---
base_model:
- meta-llama/Llama-3.1-70B
---

# NOT SUITABLE FOR CHAT INFERENCE AS-IS

llama3.1-base

Base 3.1 weights with transplanted llama 3.3 special token embeddings.

Unusable for chat in base state, intended to enable easy instruct qlora tuning of llama 3.1 base. 

No need to target embeddings or lm_head.