Baize-v2-13B

Baize-v2-13B is an open-source AI assistant developed by UCSD and Sun Yat-Sen University. This large language model is designed to engage in conversations with users, providing detailed responses in Markdown format while adhering to ethical guidelines.

Key features:
- Conversational AI: Engages in detailed dialogues with users
- Markdown Output: Provides responses formatted in Markdown
- Ethical Boundaries: Declines engagement with unethical, controversial, or sensitive topics
- Integration Options: Available for local demo, API, and SDK integration
- Non-Commercial License: Offered by UCSD for research and development purposes

How it works:
1. User inputs a query or statement
2. Baize-v2-13B processes the input using its 13 billion parameter model
3. The model generates a response based on its training
4. The response is output in Markdown format

Integrations:
Local demo, API, SDK

Use of AI:
Baize-v2-13B utilizes advanced AI techniques including supervised fine-tuning (SFT) and self-distillation with feedback (SDF). These methods enable the model to learn from large datasets and generate informative and detailed responses.

AI foundation model:
The foundation model for Baize-v2-13B incorporates supervised fine-tuning and self-distillation with feedback. It is fine-tuned with LoRA (Low-Rank Adaptation) to enhance its performance and adaptability.

Target users:
- Developers integrating conversational AI into applications
- Researchers studying large language models
- Organizations requiring detailed, Markdown-formatted AI responses

How to access:
Baize-v2-13B is available as an open-source model under a non-commercial license from UCSD. Interested parties should contact UCSD for specific access and usage details.

Additional information:
Baize-v2-13B is part of the broader Baize project, which aims to develop comprehensive open-source AI assistants for various applications, including chatbots and virtual assistants.