Goedel-LM/Goedel-Formalizer-V2-8B

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Jul 20, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

Goedel-Formalizer-V2-8B by Goedel-LM is an 8 billion parameter model designed for translating informal mathematical problems into formal Lean 4 statements. Unlike previous open-source formalizers, this model incorporates a 'think before generating' mechanism, significantly enhancing translation accuracy. It is specifically optimized for formalizing mathematical text for use in proof assistants like Lean 4.

Loading preview...

Goedel-Formalizer-V2-8B: Formalizing Math into Lean 4

Goedel-Formalizer-V2-8B is an 8 billion parameter model developed by Goedel-LM, specifically engineered to translate informal mathematical problem statements into formal Lean 4 code. This model is a core component of the Goedel-Prover-V2 project, where it aids in generating Lean 4 statements for automated theorem proving.

Key Capabilities & Differentiators

  • Enhanced Accuracy: A primary distinction of Goedel-Formalizer-V2-8B is its unique ability to "think before generating" Lean 4 statements. This pre-computation step significantly improves the accuracy of the translation from natural language to formal code.
  • Specialized for Lean 4: The model is purpose-built for the Lean 4 proof assistant, making it highly effective for users working within this formal verification ecosystem.
  • Performance Benchmarking: In internal evaluations using a 300-statement Omni-MATH dataset, the Goedel-Formalizer-V2-32B variant successfully translated 228 statements, outperforming the Kimina-formalizer-8B which translated 161 statements.

Good For

  • Automated Theorem Proving: Developers and researchers involved in projects requiring the formalization of mathematical text for proof assistants.
  • Lean 4 Development: Users who need to convert natural language mathematical problems into executable Lean 4 code.
  • Mathematical AI Research: Exploring advanced methods for bridging the gap between informal mathematical reasoning and formal verification.