kiminaembedbeta / README.md
sorgfresser's picture
Create README.md
a973a26 verified
---
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.