huanhuan-chat

Why we included this project

This project documents a complete LoRA fine-tuning run, using a fun test case: a chatbot that talks like Zhen Huan from the Chinese drama Empresses in the Palace. The guide covers the whole pipeline, from setting up the environment to turning raw script text into instruction/response pairs, isolating one character's lines, and running the training script. It also handles the messy case where your source is unstructured prose, showing how to use a large model to extract character dialogue and pointing to a companion tool for that step. The result is a working character chatbot, and the same workflow applies to any novel or screenplay, so it is a solid starting point if you want to build a character-driven model without designing the training pipeline from scratch.