xming521/WeClone

What it solves

WeClone provides an end-to-end pipeline for creating a digital avatar that mimics a specific person's communication style. It solves the difficulty of transforming raw chat history from messaging platforms into a fine-tuned Large Language Model (LLM) that captures a user's unique "flavor" and personality.

How it works

The system follows a multi-stage process:

1. Data Export & Preprocessing: It imports chat records (currently supporting Telegram) and uses tools like Microsoft Presidio to filter out sensitive private information (emails, phone numbers, etc.).
2. Profile Building: It can automatically generate a user profile based on the chat history to better define the avatar's persona.
3. Fine-tuning: It uses SFT (Supervised Fine-Tuning) with methods like LoRA or QLoRA (via LLaMA Factory) to train a model—defaulting to Qwen2.5-VL-7B-Instruct—on the processed chat data. It supports both text and image modal data.
4. Deployment: The fine-tuned model is served via an API, which can then be integrated into chatbot frameworks like AstrBot or LangBot to deploy the avatar on platforms such as Telegram, Discord, or Slack.

Who it’s for

• Individuals wanting to create a digital twin of themselves or others based on real conversation history.
• Developers looking for a complete workflow to move from raw chat logs to a deployed AI persona.

Highlights

• End-to-End Workflow: Covers everything from data export and privacy filtering to training and deployment.
• Multimodal Support: Ability to fine-tune using both text and image data.
• Privacy Focused: Includes built-in PII filtering and optional data encryption for stored datasets.
• Flexible Deployment: Compatible with multiple bot frameworks for easy integration into various messaging apps.