Asynchronous inference

SageMaker real-time endpoints are suitable for machine learning use cases that have very low latency inference requirements (up to 60 seconds), along with the data size for inference not being large (maximum 6 MB). On the other hand, batch transforms are suitable for offline inference on very large datasets. Asynchronous inference is another relatively new inference option in SageMaker that can process data up to 1 GB and can take up to 15 minutes in processing inference requests. Hence, they are useful for use cases that do not have very low latency inference requirements.

Asynchronous endpoints have several similarities to real-time endpoints. To create asynchronous endpoints, like with real-time endpoints, we need to carry out the following steps:

1. Create a model.
2. Create an endpoint configuration for the asynchronous endpoint. There are some additional parameters for asynchronous endpoints.
3. Create the asynchronous endpoint.

Asynchronous...