Let's assume you have been through the "click-through experience" of creating the object detection demo in this repo at least once and you now want to automate part of the process with Terraform. Specifically, the first step of deploying the Sagemaker endpoint and the last step of creating an App Runner Service.
Have a look at terraform.tfvars
and fill in the variable values. The model.tar.gz
file that s3_model_location
specifies needs to contain the model data as well as the inference script inference.py
in the directory code
. This means that the base_model_uri
from the sample notebook in the video won't work out of the box, since it doesn't contain the inference script (which is in deploy_source_uri
). So if you want to deploy this model without fine-tuning, you need to either manually repackage it or deploy it via notebook once and save a copy before calling base_model_predictor.delete_model()
.
After installing Terraform and the AWS CLI version 2, simply run this:
export AWS_PROFILE=<your profile>
terraform init
terraform apply
And to clean up:
terraform destroy