WeClone

English

简体中文

English

简体中文

Appearance

Deploy to a Chatbot via AstrBot ​

AstrBot is a multi-platform LLM chatbot framework that can deploy your WeClone model to platforms like QQ, WeChat, and Telegram.

Deployment steps:

  1. Deploy AstrBot: Follow the official AstrBot documentation to install and configure AstrBot on your server or local machine.

  2. Start the WeClone API service: Ensure that your weclone-cli server is running and that AstrBot can access the service's address and port.

  3. Add a new service provider in AstrBot:

    • Type: OpenAI
    • API Base URL: Fill in the WeClone API service address according to the AstrBot deployment method (e.g., http://127.0.0.1:8005/v1).
    • Model: gpt-3.5-turbo
    • API Key: Just fill in a few random letters, do not leave it blank.

  4. Deploy a messaging platform in AstrBot: Configure AstrBot to connect to the chat platform you want to use (e.g., WeChat, QQ, etc.).

  5. Disable tool calls (Important): The fine-tuned model is mainly used to imitate your language style and usually does not support complex tool calls. In the corresponding chat platform in AstrBot, send a command to your bot to turn off all default tools to ensure you can see the fine-tuning effect, or manually turn off the tools in the AstrBot WebUI:

    /tool off_all

  6. Set the system prompt: In the AstrBot configuration, set the System Prompt for your bot. This prompt must be exactly the same as the default_system you set in settings.jsonc when you fine-tuned the model.

  7. Adjust sampling parameters (Optional): You can adjust the model's sampling parameters in the infer_args of the configuration file. You can also adjust them in AstrBot, such as temperature, top_p, top_k, etc. For specific configuration methods, please refer to the section on configuring custom model parameters in the AstrBot documentation.

TIP

Frequently check the log output of api_service.py to ensure that the prompt sent by AstrBot to the large model service is consistent with what you expected during fine-tuning and testing.

Now, your exclusive digital avatar should be successfully deployed to the chatbot platform! Go and chat with "it" to see how it works~