Vicuna Chatbot Achieves 90% ChatGPT Quality According to Non-Scientific GPT-4 Evaluation (vicuna.lmsys.org)

Summary:

• Vicuna-13B is an open-source chatbot trained on user-shared conversations from ShareGPT.
• Preliminary evaluation using GPT-4 as a judge shows Vicuna-13B achieves more than 90% quality of OpenAI ChatGPT and Google Bard.

Key terms:

• Vicuna-13B: An open-source chatbot trained by fine-tuning LLaMA on user-shared conversations collected from ShareGPT.
• Preliminary evaluation: Conducted using GPT-4 as a judge to assess the quality of chatbot responses.
• 90% ChatGPT quality: Vicuna-13B achieves more than 90% quality of OpenAI ChatGPT and Google Bard in the preliminary evaluation.
• Training: Fine-tuned a LLaMA base model on approximately 70K user-shared conversations collected from ShareGPT.com.
• Online demo: A publicly available demo of Vicuna-13B for non-commercial use.

Tags:

ChatGPT Research GPT-4 LLaMA Prompt Engineering Vicuna Google Bard Stanford Alpaca Stanford UC Berkeley