Sanguoyanyi 6b

🚀 StyleLLM: A Large Model for Text Style Transfer

StyleLLM is a text style transfer project based on large language models (LLMs). It leverages LLMs to learn the writing styles (common vocabulary, sentence structures, rhetorical devices, character dialogues, etc.) of specified literary works, creating a series of models with specific styles. These models can transplant the learned styles onto other general texts. That is, given an original text, the model can rewrite it to output a text with the learned style, achieving the effects of text decoration, polishing, or style imitation.

🚀 Quick Start

Model Introduction

StyleLLM is a text style transfer project based on large language models (LLMs). The project uses LLMs to learn the writing styles (common vocabulary, sentence structures, rhetorical devices, character dialogues, etc.) of specified literary works, creating a series of models with specific styles.

With the StyleLLM models, the learned styles can be transplanted onto other general texts. That is, given an original text, the model can rewrite it to output a text with the learned style, achieving the effects of text decoration, polishing, or style imitation.

Model List

StyleLLM - Chinese Four Great Classical Novels Series includes four models fine - tuned on Yi - 6b, trained on the four Chinese classical novels: Romance of the Three Kingdoms, Journey to the West, Water Margin, and Dream of the Red Chamber.

Effect Demonstration

Using the original text of Snow White as input, here are the rewritten results of different style models.

💻 Usage Examples

Basic Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("stylellm/SanGuoYanYi-6b")
model = AutoModelForCausalLM.from_pretrained("stylellm/SanGuoYanYi-6b").eval()

messages = [{"role": "user", "content": "严冬时节，鹅毛一样的大雪片在天空中到处飞舞着，有一个王后坐在王宫里的一扇窗子边，正在为她的女儿做针线活儿，寒风卷着雪片飘进了窗子，乌木窗台上飘落了不少雪花。"}]
input_ids = tokenizer.apply_chat_template(conversation=messages, tokenize=True, add_generation_prompt=True, return_tensors='pt')
output_ids = model.generate(input_ids, do_sample=False, repetition_penalty=1.2)
response = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)
print("Output:", response)
# Output: 时值隆冬，大雪压境，王后坐于王宫之窗下，正与女作绣，寒风入室，吹落乌木窗台之雪。

📄 License

This project uses the yi-license.

👏 Acknowledgments

This project benefited from the following open - source projects during model training. We sincerely thank the relevant project teams:

• 01-ai/Yi: The base model of this project.
• hiyouga/LLaMA-Factory: The training framework of this project.