Cloud “functions” are an abstraction that allows you to run your code without managing deployments and infrastructure. Cloud Functions simplifies hosting AI inference and fine-tuning workloads in the cloud by automatically enabling access to GPU capacity and autoscaling. Cloud functions are generally considered stateless.
Therefore, function authors are only responsible for maintaining their AI models and associated code. This is highlighted in the diagram below in green. To use Cloud Functions, you create a function, then define a deployment specification for it, and deploy it on one of the available GPU-backed clusters hosted by NVIDIA.

A Cloud Functions account can contain multiple functions, each with multiple function versions. Each function created also creates a single function version.
Cloud Functions supports function invocation (calling of the function’s inference endpoint) at the function ID level or the function version ID level. You can create a single function and version and invoke only this function version, or create multiple versions of the same function and spread invocation across all versions.
See below for an overview of some key basic concepts within Cloud Functions.
A function can be in any of the following states:
ACTIVE - If the function can receive invocations. Only when a function is ACTIVE or DEGRADING can it be invoked.ERROR - If all function instances are in an ERROR state.INACTIVE - When a function is created but not yet deployed, it is INACTIVE. When a function is undeployed, the state is changed from ACTIVE to INACTIVE.DEPLOYING - When a function is being deployed and the instances are still coming up to reach the minimum instance count.DEGRADING - If ACTIVE is losing its instances and number of active is below deployment configuration field minInstance, but there are some active instances. In this state a function can be invoked. When it gets back all required instances, it will be ACTIVE again.DEGRADED - If ACTIVE or DEGRADING has lost all its instances. In this state a function can NOT be invoked. When it gets some instances it will be back DEGRADING. When it gets back all required instances, it will be ACTIVE again.The workflow when using Cloud Functions is usually as follows.
The table below provides an overview of the function lifecycle API endpoints and their respective usages.
When using the Cloud Functions API to create a function, it’s possible to specify a function description and a list of tags as strings as part of the function creation request body. This metadata is then returned in all responses that include the function definition. This is an API-only feature at this time. Please see the open-api for usage.
The table below provides an overview of the Function invocation API endpoints and their respective usages.
Read more about using the invocation API in the Function Invocation section.