top of page

Dr. Kai-Fu Lee's AI company has released its first open-source bilingual large model, "Yi"




According to the latest rankings from the Hugging Face English open-source community platform and the C-Eval Chinese evaluation, the Yi-34B pretrained model has achieved recognition for multiple SOTA (State-of-the-Art) international performance indicators.


When comparing parameter quantity and performance, Yi-34B excels, surpassing models like LLaMA2-70B and Falcon-180B. It has also outperformed the competition in evaluation sets such as MMLU and BBH, demonstrating superior performance in various test tasks worldwide.


Yi-34B demonstrates a deeper understanding of the Chinese language. It surpasses all global open-source models in the C-Eval Chinese authoritative ranking. Compared to the benchmark model GPT-4, Yi-34B exhibits an absolute advantage in the three main Chinese language benchmarks: CMMLU, E-Eval, and Gaokao, highlighting its exceptional capabilities in the Chinese language.


Moreover, the open-sourced Yi-34B model offers an extended context window version, supporting up to 200,000 characters for ultra-long text input and the comprehension of documents exceeding 1,000 pages in PDF format.


Dr. Kai-Fu Lee, founder and CEO of Zero to One Million, expressed the company's unwavering commitment to becoming the world's top player, and their achievement in delivering a globally competitive product through steady research engineering over the past half year.


In the near future, they plan to rapidly open-source a series of quantized versions, conversational models, mathematical models, code models, and multimodal models based on the Yi base model. They invite developers to actively participate and collectively promote the prosperity of the language model open-source community, nurturing a new "AI-first" innovation ecosystem.


The Yi series models are now officially available on three major global open-source community platforms: HuggingFace, ModelScope, and GitHub. They have also opened applications for commercial use, providing developers with a wider range of high-quality choices when using LLM. They hope to see more developers utilize the Yi series models to build their own "ChatGPT" applications in various contexts.






Model Links


References


8 views0 comments

Comments


bottom of page