---
datasets:
- IlyaGusev/saiga_scored
- IlyaGusev/saiga_preferences
language:
- ru
inference: false
license: gemma
---

Llama.cpp compatible versions of an original [12B model](https://huggingface.co/IlyaGusev/saiga_gemma3_12b).

Download one of the versions, for example `saiga_gemma3_12b.Q4_K_M.gguf`.
```
wget https://huggingface.co/IlyaGusev/saiga_gemma3_12b_gguf/resolve/main/saiga_gemma3_12b.Q4_K_M.gguf
```

Download [interact_gguf.py](https://raw.githubusercontent.com/IlyaGusev/saiga/refs/heads/main/scripts/interact_gguf.py)
```
https://raw.githubusercontent.com/IlyaGusev/saiga/refs/heads/main/scripts/interact_gguf.py
```

How to run:
```
pip install llama-cpp-python fire

python3 interact_gguf.py saiga_gemma3_12b.Q4_K_M.gguf
```

System requirements:
* 13GB RAM for q8_0 and less for smaller quantizations