---
base_model:
- AI-MO/Kimina-Autoformalizer-7B
---

A GRPO version on RLMEval, with embedding similarity and type-check accuracy as rewards, both with weight 1.0, and a KL divergence beta of 0.04.