---
base_model:
- google/gemma-3-12b-pt
---

This is the same datamix as [Glitter](https://huggingface.co/allura-org/Gemma-3-Glitter-12B), but trained on the base model (gemma-3-12b-pt) instead of instruct. Gemma 3 instruct format was used for the instruct portions.