---
license: apache-2.0
datasets:
- ConicCat/AntiRepV0.2
language:
- en
base_model:
- swiss-ai/Apertus-8B-2509
---


<p align="left">
  <img width="60%" src="capertus.jpg">
</p>

TODO: improve model card

Trained from Apertus-8B Base using liger kernel with a two step Anchored SFT > LD-DPO recipe.

Alpaca template, no system.

.7 temp, top_p .95, no rep pen or dry