What you call for inference
A catalog call names the model id. text-embed-v1 is 3 credits per 1,000 tokens. sentiment-multilingual-v1 is 2 credits per 1,000 tokens. Vision catalog models are 3 credits per image. Compute beside the call is 0.01 credits per GiB-second. The charge lands on the project balance.
A custom endpoint, when you need one, is an inference_endpoints entry with a name and a runtime. That entry is something you write. It is not produced by reading a SageMaker endpoint description.
What stays behind in SageMaker
Notebooks, training jobs, model packages, endpoint configurations, and instance types are not imported. There is no block in stack.json that receives a SageMaker export.
You choose a catalog id or you declare an endpoint yourself. You do not point KERNSTACK at an AWS account and expect the endpoint to reappear with its production variant and its autoscaling policy.
Amazon SageMaker vs KERNSTACK on European Bare Metal
| Dimension | Amazon SageMaker | KERNSTACK Inference |
|---|---|---|
| Operator | AWS account | KERNSTACK project |
| Manifest | Amazon SageMaker console or API | catalog model id in the SDK call |
| Region | AWS region list | European bare metal |
| Bill | Amazon SageMaker line item | One project credit balance |
Frequently asked questions
No. Catalog models are SDK calls by id. A custom endpoint is an inference_endpoints entry you write, not a copy of a SageMaker configuration.
text-embed-v1, sentiment-multilingual-v1, and the vision catalog models. The model id is in the SDK call.
text-embed-v1 is 3 credits per 1,000 tokens. sentiment-multilingual-v1 is 2. Vision models are 3 credits per image.
Related capabilities, architectures, and comparisons
Ship your stack on the Software Lifecycle Cloud.
Closed preview on dedicated European infrastructure. Bring a repository or start from a blueprint.