firefly-roleplay

This project is for role-playing with the Firefly model.

中文文档 (Chinese Documentation) | English Documentation

Install

Please follow the Unsloth Installation Guide to set up the environment.

Usage

Dataset

The data can be downloaded from: Firefly-Roleplay Dataset

The project provides scripts to construct data for pre-training and fine-tuning.

• construct_continue_pretrain_data.py: Constructs data for continued pre-training.

  python script/construct_continue_pretrain_data.py --merge-mode role --merge-max-messages 10 --lang zh

• construct_firefly_data.py: Constructs data in the Firefly format.

For more detailed information on the data processing methods, please refer to: https://github.com/linden713/firefly-roleplay

Training

• continue_training_gemma3.py: Script for continuing the pre-training of a Gemma model.

  python script/continue_training_gemma3.py

• fine_tune_gemma3.py: Script for fine-tuning a Gemma model.

  python script/fine_tune_gemma3.py

Inference

• inference_gr_gemma.py: Launch a Gradio interface for inference with the trained model.

  python script/inference_gr_gemma.py

Weight

Firefly_gemma3n_adapter

Edge Deployment

App Web

© All rights reserved by miHoYo

Other properties and any right, title, and interest thereof and therein (intellectual property rights included) not derived from Honkai: Star Rail belong to their respective owners.