Lean Sft Dataset: Merged Theorem-Proving Exercises and Proofs

A merged dataset for theorem proving combines problem statements from internlm/Lean-Workbook with full proofs from Goedel-LM/Lean-workbook-proofs. It also includes a human-annotated subset from AI-MO/NuminaMath-LEAN. The dataset was last updated on February 11, 2026.

Use Cases

• Training language models for formal mathematics based on annotated Lean code.
• Benchmarking automated theorem-proving systems based on structured problem statements.
• Fine-tuning models for proof generation based on the provided full proofs.
• Studying human-annotated mathematical reasoning based on the NuminaMath-LEAN subset.

Strengths

• Combines three distinct sources: problem statements, full proofs, and human annotations.
• Last updated on February 11, 2026, suggesting recent maintenance.

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

Merged from internlm/Lean-Workbook, Goedel-LM/Lean-workbook-proofs, and AI-MO/NuminaMath-LEAN.

Collection Method

Merging of existing datasets from the Hugging Face platform.

Freshness

Last updated 2026-02-11 13:55:20