OProofs: 6.8 Million Formal Theorem-Proof Pairs in Lean 4

Formal Lean 4 theorem-proof pairs produced as part of the OProver project. The dataset contains 6,804,694 records across 73 parquet shards, uploaded by author m-a-p. It was last updated on the Hugging Face platform in May 2026.

Use Cases

• Training language models for formal mathematics based on the structured theorem-proof pairs.
• Developing automated proof assistants by learning from the formal_statement and formal_proof fields.
• Studying chain-of-thought reasoning in theorem proving using the optional cot_proof field.
• Analyzing the structure of machine-generated mathematical proofs based on the provided prompt field.

Strengths

• 6,804,694 records provide a substantial corpus for machine learning.
• Data is structured with specific fields for statements, proofs, and reasoning.
• Records are stored in 73 parquet shards compressed with zstd for efficient handling.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last updated 2026-05-19 08:07:13; freshness should be verified.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Hugging Face dataset uploaded by author m-a-p as part of the OProver project.

Collection Method

Formal Lean 4 theorem-proof pairs produced as part of the OProver project.

Freshness

Last updated 2026-05-19 08:07:13.