01.AI, an artificial intelligence company founded by Kai-Fu Lee, has unveiled the Yi model version 1.5, highlighting that the 9B model is capable of outperforming similar-sized open-source models in test scores (even though technically, the Yi 9B might be considered larger than necessary).
The 01.AI team stated that Yi was trained on an incredibly vast dataset of up to five trillion tokens and further fine-tuned with an additional three million samples.
In an interview with The Japan Times, Kai-Fu Lee mentioned that the Chinese chatbot market has not been as conducive for usage, unlike ChatGPT, which has proven to be highly functional. Additionally, 01.AI introduced Wanzhi, an AI assistant similar to Microsoft Copilot, designed to assist users effectively.
Moreover, 01.AI also unveiled the Yi-Large model, a large-scale model with undisclosed internal architecture, but with test results comparable to leading models such as GPT-4, Claude 3 Sonnet, or Gemini 1.5 Pro. Yi-Large will be offered through an API compatible with OpenAI.
Source: HuggingFace, Japan Times
TLDR: 01.AI, led by Kai-Fu Lee, introduces the Yi model 1.5, showcasing its superior performance and unveiling the Yi-Large model that competes with top models in the industry.
Leave a Comment