www.chinaservicesinfo.com

News

Advanced large language model launched by RockAI

Updated: Jan 26, 2024 chinadaily.com.cn Print
Share - WeChat

RockAI in Shanghai, a subsidiary of Stonehill, released its new general-purpose natural large language model called Yan on Wednesday.

Scholars from the Shanghai Artificial Intelligence Technology Association; the Industrial Internet Innovation Center (Shanghai) of the China Academy of Information and Communications Technology; the Shanghai University of Finance and Economics; along with experts from industries and institutions, such as finance and internet, attended the launch.

According to the company, the model is one of the few in the industry that does not use Transformer architecture. It boasts seven times the training efficiency, five times the inference throughput and three times the memory capacity compared to a Transformer model with the same parameters. Additionally, it supports lossless operation on CPUs, low hallucination expression and 100 percent support for private applications.

The research team of RockAI also demonstrated a large number of experimental comparisons between the Yan model and a Transformer model of the same parameter scale. In addition to its technological breakthroughs, the Yan model also possesses six outstanding capabilities. These are privacy, cost-effectiveness, precision, real-time performance, professionalism and universality, making it "born for practical application."

RockAI is among the pioneering enterprises in China to promote the development of large-scale artificial intelligence models. The company has consistently upheld a commitment to technological innovation and product development, offering controllable and reliable AI solutions.

Liu Fanping, CEO of RockAI, expressed his hope that the Yan model could serve as foundational infrastructure for artificial intelligence, assisting the establishment of a robust developer ecosystem within the field.

Copyright©2024 China Daily. All rights reserved.

京ICP备13028878号-6

京公网安备 11010502032503号 京公网安备 11010502032503号