def deploy_xgboost(self,
s3_model_location,
instance_count,
instance_type,
framework_version,
model_server_workers=None,
tags=None,
endpoint_name=None):
model = sagemaker.xgboost.model.XGBoostModel(
role=self.role,
model_data=s3_model_location,
framework_version=framework_version,
py_version="py3",
source_dir=os.path.join(_FILE_DIR_PATH, "xgboost_code"),
entry_point="xgboost_inference.py",
model_server_workers=model_server_workers,
sagemaker_session=self.sagemaker_session)
try:
predictor = model.deploy(instance_type=instance_type,
initial_instance_count=instance_count,
tags=tags,
endpoint_name=endpoint_name)
except botocore.exceptions.ClientError:
# ValueError raised if there is no endpoint already
predictor = sage.Predictor(
endpoint_name=endpoint_name,
sagemaker_session=self.sagemaker_session)
predictor.update_endpoint(initial_instance_count=instance_count,
instance_type=instance_type,
tags=tags,
model_name=model.name)
return predictor.endpoint_name
def deploy(self,
image_name,
s3_model_location,
train_instance_count,
train_instance_type,
tags=None,
endpoint_name=None):
"""
Deploy model to SageMaker
:param image_name: [str], name of Docker image
:param s3_model_location: [str], model location in S3
:param train_instance_count: [str], number of ec2 instances
:param train_instance_type: [str], ec2 instance type
:param tags: [optional[list[dict]], default: None], List of tags for labeling a training
job. For more, see https://docs.aws.amazon.com/sagemaker/latest/dg/API_Tag.html. Example:
[
{
'Key': 'key_name_1',
'Value': key_value_1,
},
{
'Key': 'key_name_2',
'Value': key_value_2,
},
...
]
:param endpoint_name: [optional[str]], Optional name for the SageMaker endpoint
:return: [str], endpoint name
"""
image = self._construct_image_location(image_name)
model = sage.Model(model_data=s3_model_location,
image_uri=image,
role=self.role,
sagemaker_session=self.sagemaker_session)
try:
model.deploy(initial_instance_count=train_instance_count,
instance_type=train_instance_type,
tags=tags,
endpoint_name=endpoint_name)
return model.endpoint_name
except botocore.exceptions.ClientError:
# ValueError raised if there is no endpoint already
predictor = sage.Predictor(
endpoint_name=endpoint_name,
sagemaker_session=self.sagemaker_session)
predictor.update_endpoint(
initial_instance_count=train_instance_count,
instance_type=train_instance_type,
tags=tags,
model_name=model.name)
return predictor.endpoint_name
def deploy_hugging_face(self,
instance_count,
instance_type,
transformers_version=None,
pytorch_version=None,
tensorflow_version=None,
s3_model_location=None,
hub=None,
model_server_workers=None,
tags=None,
endpoint_name=None):
def _validate_either_of_them(name_a, name_b, var_a, var_b):
if var_a is not None and var_b is not None:
raise ValueError(f'{name_a} and {name_b} are both not None. '
f'Specify only {name_a} or {name_b}.')
if var_a is None and var_b is None:
raise ValueError("{name_a} and {name_b} are both None. "
"Specify either {name_a} or {name_b}.")
_validate_either_of_them(name_a='pytorch_version',
name_b='tensorflow_version',
var_a=pytorch_version,
var_b=tensorflow_version)
_validate_either_of_them(name_a='model_location',
name_b='hub',
var_a=s3_model_location,
var_b=hub)
model = sagemaker.huggingface.HuggingFaceModel(
role=self.role,
model_data=s3_model_location,
transformers_version=transformers_version,
pytorch_version=pytorch_version,
tensorflow_version=tensorflow_version,
model_server_workers=model_server_workers,
py_version='py36',
env=hub,
sagemaker_session=self.sagemaker_session)
try:
predictor = model.deploy(instance_type=instance_type,
initial_instance_count=instance_count,
tags=tags,
endpoint_name=endpoint_name)
except botocore.exceptions.ClientError:
# ValueError raised if there is no endpoint already
predictor = sage.Predictor(
endpoint_name=endpoint_name,
sagemaker_session=self.sagemaker_session)
predictor.update_endpoint(initial_instance_count=instance_count,
instance_type=instance_type,
tags=tags,
model_name=model.name)
return predictor.endpoint_name
def test_predict_jsonlines(tfs_predictor):
input_data = "[1.0, 2.0, 5.0]\n[1.0, 2.0, 5.0]"
expected_result = {"predictions": [[3.5, 4.0, 5.5], [3.5, 4.0, 5.5]]}
predictor = sagemaker.Predictor(
tfs_predictor.endpoint_name,
tfs_predictor.sagemaker_session,
serializer=IdentitySerializer(content_type="application/jsonlines"),
deserializer=JSONDeserializer(),
)
result = predictor.predict(input_data)
assert expected_result == result
def predict_wrapper(endpoint, session):
return sagemaker.Predictor(endpoint,
session,
serializer=CSVSerializer())
