Overview

shenzhi-wang/Llama3-8B-Chinese-Chat is an 8 billion parameter instruction-tuned language model, developed by Shenzhi Wang and his team, based on Meta-Llama-3-8B-Instruct. This model is specifically fine-tuned for Chinese and English users, utilizing the ORPO method on a mixed Chinese-English dataset of approximately 100K preference pairs for its v2.1 iteration. It aims to provide enhanced performance in bilingual contexts.

Key Capabilities

• Bilingual Proficiency: Significantly reduces "Chinese questions with English answers" and mixed language responses compared to the base Llama-3 model.
• Enhanced Performance: The v2.1 version, trained on a 5x larger dataset than v1, shows substantial improvements in:
  • Roleplay: Demonstrates strong ability to adopt and maintain specific personas.
  • Function Calling: Capable of interpreting user requests to call predefined tools.
  • Math: Improved accuracy and reasoning for mathematical problems.
  • Reduced English Inclusion: Less prone to including English words in Chinese responses compared to v2.
• Context Length: Supports a context length of 8192 tokens.

Good for

• Applications requiring robust bilingual (Chinese-English) conversational AI.
• Use cases involving role-playing scenarios or character-based interactions.
• Integrating with tools or systems via function calling.
• Educational or problem-solving platforms needing mathematical reasoning capabilities.
• Developers looking for a Llama-3 based model optimized for Chinese language tasks.