InfinityMATH: Scalable Programmatic Math Instruction Tuning Dataset

InfinityMATH is a scalable instruction tuning dataset for programmatic mathematical reasoning. The dataset was created by BAAI and was last updated on September 3, 2024. Its construction pipeline emphasizes decoupling numbers from problems to synthesize number-independent programs.

Use Cases

• Fine-tuning language models for mathematical reasoning based on instruction tuning data.
• Training code models to generate number-independent programs based on the described synthesis method.
• Benchmarking model performance on scalable mathematical problem-solving based on programmatic reasoning.

Strengths

• Construction pipeline emphasizes decoupling numbers from problems, enabling flexible scaling.
• Fine-tuning experiments demonstrate practical benefits with models like Llama2 and CodeLlama.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

BAAI

Collection Method

Synthesized via a pipeline emphasizing decoupling numbers from mathematical problems.

Freshness

Last updated 2024-09-03 07:24:26; freshness should be verified.