Muice-Chatbot is an AI companion designed to proactively engage users in conversation with a distinct tsundere anime girl persona. It leverages a Qwen model fine-tuned on a 3k+ dialogue dataset, offering features like proactive chat initiation, image recognition for memes, and a Faiss-based memory module for contextual recall. The project is intended for users seeking an interactive AI friend with a specific personality.

How It Works

The core of Muice-Chatbot is a Qwen language model fine-tuned with a custom dataset to emulate an anime girl's personality. It utilizes a LoRA adapter for efficient fine-tuning. The system supports multiple model loaders, allowing flexibility with base models like Qwen and ChatGLM, and integrates a Faiss index for long-term memory retrieval, enhancing conversational context.

Quick Start & Requirements

• Install: pip install -r requirements.txt (or use install_env.bat for automated setup).
• Prerequisites: Python 3.10, GPU with 6GB+ VRAM (4GB for int4 quantization, 16GB+ RAM for CPU inference), CUDA 12.4 recommended for GPU.
• Setup: Automated environment setup available. Model download is manual.
• Docs: Muice-Chatbot

Highlighted Details

• Proactively initiates conversations at random or scheduled times.
• Supports OFA for image recognition (emojis, memes).
• Integrates Faiss for long-term memory retrieval.
• Offers real-time voice chat in the console.
• Supports OneBot V11 protocol for QQ integration.

Maintenance & Community

• Status: Project officially stopped updating on 2025.02.19 and will be archived on 2025.07.16. Migration to MuiceBot (based on Nonebot2) is recommended.
• Community: MuikaAI Official Channel

Licensing & Compatibility

• License: Source code is MIT licensed. Fine-tuned models are not recommended for commercial use.

Limitations & Caveats

The project is no longer actively maintained and will be archived soon. A warning exists regarding potential account bans when using QQNT with LiteLoader, recommending specific downgrade and configuration steps.