bilibot  by linyiLYi

Local chatbot fine-tuned with user comments

created 1 year ago
3,183 stars

Top 15.5% on sourcepulse

GitHubView on GitHub
Project Summary

This project provides a local chatbot fine-tuned on Bilibili user comments, offering text-based conversations and text-to-speech capabilities using a specific voice model. It's designed for users interested in creating custom, localized AI companions with a unique conversational style derived from a popular Chinese video platform.

How It Works

The core of the chatbot utilizes the Qwen1.5-32B-Chat model, fine-tuned using the MLX-LM LoRA example project on Apple's MLX framework. This approach allows for efficient fine-tuning on Apple Silicon hardware. For speech generation, it integrates the GPT-SoVITS open-source project, specifically using a Paimon voice model trained by a Bilibili user.

Quick Start & Requirements

  • Installation: Requires Python 3.10 and Anaconda for environment management.
  • Prerequisites: macOS is the tested platform. Dependencies are listed in requirements.txt.
  • Setup: Involves creating a Conda environment (bilibot), activating it, and installing requirements.
  • Resources: Fine-tuning and inference are performed using MLX, optimized for Apple Silicon.
  • Documentation: Links to MLX, Qwen1.5, and GPT-SoVITS are provided.

Highlighted Details

  • Fine-tuning is performed using LoRA on the Qwen1.5-32B-Chat model.
  • Model quantization is supported via tools/compress_model.py for faster inference.
  • Speech generation leverages GPT-SoVITS with a Paimon voice model.
  • A start_qa_dialogue.py script enables voice dialogue generation from text questions.

Maintenance & Community

The project references external projects like MLX, Qwen1.5, and GPT-SoVITS, indicating reliance on their development and community support. Specific maintainer or community links for this project are not detailed in the README.

Licensing & Compatibility

The README does not explicitly state the license for this project. It relies on external projects with their own licenses. Commercial use compatibility would depend on the licenses of Qwen1.5, MLX, and GPT-SoVITS.

Limitations & Caveats

The project is tested on macOS and may not be directly compatible with other operating systems without modifications. The setup for GPT-SoVITS requires separate environment configuration and execution as per its official guidelines.

Health Check
Last commit

1 year ago

Responsiveness

1 day

Pull Requests (30d)
0
Issues (30d)
0
Star History
12 stars in the last 90 days

Explore Similar Projects

Feedback? Help us improve.