Sagemaker Pipeline Deploy Model Step

0

According to Sagemaker's Pipeline Python SDK documenation, looks like there is no specific pipeline step for model deployment.

Can you please confirm this and, also, if there is a plan to have such a step?

What is the recommended way to add a pipeline step to deploy the trained model, resulting in an enpoint being created?

1 Answer
2
Accepted Answer

Hi, there is indeed no specific pipeline step for model deployment. The idea is that SageMaker Pipelines is more about "batch mode", but customers do ask for this feature, so it might be added.

You can implement it quite easily using Lambda Step.

1st create a Lambda function to deploy/update the model:

%%writefile deploy_model_lambda.py


"""
This Lambda function deploys the model to SageMaker Endpoint. 
If Endpoint exists, then Endpoint will be updated with new Endpoint Config.
"""

import json
import boto3
import time


sm_client = boto3.client("sagemaker")


def lambda_handler(event, context):

    print(f"Received Event: {event}")

    current_time = time.strftime("%m-%d-%H-%M-%S", time.localtime())
    endpoint_instance_type = event["endpoint_instance_type"]
    model_name = event["model_name"]
    endpoint_config_name = "{}-{}".format(event["endpoint_config_name"], current_time)
    endpoint_name = event["endpoint_name"]

    # Create Endpoint Configuration
    create_endpoint_config_response = sm_client.create_endpoint_config(
        EndpointConfigName=endpoint_config_name,
        ProductionVariants=[
            {
                "InstanceType": endpoint_instance_type,
                "InitialVariantWeight": 1,
                "InitialInstanceCount": 1,
                "ModelName": model_name,
                "VariantName": "AllTraffic",
            }
        ],
    )
    print(f"create_endpoint_config_response: {create_endpoint_config_response}")

    # Check if an endpoint exists. If no - Create new endpoint, if yes - Update existing endpoint
    list_endpoints_response = sm_client.list_endpoints(
        SortBy="CreationTime",
        SortOrder="Descending",
        NameContains=endpoint_name,
    )
    print(f"list_endpoints_response: {list_endpoints_response}")

    if len(list_endpoints_response["Endpoints"]) > 0:
        print("Updating Endpoint with new Endpoint Configuration")
        update_endpoint_response = sm_client.update_endpoint(
            EndpointName=endpoint_name, EndpointConfigName=endpoint_config_name
        )
        print(f"update_endpoint_response: {update_endpoint_response}")
    else:
        print("Creating Endpoint")
        create_endpoint_response = sm_client.create_endpoint(
            EndpointName=endpoint_name, EndpointConfigName=endpoint_config_name
        )
        print(f"create_endpoint_response: {create_endpoint_response}")

    return {"statusCode": 200, "body": json.dumps("Endpoint Created Successfully")}

Then create the Lambda step:

deploy_model_lambda_function_name = "sagemaker-deploy-model-lambda-" + current_time

deploy_model_lambda_function = Lambda(
    function_name=deploy_model_lambda_function_name,
    execution_role_arn=lambda_role,
    script="deploy_model_lambda.py",
    handler="deploy_model_lambda.lambda_handler",
)

You can see a full working example in this notebook.

profile pictureAWS
answered a year ago
profile picture
EXPERT
reviewed 18 hours ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.

Guidelines for Answering Questions