Zero One Universe Unveils Yi-9B Model: Crowned the "Top Scholar in Sciences" Among Current Yi Series

2024-03-07

The official WeChat account "01AI" has released an important announcement, declaring the launch of its new open-source model, Yi-9B. This model is hailed as the "top scorer in the science category" within the Yi series, showcasing its powerful technical capabilities through outstanding code and mathematical abilities. Yi-9B boasts impressive specifications, with a default context length of 4K tokens and a total of 8.8B actual parameters, providing users with enhanced information processing capabilities. Building upon the foundation of Yi-6B, the Yi-9B model has undergone further optimization and upgrades, utilizing a dataset of 0.8T tokens for continued training, with data up until June 2023, ensuring the model's cutting-edge and accurate performance. Yi-9B has demonstrated exceptional performance in comprehensive ability assessments. It stands out among similar-sized open-source models, surpassing numerous competitors such as DeepSeek-Coder, DeepSeek-Math, Mistral-7B, SOLAR-10.7B, and Gemma-7B. Particularly in terms of code and mathematical abilities, Yi-9B exhibits strong competitiveness, albeit slightly inferior to DeepSeek-Coder-7B and DeepSeek-Math-7B in certain aspects. Nevertheless, its overall performance remains remarkable. Furthermore, Yi-9B excels in common sense and reasoning abilities, showcasing outstanding performance comparable to models like Mistral-7B, SOLAR-10.7B, and Gemma-7B. This advantage enables Yi-9B to better comprehend and handle complex language tasks, providing users with more precise and intelligent services. It is worth mentioning that the Yi-9B model not only delivers exceptional performance but also possesses significant advantages in terms of usability and cost-effectiveness. The official statement emphasizes that both the BF 16 version of Yi-9B and its quantized version, Int8, can be easily deployed on consumer-grade graphics cards, greatly reducing the threshold and cost of usage. This feature enables more developers to effortlessly utilize the Yi-9B model, promoting the popularization and application of AI technology. Previously, under the leadership of Dr. Kai-Fu Lee, the Chairman and CEO of Innovation Works, the company had already launched two open-source large-scale models, Yi-34B and Yi-6B, providing abundant resources and support for academic research. Now, with the release of Yi-9B, the company further solidifies its leading position in the AI field, injecting new vitality into the entire industry.