---
language:
- ja
---

[google/gemma-3-12b-it-qat-q4_0-unquantized](https://huggingface.co/google/gemma-3-12b-it-qat-q4_0-unquantized)を[日本語が多く含まれるimatrixを使って量子化](https://huggingface.co/dahara1/imatrix-jpn-test)したモデルです  
This is a model that quantizes [google/gemma-3-12b-it-qat-q4_0-unquantized](https://huggingface.co/google/gemma-3-12b-it-qat-q4_0-unquantized) using an [imatrix that contains a lot of Japanese](https://huggingface.co/dahara1/imatrix-jpn-test)..  
https://huggingface.co/dahara1/imatrix-jpn-test).  

最新の[llama.cpp](https://github.com/ggml-org/llama.cpp)を使って動かしてください。  
Please use the latest [llama.cpp](https://github.com/ggml-org/llama.cpp).   

llama-mtmd-cliコマンドとmmproj.ggufファイルを使うと画像を読みこむ事ができます  
You can use llama-mtmd-cli for image reading.  
```
llama-mtmd-cli -m gemma-3-4b-it-qat-q4_0-japanese-imatrix-Q4_K_L.gguf --mmproj mmproj.gguf --image ./test.png -p "この画像はなんですか？(What is this image?)"
```

## auto download command example.
```
llama-server.exe -hf dahara1/gemma-3-12b-it-qat-japanese-imatrix:gemma-3-12b-it-qat-q4_0-japanese-imatrix-Q4_0.gguf
```
then access to http://127.0.0.1:8080