(资料图片)

Credit: Visual China

BEIJING, June 29 (TiPost) —— Volcengine, a subsidiary of ByteDance, has launched an enterprise-focused big model service platform called Volcano, which provides model fine-tuning, evaluation, and reasoning services for enterprises.

Unlike Ali Cloud and Tencent Cloud, which applied their self-researched big models directly on cloud services, ByteDance is still in the early stage of big models. Volcano Ark is a big model service platform that offers businesses Model-as-a-Service solutions (MaaS). It provides services by integrating big models from other companies. The companies that provide big models for Volcano Ark include seven AI companies and research institutes, such as Baichuan Intelligence, MOSS of Fudan University, International Digital Economy Academy, Langboat, and MiniMax.

This means that Volcengine is expanding its cloud services from the model-building side to the model-application side. Volcengine released a series of cloud products on April 18, such as DPU developed by Volcengine, and launched an upgraded machine-learning platform to support large model training. Big model companies, including Zhipu AI and MiniMax, are customers of Volcengine.

Volcano Ark has started invitation-based testing. The first batch of invited companies includes customers in the field of finance, automotive, and retail. Tan Zai, the president of Volcengine, said: "Most of the training demands in April and May were from model manufacturers, who want to train their models with large parameters as soon as possible, and these were the early demands. Recently many companies also came over to discuss with us, such as companies in the area of manufacturing, finance, and retail, hoping to explore the needs of large models in marketing, customer service, and other scenarios. The launch of the Volcano Ark intends to meet these needs."

More than a dozen business teams of ByteDance have tried Volcano Ark, such as groups in charge of the intelligent office assistant of Lark, the intelligent customer service of the TikTok e-commerce, and the intelligent marketing copy generation of Capcut, using the large model capabilities to reduce costs and increase efficiency. " If ByteDance"s internal teams do a good job in building the models, services are likely to be offered to the public on Volcano Ark," said Tan.

As the "youngest" cloud vendor which receives continuous investment from ByteDance in the enterprise-focused business, Volcengine needs customers. According to Volcengine, its large customers include intelligent driving companies, large financial institutions, Internet giants, large model startups, and research institutions. Volcengine is also collaborating with huge engines to reduce the total cost of retaining customers and acquiring more customers through cloud and advertising.

As the threshold of large model production is extremely high, and AIGC has been able to significantly improve efficiency in the case of rapid iteration of large models, there is a demand for enterprise customers to optimize their business by using big models. And due to export restrictions on high-end GPUs, most of the domestic GPUs (graphics processing units) as good as Nvidia A100 are held by Internet giants. ByteDance is currently one of a few domestic companies that have more than 10,000 Nvidia GPUs, which is an advantage of Volcengine.