Vicuna-13B: the open chatbot that reaches 90% of ChatGPT quality
In one sentence LMSYS fine-tunes LLaMA-13B on 70,000 ShareGPT conversations and produces an open-source chatbot that GPT-4, used as judge, rates at 90% of ChatGPT quality.
Until spring 2023, having a chatbot as good as ChatGPT seemed impossible without OpenAI API access. A group of Berkeley researchers proved otherwise.
They took LLaMA — the model Meta had made available to researchers — and further trained it on 70,000 real conversations copied from ShareGPT, the site where people shared their ChatGPT chats. Result: Vicuna, a chatbot that an AI judge (GPT-4) rates equivalent to 90% of ChatGPT in response quality.
The important lesson isn't just the number: quality of data beats quantity. 70,000 well-chosen conversations outperform millions of generic texts. Vicuna opened everyone's eyes to how feasible it was to build a capable chatbot with limited resources.
Companies
LMSYS
Tools
—
Tags
Sources