If we run All the inference code within the celery task or another task queue like rabbit MQ, we will get the same performance ( if we ignore the impact of ML type instance ).

What are the major benefits we will get if we are using sagemaker async for a non-conventional ML model?

CodePudding user response：

1. SageMaker Async is managed: you don't need a team/skills to choose, develop, test and maintain queuing software.
2. You can also configure autoscaling, to scale up and down (eg to reduce fleet size to zero if queue is empty and reduce costs).
3. It also does S3 interactions for you, (copy to/from S3), you don't need to learn to use S3 code. You would need to add S3 interaction to your ML inference code you were to develop the same thing from scratch.