This project is a local chatbot fine-tuned using Bilibili user comments as training data. It uses Qwen1.5-32B-Chat as the base language model and applies LoRA fine-tuning via Apple’s mlx-lm framework. For voice interactions, it leverages GPT-SoVITS for speech synthesis using a Paimon voice model. The repository includes tools for model fusion and quantization to optimize performance.