StepFun Formalizer 32B
StepFun Formalizer 32B is the larger model in StepFun's autoformalization release. It builds on DeepSeek-R1-Distill-Qwen-32B and targets natural-language-to-Lean 4 conversion.
Research release
Combine formal knowledge with reasoning
StepFun describes training that combines knowledge of formal mathematics with reasoning about the mapping from informal problems. Its evaluation uses checks of statement equivalence, rather than treating any compilable declaration as correct.StepFun StepFun-Formalizer-32B model card
Choose by statement fidelity
Practical guidance. Compare the 7B and 32B releases on your notation and subject area. Count semantic mistakes separately from syntax errors. A larger model is useful only if the improvement justifies the additional memory and inference cost for that workload.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | StepFun-Formalizer-32B |
| Base model | DeepSeek-R1-Distill-Qwen-32B.StepFun StepFun-Formalizer-32B model card |
| License | Apache 2.0 for code and weights.StepFun StepFun-Formalizer-32B model card |
| Published materials | The card links training data, research code and both Formalizer sizes.StepFun StepFun-Formalizer-32B model card |
Related models
- StepFun Formalizer 7B
StepFun · StepFun Formalizer
StepFun Formalizer 7B converts natural-language mathematical problems into formal Lean 4 statements.
Launch date not confirmedResearch releaseDownloadable - StepFun Prover Preview 32B
StepFun · StepFun Prover
A DeepSeek-R1-Distill-Qwen-32B-based model for iterative Lean 4 proof search.
Launch date not confirmedResearch releaseDownloadable