Kimina-Autoformalizer-7B: Natural Language to Lean 4 Autoformalization

Kimina-Autoformalizer-7B, developed by Project Numina, is a 7.6 billion parameter model engineered for the specific task of autoformalization. Its core function is to convert natural language descriptions of mathematical problems, particularly those found in competitive programming contexts, into Lean 4 code.

Key Capabilities

• Natural Language to Lean 4 Translation: Translates problem statements into formal Lean 4 code.
• Competition Problem Focus: Optimized for the structure and style of competition-level mathematical problems.
• Code Generation with by sorry: Generates Lean 4 code snippets that conclude with by sorry, indicating a formal statement ready for proof.

Use Cases

• Formal Mathematics Research: Aids mathematicians and researchers in formalizing conjectures and theorems in Lean 4.
• Educational Tools: Can be used to help students understand the process of formalizing mathematical statements.
• Automated Theorem Proving Pipelines: Serves as a front-end for systems that require formal input for automated theorem provers.

This model is particularly useful for developers and researchers working on formal verification, automated reasoning, or educational platforms that leverage Lean 4.