project-baize/baize-v2-7b
Baize-v2-7b is a 7 billion parameter open-source chat model developed by UCSD and Sun Yat-Sen University, fine-tuned with LoRA using supervised fine-tuning (SFT) and self-distillation with feedback (SDF). This model is designed for conversational AI, providing detailed responses in a specific chat format. It is based on the LLaMA architecture and is intended for general-purpose chat applications.
Loading preview...
Overview
Baize-v2-7b is an open-source 7 billion parameter chat model developed by UCSD and Sun Yat-Sen University. It is fine-tuned using LoRA with a combination of supervised fine-tuning (SFT) and self-distillation with feedback (SDF) techniques. The model is named after Baize, a mythical creature in Chinese folklore known for its extensive knowledge, reflecting the developers' aspirations for the chat model's capabilities. This specific checkpoint has been merged with LLaMA, making it ready for immediate use in conversational AI applications.
Key Capabilities
- Conversational AI: Designed to engage in detailed human-like conversations.
- Specific Formatting: Requires a precise input format (
[|Human|]and[|AI|]) for optimal performance, ensuring structured dialogue. - Ethical Guidelines: Programmed to decline engagement with unethical, controversial, or sensitive topics.
- Markdown Output: Always provides responses in Markdown format, enhancing readability and structure.
Usage Considerations
- Strict Input Format: Users must adhere to the specified
[|Human|]...[|AI|]...conversation format for the model to function correctly. - Open-Source: The project provides access to its GitHub repository for further details on usage, demo, and integration with tools like Fastchat.
- Research-Backed: The underlying methodology and details are available in their academic paper.