---
inference: false
language:
- zh
- en
license: unknown
model_name: Rain-2x7B-MoE-32k-v0.2
pipeline_tag: text-generation
prompt_template: '<s> SYS_PROMPT [INST] QUERY1 [/INST] RESPONSE1 [INST] QUERY2 [/INST]'
tags:
- nlp
- chinese
- mistral
- mixtral
- traditional_chinese
- merge
- mergekit
- MediaTek-Research/Breeze-7B-Instruct-v0_1
- beowolx/CodeNinja-1.0-OpenChat-7B
- mlabonne/Marcoro14-7B-slerp
---

<br/>

# 小雨同學 2x7B

採用聯發科 Breeze 7B Instruct 為基底的國語 MoE (Mixture-of-Experts) 模型，共有兩個 Expert model。

請用 Marcoro14-7B 或是 Breeze-7B-Instruct 所推薦的 Prompt 格式進行操作；以下為模型配置。

- v0.2 更新了 tokenizer parameters

![](https://i.imgur.com/f3Ro6Fu.png)

### Rain-2x7B-MoE-32k-v0.2

This is an experimental Mixtral-architecture MoE model with 2 of 7B sized fine-tunes. Breeze and CodeNinja are used on top of [Marcoro14-7B-slerp](https://huggingface.co/mlabonne/Marcoro14-7B-slerp).

Model configuration is as follows:

* [Marcoro14-7B-slerp](https://huggingface.co/mlabonne/Marcoro14-7B-slerp) as base.
* [Breeze-7B-Instruct-v0_1](https://huggingface.co/MediaTek-Research/Breeze-7B-Instruct-v0_1) as model 0.
* [CodeNinja-1.0-OpenChat-7B](https://huggingface.co/beowolx/CodeNinja-1.0-OpenChat-7B) as model 1.

To use the model, please use either prompt templates suggested by the base models.

## Notes

Please evaluate before use in any application pipeline. Activation for coding part of the model would be `'code'`, `'python'`, `'typescript'`, `'javascript'`, `'programming'`, `'algorithm'`.