Alibaba Cloud launches open source Model-as-a-Service platform
Alibaba Cloud, the digital technology and intelligence backbone of Alibaba Group, has launched ModelScope, an open source Model-as-a-Service (MaaS) platform that comes with hundreds of AI models, including large pre-trained models for global developers and researchers.
During its flagship conference, the global cloud provider also introduced a range of serverless database products and upgraded its integrated data analytics and intelligent computing platform to help customers further achieve business innovation through cloud technologies.
Jeff Zhang, President of Alibaba Cloud Intelligence, says, "Cloud computing has given rise to a fundamental revolution in the way computing resources are organised, produced and put to commercial use while shifting the paradigm of software development and speeding up the integration of the cloud and endpoint terminals.
"As more customers are speeding up with their cloud adoption, we have been upgrading our cloud-based resources, services and tools to become serverless, more intelligent and digitised in order to lower the barrier for companies to adopt new technology and capture more opportunities in the cloud era."
The ModelScope platform was launched with more than 300 ready-to-deploy AI models developed by Alibaba DAMO Academy (‘DAMO’), Alibaba's global research initiative, in the past five years. These models cover various fields from computer vision to natural language processing (NLP) and audio. The platform also includes more than 150 state-of-the-art (SOTA) models, which are recognised globally as the best in their respective fields for achieving the results in a given task, the company states.
Also made available on the platform are Alibaba’s proprietary large pre-trained models such as Tongyi, which is capable of turning text into image with five billion parameters, and OFA (One-For-All), a six billion-parameter pre-trained model that excels at cross-modal tasks such as image captioning and visual question answering.
Independent developers have also contributed dozens of models to the open source platform to date. As an open source community, ModelScope aims to make developing and running AI models easier and more cost effective.
Developers and researchers can test the models online for free and get the results of their tests within minutes. They can also develop customised AI application by fine-tuning existing models, and run the models online backed by Alibaba Cloud, or deploy them on other cloud platforms or in a local setting, the company states.
According to the company, Alibaba Cloud is making its key cloud products serverless to enable customers to concentrate on product deployment and development without worrying about managing servers and infrastructure. Essentially, Alibaba Cloud’s updated products focus on turning computing power into an on-demand capability for users.
Examples of these are the cloud native database PolarDB, the cloud-native data warehouse AnalyticDB (ADB) and ApsaraDB for Relational Database Service (RDS). Leveraging Alibaba Cloud’s serverless technologies, customers can enjoy automatic scaling with extreme elasticity based on actual workloads and a pay-as-you-go billing model to reduce costs.
The automatic elastic scaling time on demands can be as little as one second. The use of updated database products can help businesses in the internet industry reduce their costs by 50%, on average, compared to using traditional ones.
Currently, Alibaba Cloud has more than 20 serverless key products in total and is adding more product categories to become serverless. Alibaba Cloud also upgraded its ODPS (Open Data Platform and Services), a self-developed integrated data analytics and intelligent computing platform, to provide companies with diversified data processing and analytics services.
The platform can handle both online and offline data simultaneously in one system, providing businesses dealing with complex workloads with analytics for business decision-making with reduced cost and increased efficiency.
ODPS has refreshed global records for big data performance, according to recent results from the Transaction Processing Performance Council (TPC), an industry council that sets the standards for transaction processing and database benchmarking. Evaluated based on a 100 TB data benchmark, the performance of ODPS Maxcompute Cluster attained the top score for the sixth consecutive year. ODPS Hologres Cluster also shown a record-breaking result in the TPC-H 30000GB decision support benchmark test.
To drive workload collaboration between cloud and local hardware, Alibaba Cloud has announced the launch of the Wuying Architecture with a showcase on its applications on the Wuying Cloudbook. The Cloudbook with dedicated Architecture is designed to help users access unlimited computing power on the cloud in a more secure and agile manner while supporting collaboration and flexibility at a workplace.