Shulou(Shulou.com)11/24 Report--

CTOnews.com Sept. 8, at the 2023 Bund Conference, Ant Group announced the release of a financial model and an open source generative AI programming platform CodeFuse.

He Zhengyu, chief technology officer of Ant Group and president of the platform Technology Group, said in an interview that the ant model takes the technical route of pure self-research, takes the full stack layout and long-term development as the principle, and aims to create industrial value. To this end, ants are determined to invest in the underlying infrastructure of the large model, and have built a million-card AI cluster, leading the industry in training efficiency, providing strong support for the industrial application of the large model.

According to he Zhengyu, Ant has always adhered to independent innovation of core technology, and formally established a large model research and development project at the end of 2022. At present, it has formed a full stack layout from basic large model to industry large model and industrial application.

The ant financial model released today is based on the ant foundation model and is deeply customized for the financial industry. According to he Zhengyu, the ant-based large model platform has a million-card heterogeneous cluster, in which the kcal scale training MFU can reach 40%, and the effective training time of the cluster accounts for more than 90%. Under the same model effect, the training throughput performance of RLHF training is 3.59 times higher than that of the industry solution, and the reasoning performance is about 2 times higher than that of the industry solution, which is at the advanced level in the industry.

He Zhengyu said that in the future, ants will continue to explore and refine the five major capability directions of the big model. CTOnews.com summarizes as follows:

First, build a high-quality data labeling team and precipitate a high-quality data system.

Second, attack the basic large model algorithm, as well as the efficient green engineering ability, improve the model logic reasoning ability and so on.

Third, from the general language model to the general multimodal model, from general knowledge to comprehensive specialization.

Fourth, build efficient large model evaluation standards and evaluation system to speed up the iterative speed of large models.

Fifth, build the safety capacity of the large model to ensure the healthy and sustainable development of the large model.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.