---
thumbnail: https://github.com/rinnakk/japanese-pretrained-models/blob/master/rinna.png
language:
- ja
- en
tags:
- qwen
inference: false
---

# `rinna/nekomata-14b-instruction-gguf`

![rinna-icon](./rinna.png)

# Overview
The model is the GGUF version of [`rinna/nekomata-14b-instruction`](https://huggingface.co./rinna/nekomata-14b-instruction). It can be used with [llama.cpp](https://github.com/ggerganov/llama.cpp) for lightweight inference.

Quantization of this model may cause stability issue in GPTQ, AWQ and GGUF q4_0. We recommend **GGUF q4_K_M** for 4-bit quantization.

See [`rinna/nekomata-14b-instruction`](https://huggingface.co./rinna/nekomata-14b-instruction) for details about model architecture and data.

* **Authors**

    - [Toshiaki Wakatsuki](https://huggingface.co./t-w)
    - [Tianyu Zhao](https://huggingface.co./tianyuz)
    - [Kei Sawada](https://huggingface.co./keisawada)
    
---

# How to use the model

See [llama.cpp](https://github.com/ggerganov/llama.cpp) for more usage details.

~~~~bash
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
make

MODEL_PATH=/path/to/nekomata-14b-instruction-gguf/nekomata-14b-instruction.Q4_K_M.gguf
MAX_N_TOKENS=512
PROMPT_INSTRUCTION="次の日本語を英語に翻訳してください。"
PROMPT_INPUT="大規模言語モデル（だいきぼげんごモデル、英: large language model、LLM）は、多数のパラメータ（数千万から数十億）を持つ人工ニューラルネットワークで構成されるコンピュータ言語モデルで、膨大なラベルなしテキストを使用して自己教師あり学習または半教師あり学習によって訓練が行われる。"
PROMPT="以下は、タスクを説明する指示と、文脈のある入力の組み合わせです。要求を適切に満たす応答を書きなさい。\n\n### 指示:\n${PROMPT_INSTRUCTION}\n\n### 入力:\n${PROMPT_INPUT}\n\n### 応答:\n"

./main -m ${MODEL_PATH} -n ${MAX_N_TOKENS} -p "${PROMPT}"
~~~~

---

# Tokenization
Please refer to [`rinna/nekomata-14b`](https://huggingface.co./rinna/nekomata-14b) for tokenization details.

---

# How to cite
~~~
@misc{RinnaNekomata14bInstructionGGUF, 
    url={https://huggingface.co./rinna/nekomata-14b-instruction-gguf}, 
    title={rinna/nekomata-14b-instruction-gguf}, 
    author={Wakatsuki, Toshiaki and Zhao, Tianyu and Sawada, Kei}
}
~~~
---

# License
[Tongyi Qianwen LICENSE AGREEMENT](https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT)