deploy_model.py

import sagemaker as sage

# get and pass the auth role and image path, same as before
# this step is unchanged from the training script
role = sage.get_execution_role()
sess = sage.Session()
account = sess.boto_session.client('sts').get_caller_identity()['Account']
region = sess.boto_session.region_name
image = '{}.dkr.ecr.{}.amazonaws.com/quiltdata/sagemaker-demo'.format(account, region)

# create a new Model object
clf = Model(
    # insert model path below
    model_data='s3://quilt-example/quilt/quilt_sagemaker_demo/model/sagemaker-demo-[...]/output/model.tar.gz',
    image=image,
    role=role,
    sagemaker_session=sess
)

# deploy it to an endpoint
predictor = clf.deploy(1, 'ml.c4.2xlarge')

# connect to the endpoint
predictor = sage.predictor.RealTimePredictor(
    'sagemaker-demo-[...]',  # insert model name here 
    sagemaker_session=sess, 
    content_type="text/csv"
)

Hi, awesome tutorial! But where does the Model in line 12 come from?

@nielsrolf

Hi, awesome tutorial! But where does the Model in line 12 come from?

sagemaker.model.Model

why you store clf.deploy(1, 'ml.c4.2xlarge') in predictor if later your overwrite predictor with sage.predictor.RealTimePredictor(...?