hiyouga/Llama-2-Chinese-13b-chat

TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kPublished:Jul 24, 2023License:llama2Architecture:Transformer0.0K Open Weights Cold

The hiyouga/Llama-2-Chinese-13b-chat is a 13 billion parameter Llama-2-based model, instruction-tuned using LoRA for bilingual capabilities. Developed by hiyouga, it leverages instruction-following datasets including alpaca, alpaca-zh, and open assistant. This model is specifically designed for chat applications requiring both English and Chinese language understanding and generation, with a context length of 4096 tokens. It excels in providing helpful and polite answers to user questions in a conversational format.

Loading preview...

Overview

The hiyouga/Llama-2-Chinese-13b-chat is a 13 billion parameter model built upon the Llama-2 architecture, specifically fine-tuned using the LoRA method. Its primary distinction is its bilingual instruction-following capability, trained on a combination of English and Chinese datasets including alpaca, alpaca-zh, and open assistant. This model is designed to understand and generate responses in both languages, making it suitable for diverse conversational AI applications.

Key Capabilities

  • Bilingual Chat: Proficient in generating helpful, detailed, and polite answers in both English and Chinese.
  • Instruction Following: Trained to adhere to user instructions for various tasks.
  • Code Generation: Demonstrated ability to generate Python code snippets based on natural language requests.
  • Translation: Capable of translating text between English and Chinese.
  • Explanation & Definition: Can provide clear explanations for concepts and abbreviations.

Good For

  • Developing conversational agents that need to operate in both English and Chinese.
  • Applications requiring code generation or technical explanations.
  • Use cases where bilingual question-answering and instruction-following are critical.
  • Leveraging the LLaMA-Factory framework for deployment and further customization.