🧠 WeClone Deployment Note¶

Project Overview

This note records my complete deployment journey of WeClone, a local fine-tuning tool that turns chat history into a personalized language model, and its integration with AstrBot.

🚀 Environment Setup¶

✅ WSL2 Installation and Linux Environment¶

Installed WSL2 with Ubuntu.
Installed Git, Python 3.10+, pip, and uv as a fast dependency manager.

Virtual Environment with uv

Use uv venv .venv to create a virtual environment and source .venv/bin/activate to activate it. This avoids interfering with system Python and supports pyproject.toml based dependency management.

✅ WeClone Project Setup¶

GitHub Repo: WeClone
Official Guide: weclone.love

# Clone project
cd ~
git clone https://github.com/xming521/WeClone.git
cd WeClone

# Create and activate venv
uv venv .venv
source .venv/bin/activate

# Install dependencies
uv pip install --group main -e .

flash-attn optional dependency

FlashAttention requires CUDA and nvcc. In WSL2, you must install the full CUDA toolkit and set CUDA_HOME before installing: bash uv pip install flash-attn --no-build-isolation Skip this step if you're not training large models or running on CPU.

📦 Model Download via ModelScope¶

uv pip install modelscope
modelscope download --model Qwen/Qwen2.5-7B-Instruct --local_dir ./models/Qwen2.5-7B-Instruct

What is ModelScope?

ModelScope is a model hosting service by Alibaba DAMO Academy. It provides an easy CLI to download open-source foundation models. Model page

🗂️ Data Preprocessing¶

Export WeChat chat history using PyWxDump.
Place .csv in ./data/csv/ folder.
Preprocess using:

1	`weclone-cli make-dataset`

Automated startup shortcut

Add to .bashrc: bash alias start-weclone='source ~/WeClone/.venv/bin/activate && weclone-cli server'

🧪 Model Finetuning & Testing¶

# Fine-tuning (Single GPU)
weclone-cli train-sft

# Test model response
weclone-cli test-model

🌐 API & Interface¶

1 2	`# Start REST API service weclone-cli server`

Default listening on: http://127.0.0.1:8005/v1

WSL Networking Tip

If WeClone runs in WSL2 and AstrBot runs in Docker, you must connect via WSL's internal IP (e.g. 172.24.x.x) rather than localhost. Use ip addr show inside WSL to find this IP.

🤖 AstrBot Integration¶

GitHub: AstrBot
Docs: docs.astrbot.app

🌐 Model Provider Setup¶

If WeClone is hosted in WSL2:
Set API Base: http://172.24.x.x:8005/v1
Ensure WeClone server is running and accessible

💬 Messaging Platform Tips¶

Feishu: Bot name in AstrBot must exactly match the name registered in Feishu platform.
WeChat Personal: Use wechatpadpro with Docker, and ensure proper config for admin_key, host, and port. Docker network should allow communication between containers.

🧰 Other Tools & Tech Used¶

🐳 Docker Compose¶

Used to deploy supporting services like wechatpadpro:

Web docs often at: http://localhost:38849/docs

📦 Python Modern Packaging¶

uv + pyproject.toml for dependency management
Supports multiple dependency groups (main/dev/test)

🔬 LLaMA Factory¶

WeClone uses LLaMA-Factory under the hood to support LLM training and inference.

✅ Summary¶

This setup combines LLM training, API deployment, chat integration, and WSL networking. I now understand how to:

Export and clean real-world chat data
Run models locally with REST API
Connect chatbots via Docker
Build with modern Python toolchains (uv + pyproject.toml)

What’s Next?

I plan to try: - Multi-GPU or LoRA fine-tuning - Deploying the WeClone API via cloud (e.g. HuggingFace, EC2) - Integrating more platforms via AstrBot (Telegram, Discord, etc)

¶

# 用到的知识点
使用WeClone的模型根据wechat chat history 训练模型
    -github repo:https://github.com/xming521/WeClone
    -文档指南 https://www.weclone.love/what-is-weclone.html
    - 常用指令 ：
        1.数据预处理 weclone-cli make-dataset
        2.模型微调（单卡训练） weclone-cli train-sft
        3. 使用浏览器Demo简单退推理 weclone-cli webchat-demo； 使用API接口 weclone-cli server，接口监听在http://127.0.0.1:8005/v1
        4.测试模型 ：weclone-cli test-model

window 创建WSL2子系统-linux

PyWxDump
    -基本简介，用于导出微信聊天记录, 提供可视化的工具使用
    -github https://github.com/xaoyaoo/PyWxDump/
    -下载 https://github.com/xaoyaoo/PyWxDump/releases

linux下自动运行设定好的脚本
    。bashrc中写 alias start-weclone='source ~/WeClone/.venv/bin/activate && weclone-cli server'
    然后直接start-weclone

从ModelScope下载模型,例如Qwen2.5-7B-Instruct
    - 可以使用uv pip install modelscope
    -ModelScope 是什么，如何使用，link   https://modelscope.cn/models/Qwen/Qwen2.5-VL-32B-Instruct
    - modelscope download --model Qwen/Qwen2.5-7B-Instruct --local_dir ./models/Qwen2.5-7B-Instruct

AstrBot的部署
    - 文档：https://docs.astrbot.app/what-is-astrbot.html
    - github https://github.com/AstrBotDevs/AstrBot
    - 注意使用astrbot设置大模型的时候有如下问题
        1.服务提供商管理：连接大模型的API base 填什么？
            如果都是在同一主机可以填localhost，但是如果astrbot使用docker部署，大模型在wsl内，需要填wsl的ip，使用ip addr show查看
        2.消息平台，成功连接了飞书和微信个人：注意飞书机器人名必须和飞书中创建的bot名字一致，否则无法调用; 使用wechatpadpro连接个人微信，注意填写admin_key和host,port，还是 需要确定部署的astrbot和wechatpadpro是什么环境 (docker?) 是否共享网络


部署到飞书

# 学到的技术
docker compose部署多个应用
-例如部署WechatpadPro
    - 文档页面：http://localhost:38849/docs


uv和pyproject.toml
    - 使用uv venv .venv在当前目录创建python虚拟环境
    - source .venv/bin/activate激活
    - pyproject.toml如何使用，好处有哪些


了解 LLaMA Factory 