.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count, file formats, and license are unknown, which may limit suitability assessment.
Provenance
- Source
- huggingface
- Collection Method
- Synthetic dataset created by translating and adapting English educational resources.
- Time Range
- null
- Freshness
- Last updated 2025-03-10 04:57:32; freshness should be verified.
- Geography
- null