OpenAPI Overview
OpenAPI is a GPU-focused, API-driven platform that turns Virtual Kubernetes Services (VKS)-based AI model training and fine-tuning into a serverless-style service that users can fully automate from their own systems.
OpenAPI simplifies and accelerates the development and training of AI model. Users only need to provide basic information—such as the container image and key configuration parameters—to quickly deploy AI training or fine-tuning services. In addiong, flexible scaling of compute resources is offered, ensuring that both small experiments and large-scale production workloads can achieve optimal performance and efficiency.
Features
OpenAPI offers comprehensive capabilities across both infrastructure and management. On the infrastructure side, it provides solutions for VKS clusters, storage, and container image registries. On the management side, it supports resource provisioning, configuration management, compute billing, log management, and more. By integrating with OpenAPI, users can automate end-to-end workflows and significantly improve the efficiency of AI model training and fine-tuning.
In addition, OpenAPI offers flexible integration capabilities, allowing users to seamlessly extend their existing systems. It supports distributed GPU training tasks across single-node single-GPU and multi-node multi-GPU configurations, meeting the needs of projects of various scales and technical requirements and helping realize the vision of GPU serverless. In this way, whether for small or large-scale deployments, users can enjoy an experience with efficient and convenient service.
-
Users can easily utilize high-performance GPU compute services without having to manage the lifecycle of the underlying VKS clusters.
-
A comprehensive log panel lets you query real-time log streams, helping users quickly locate and resolve potential issues.
-
The platform uses per-second billing: resources start when used and stop when idle, saving compute resources and truly enabling pay-as-you-go billing to reduce user costs.
-
By leveraging GPU container instance retention, image warm-up, and high-performance hardware, the platform delivers cold start within seconds, helping users confidently handle traffic spikes.
-
Ready-to-use task templates are provided to simplify deployment of AI model training or fine-tuning services for users.
Task Example
- Visit the Alaya NeW Cloud official website to complete registration. If you need help, see Account Registration.
- Sign in to Alaya NeW Cloud and obtain a Serverless API Key.
- Perform the subsequent workflow as shown below.
*: Indicates that this operation is optional.
The workflow diagram is explained as follows:
-
After signing in, the user queries the VKS cluster list, creates a task template from the selected VKS resources, and then generates a task from the template.
-
After the task is created, the task is started and its details and logs are queried. Monitoring is then enabled to track the task status in real time and view metrics for related resources.
The above diagram indicates a sample workflow. The actual workflow may vary by use case. For more information, refer to the relevant API documentation.