| from sagemaker.huggingface import HuggingFaceModel | |
| import sagemaker | |
| # IAM role with permissions to create endpoint | |
| role = sagemaker.get_execution_role() | |
| # public S3 URI to gpt-j artifact | |
| model_uri="s3://huggingface-sagemaker-models/transformers/4.12.3/pytorch/1.9.1/gpt-j/model.tar.gz" | |
| # create Hugging Face Model Class | |
| huggingface_model = HuggingFaceModel( | |
| model_data=model_uri, | |
| transformers_version='4.12.3', | |
| pytorch_version='1.9.1', | |
| py_version='py38', | |
| role=role, | |
| ) | |
| # deploy model to SageMaker Inference | |
| predictor = huggingface_model.deploy( | |
| initial_instance_count=1, # number of instances | |
| instance_type='ml.g4dn.xlarge', #'ml.p3.2xlarge' # ec2 instance type | |
| endpoint_name='sm-endpoint-gpt-j-6b' | |
| ) |