This message was deleted.

yes that’s recommended, since the Runners are running in a different processes/pods with BentoML, it doesn’t limit the model inference. We also recommend using async in the API server side if an async client is available for dynamodb

Thanks @Chaoyu. So in my code example, you're saying it's best practice to use an async boto3 call to retrieve the DynamoDB features in

get_features_from_dynamodb()

? Can you help me understand the benefit of that? Don't we need the features in a data structure before we pass them to

runner.predict.run()

?