Recently, there have been frequent reports of Chinese companies announcing their own big models. In this wave of big models, the biggest unicorn in China finally disclosed the latest news about big models to the outside world.

According to the official official account of Volcano Engine, on June 28, Volcano Engine, a cloud service platform under ByteDance, hosted the "V-Tech Experience Innovation Technology Summit". At the meeting, Volcano Engine released the large model service platform "Volcano Ark", which will provide enterprises with comprehensive platform services such as model fine-tuning, evaluation, and inference (MaaS, also known as Model as a Service).

It is worth noting that unlike Baidu, Alibaba, Tencent and other domestic Internet technology enterprises, the Volcano Ark released by ByteDance is not a self researched large model, but a large model service platform. Tan Dai, the President of Volcano Engine, said, "We will put the selected large models on the platform for customers to choose and use." In Tan Dai's view, Volcano Ark is like a "preferred store" of large models. Based on Volcano Ark, enterprises can simultaneously try multiple large models and choose a combination of models that are more suitable for their business needs.

At the same time, Tan Dai also stated, "Volcano Ark is an open platform. If other teams within ByteDance have completed their models, they will probably provide them to the public on the Ark platform. It is one of many models on the Ark platform."

In terms of volcanic engines, the future large model market will be a diverse multi model ecosystem. In the future, the application of large models by enterprises themselves will be a "1+N" application model, which involves a collaboration between a main model and N external models. It is reported that the Volcano Ark will create an ecological panorama consisting of the Volcano Ark, model providers, and model users.

For model providers, using "Volcano Ark" can reach a large number of customers at a lower cost and achieve scale in the ToB market at a lower cost; A rigorous security and mutual trust mechanism, taking into account flexibility and security; A continuous stream of computing power forms the most competitive cost-effectiveness. For model users, the Volcano Ark provides convenient access to numerous high-quality base models, enabling one-stop integration with multiple model providers and selecting the most suitable model for different scenarios.

Specifically, the Volcano Ark has created multiple core components based on the working habits of large model applications. The Model Square will have different model suppliers providing different versions and sizes of models, and users can directly interact with the models, call inference APIs, and connect to the production environment; "Model evaluation" is the key entry point for the construction of "Volcano Ark". Users can design a set of quantifiable evaluation indicators based on business needs, and select the most suitable model after model evaluation; Model fine-tuning can help customers use their own data for continuous training, build and accumulate their own fine-tuning datasets, and reduce inference costs.

Wu Di, the head of the Volcano Engine Intelligent Algorithm, said that a well tuned small to medium-sized model may perform as well in specific tasks as a universal, massive base model, and the inference cost can be reduced to one tenth of the original.At the same time, in order to promote mutual trust between model providers and users, "Volcano Ark" has launched a large model security mutual trust computing scheme based on a secure sandbox, utilizing methods such as computation isolation, storage isolation, network isolation, and traffic auditing to achieve model confidentiality, integrity, and availability. This scheme is suitable for customers with low training and inference delay requirements.

According to the official website, the industry scene of Volcano Ark includes automobile, finance, mass consumption, pan Internet and education office. In the automotive field, Volcano Ark provides services such as intelligent cockpit interaction, after-sales knowledge base, vehicle health monitoring, and vehicle operation guidance. In the financial field, Volcano Ark provides intelligent investment research assistants, intelligent risk control, online store robots, intelligent outbound calls, and more. In terms of big consumption, there are intelligent marketing, intelligent customer service, public opinion analysis, and advertising copy generation. In terms of pan Internet, there are two services: game character building and smart digital people. Education office includes intelligent collaborative office and intelligent education assistants.

It is worth noting that Nvidia is a co organizer of this technology summit. Li Xipeng, General Manager of NVIDIA's Asia Pacific Development and Technology Department, stated that NVIDIA and Volcano Engine have achieved fruitful cooperation in the past. Previously, both parties jointly opened up the high-performance image processing acceleration library CV-CUDA, and achieved results in technical cooperation in large-scale stable training, multi model hybrid deployment, and other areas.

In the future, Nvidia and the Volcano Engine team will continue to deepen cooperation, including adaptation and optimization of the NVIDIA Hopper architecture, confidential computing, key model collaboration optimization, joint support for key customers, and NeMo Framework adaptation, to jointly help promote the prosperity of the large model industry.

Wu Di stated that Volcano Ark is still exploring various security and mutual trust computing solutions based on NVIDIA's next-generation hardware support for a trusted computing environment, as well as data asset separation based on federated learning, to more comprehensively meet the data security requirements of large models in different business scenarios.

At present, more than ten business teams within Douyin Group have tried the "Volcano Ark" to explore the efficiency improvement scenarios such as code correction, knowledge management scenarios such as text classification and summary, as well as data annotation and attribution analysis, and promote cost reduction and efficiency increase by using large model capabilities.

According to the Volcano Engine, the "Volcano Ark" currently integrates large models from multiple AI technology companies and research institutes, including Baichuan Intelligence, Go Out and Ask, Fudan University MOSS, IDEA Research Institute, Lanzhou Technology, MiniMax, and Zhipu AI, and has launched an invitation for testing. The first batch of invited testing companies included customers from various industries such as finance, automotive, and consumer goods.

"Every major technological change will bring new opportunities for experiential innovation," Tan Dai said. "The Ark of the Volcano is still in its early stages, and the toolchain and downstream application plugins need to be continuously improved. In the future, the platform will also integrate more large models and gradually expand the scope of invitation for testing, jointly building an open and cooperative multi model ecosystem with enterprise customers, and accelerating the application of large models in various industries."