metadata

language:
  - ru
tags:
  - llama
license: apache-2.0

Model Card for Mira 70B v0.8

The smartest Russian model based on LLaMA 70B v2 architecture. So maybe it's the last Big Russian Model of that family.

Trained with TRINITY method for 60 hours of Nvidia H100.

Please use the prompt formatted like this:

You are an AI assistant that follows instruction. Help as much as you can.

### Human:

Your question here.

### Assistant:

Please pay attention for double new lines everywhere - and after "### Assistant:" line too!