https://github.com/zyayoung/wechatpersona
LLM agents that mimic your friends.
https://github.com/zyayoung/wechatpersona
Last synced: 2 days ago
JSON representation
LLM agents that mimic your friends.
- Host: GitHub
- URL: https://github.com/zyayoung/wechatpersona
- Owner: zyayoung
- Created: 2024-02-08T08:23:00.000Z (over 1 year ago)
- Default Branch: master
- Last Pushed: 2024-02-08T08:52:31.000Z (over 1 year ago)
- Last Synced: 2025-03-15T02:22:32.522Z (3 months ago)
- Language: Python
- Homepage:
- Size: 84 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: Readme.md
Awesome Lists containing this project
README
# WeChatPersona 🤖
Introducing WeChatPersona, virtual agents to replicate the conversational style of your real friends on WeChat. By leveraging the powerful Baichuan2-7B-Chat model and fine-tuning it with your WeChat history, WeChatPersona delivers personalized virtual agents that emulate the language patterns and style of your friends.
## Demo
Witness the capabilities of WeChatPersona in action with our demonstration (trained on 4090*8 for 10 hours):
Prompt:
```
USER:生成与小明的微信聊天记录
ASSISTANT:
我：a6000好慢啊
我：oh好像不慢 是nfs慢
```
Generated:
```
对方：[图片]
对方：这个怎么是英文的呀，我应该下载哪个?
我：oh
对方：哦哦我：就下第一个吧 好像都是中文的
我：<哈哈>
对方：okk
我：oh对
对方：
我：<紧张>
我：你咋了
对方：没事，就是问问
我：<问号>
对方：<哼歌>
我：<哈哈哈>
我：<爱你>
```
Note: Pre-trained weights are not provided for privacy reasons
## Usage 🛠️
Unlock the full potential of WeChatPersona with these simple steps:
1. **Install Dependencies:** Begin by installing the necessary dependencies using pip:
```bash
pip install transformers deepspeed
```
2. **Prepare Training Data:** Export your [WeChatMsg](https://github.com/LC044/WeChatMsg) message history to a CSV file using WeChatMsg. Optionally, you can utilize `StoreEmotion.db` for additional emotion descriptions. Set configs in `prepare_data.py`. Then run
```bash
python src/prepare_data.py
```
3. **Run LoRA Finetuning:** Initiate the LoRA fine-tuning process with the provided script:
```bash
bash src/ft.sh
```
4. **Run the demo:** Inference using fine-tuned model:
```bash
python src/demo.py
```
## Acknowledgements
- [WeChatMsg](https://github.com/LC044/WeChatMsg)
- [Baichuan2-7B-Chat](https://huggingface.co/baichuan-inc/Baichuan2-7B-Chat)