src/graph_notebook/notebooks/03-Neptune-ML/02-SPARQL/neptune_ml_sparql_utils.py [518:608]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - config = json.load(f) region_name = boto3.session.Session().region_name if region_name in ['cn-north-1', 'cn-northwest-1']: self.PRETRAINED_MODEL = config['models_cn'] else: self.PRETRAINED_MODEL = config['models'] self.PYTORCH_CPU_CONTAINER_IMAGE = config['container_images'][region_name] def __run_create_model(self, sm_client, name, role, image_uri, model_s3_location, container_mode='SingleModel', script_name='infer_entry_point.py', ): model_environment_vars = {self.SCRIPT_PARAM_NAME.upper(): script_name, self.DIR_PARAM_NAME.upper(): model_s3_location, self.CONTAINER_LOG_LEVEL_PARAM_NAME.upper(): str(20), self.MODEL_SERVER_TIMEOUT_PARAM_NAME.upper(): str(1200), self.MODEL_SERVER_WORKERS_PARAM_NAME.upper(): str(1), self.SAGEMAKER_REGION_PARAM_NAME.upper(): boto3.session.Session().region_name, self.ENABLE_CLOUDWATCH_METRICS_PARAM.upper(): "false" } container_def = [{"Image": self.PYTORCH_CPU_CONTAINER_IMAGE, "Environment": model_environment_vars, "ModelDataUrl": model_s3_location, "Mode": container_mode }] request = {"ModelName": name, "ExecutionRoleArn": role, "Containers": container_def } return sm_client.create_model(**request) def __run_create_endpoint_config(self, sm_client, model_name, instance_type='ml.m5.2xlarge', initial_instance_count=1, initial_weight=1, variant_name='AllTraffic' ): production_variant_configuration = [{ "ModelName": model_name, "InstanceType": instance_type, "InitialInstanceCount": initial_instance_count, "VariantName": variant_name, "InitialVariantWeight": initial_weight, }] request = {"EndpointConfigName": model_name, "ProductionVariants": production_variant_configuration } return sm_client.create_endpoint_config(**request) def __create_model(self, name: str, model_s3_location: str): image_uri = self.PYTORCH_CPU_CONTAINER_IMAGE instance_type = self.INSTANCE_TYPE role = self.__get_neptune_ml_role() sm = boto3.client("sagemaker") name = "{}-{}".format(name, strftime("%Y-%m-%d-%H-%M-%S", gmtime())) create_model_result = self.__run_create_model( sm, name, role, image_uri, model_s3_location) create_endpoint_config_result = self.__run_create_endpoint_config( sm, name, instance_type=instance_type) create_endpoint_result = sm.create_endpoint( EndpointName=name, EndpointConfigName=name) return name def __get_neptune_ml_role(self): with open(f'{HOME_DIRECTORY}/.bashrc') as f: data = f.readlines() for d in data: if str.startswith(d, 'export NEPTUNE_ML_ROLE_ARN'): parts = d.split('=') if len(parts) == 2: return parts[1].rstrip() logging.error("Unable to determine the Neptune ML IAM Role.") return None def __copy_s3(self, s3_bucket_uri: str, source_s3_uri: str): path = urlparse(s3_bucket_uri, allow_fragments=False) bucket = path.netloc file_path = path.path.lstrip('/').rstrip('/') source_path = urlparse(source_s3_uri, allow_fragments=False) source_bucket = source_path.netloc source_file_path = source_path.path.lstrip('/').rstrip('/') s3 = boto3.resource('s3') s3.meta.client.copy( {"Bucket": source_bucket, "Key": source_file_path}, bucket, file_path) - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - src/graph_notebook/notebooks/03-Neptune-ML/03-Sample-Applications/04-Telco-Networks/neptune_ml_utils.py [490:580]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - config = json.load(f) region_name = boto3.session.Session().region_name if region_name in ['cn-north-1', 'cn-northwest-1']: self.PRETRAINED_MODEL = config['models_cn'] else: self.PRETRAINED_MODEL = config['models'] self.PYTORCH_CPU_CONTAINER_IMAGE = config['container_images'][region_name] def __run_create_model(self, sm_client, name, role, image_uri, model_s3_location, container_mode='SingleModel', script_name='infer_entry_point.py', ): model_environment_vars = {self.SCRIPT_PARAM_NAME.upper(): script_name, self.DIR_PARAM_NAME.upper(): model_s3_location, self.CONTAINER_LOG_LEVEL_PARAM_NAME.upper(): str(20), self.MODEL_SERVER_TIMEOUT_PARAM_NAME.upper(): str(1200), self.MODEL_SERVER_WORKERS_PARAM_NAME.upper(): str(1), self.SAGEMAKER_REGION_PARAM_NAME.upper(): boto3.session.Session().region_name, self.ENABLE_CLOUDWATCH_METRICS_PARAM.upper(): "false" } container_def = [{"Image": self.PYTORCH_CPU_CONTAINER_IMAGE, "Environment": model_environment_vars, "ModelDataUrl": model_s3_location, "Mode": container_mode }] request = {"ModelName": name, "ExecutionRoleArn": role, "Containers": container_def } return sm_client.create_model(**request) def __run_create_endpoint_config(self, sm_client, model_name, instance_type='ml.m5.2xlarge', initial_instance_count=1, initial_weight=1, variant_name='AllTraffic' ): production_variant_configuration = [{ "ModelName": model_name, "InstanceType": instance_type, "InitialInstanceCount": initial_instance_count, "VariantName": variant_name, "InitialVariantWeight": initial_weight, }] request = {"EndpointConfigName": model_name, "ProductionVariants": production_variant_configuration } return sm_client.create_endpoint_config(**request) def __create_model(self, name: str, model_s3_location: str): image_uri = self.PYTORCH_CPU_CONTAINER_IMAGE instance_type = self.INSTANCE_TYPE role = self.__get_neptune_ml_role() sm = boto3.client("sagemaker") name = "{}-{}".format(name, strftime("%Y-%m-%d-%H-%M-%S", gmtime())) create_model_result = self.__run_create_model( sm, name, role, image_uri, model_s3_location) create_endpoint_config_result = self.__run_create_endpoint_config( sm, name, instance_type=instance_type) create_endpoint_result = sm.create_endpoint( EndpointName=name, EndpointConfigName=name) return name def __get_neptune_ml_role(self): with open(f'{HOME_DIRECTORY}/.bashrc') as f: data = f.readlines() for d in data: if str.startswith(d, 'export NEPTUNE_ML_ROLE_ARN'): parts = d.split('=') if len(parts) == 2: return parts[1].rstrip() logging.error("Unable to determine the Neptune ML IAM Role.") return None def __copy_s3(self, s3_bucket_uri: str, source_s3_uri: str): path = urlparse(s3_bucket_uri, allow_fragments=False) bucket = path.netloc file_path = path.path.lstrip('/').rstrip('/') source_path = urlparse(source_s3_uri, allow_fragments=False) source_bucket = source_path.netloc source_file_path = source_path.path.lstrip('/').rstrip('/') s3 = boto3.resource('s3') s3.meta.client.copy( {"Bucket": source_bucket, "Key": source_file_path}, bucket, file_path) - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -