Ant Group leverages domestic chips to cut AI training costs

Chinese fintech company Ant Group said it has continuously optimized the training of artificial intelligence-powered models by using different chips to lower the costs of AI applications, and will gradually make these models open-source.

The company said it has leveraged domestically produced chips to train its AI models, reducing computing costs by around 20 percent. It also utilized Nvidia's chips to train such models.

In a research paper, Ant addressed the challenges associated with training large-scale Mixture of Experts or MoE models, including cost inefficiency and resource limitations, by introducing innovative strategies to improve efficiency in resource-constrained environments.

The remarks came after a report from Bloomberg stated that Ant used chips from Chinese tech giants Alibaba and Huawei to train its AI models, achieving performance levels comparable to Nvidia's H800 chips.

fanfeifei@chinadaily.com.cn