---
license: apache-2.0
datasets:
- mlabonne/CodeLlama-2-20k
language:
- zh
---
# Chinese-CodeLlama-7B-SFT-V1

We implemented SFT based on our [Chinese-CodeLlama-7B-PT](https://huggingface.co/frankminors123/Chinese-CodeLlama-7B-PT). The dataset comes from [CodeLlama-2-20k](https://huggingface.co/datasets/mlabonne/CodeLlama-2-20k), we used Google Translate to translate it into Chinese. 

In addition, we designed appropriate Chinese prompt template for coding tasks, and during the fine-tuning stage, `memory efficient attention` was applied which save us a lot of GPU memory space.

The Chinese prompt template used is as follows:
```python
PROMPT_TEMPLATE = (
  "下面是描述一项任务的指令，并且与一则输入配对用来提供更多的上下文。请给出尽可能满足请求的回答.\n"
  "### 指令:\n{instruction}\n### 输入:\n{input}\n### 回答:\n"
)
```