Hello @Steve Zaske ,
I believe that you are trying to create a compute instance from ML portal ml.azure.com which gives you an option to stop or delete the instance but does not have an option to auto shutoff the instance since this feature is not available for compute instance.
You have an option to use the compute cluster type of compute which allows you to set the minimum/maximum number of nodes. In this case to avoid being billed when the training is not in progress the minimum number of 0 will ensure you do not have any nodes running when not required. You can also set this from the SDK if required.
For example, if you are using the Basic plan and you you train a model for 100 hours using 10 DS14 v2 VMs on an Basic workspace in US West 2. For a billing month of 30 days, your bill will be as follows:
Azure VM Charge: (10 machines * $1.196 per machine) * 100 hours = $1,196
Azure Machine Learning Charge: (10 machines * 16 cores * $0 per core) * 100 hours = $0
Total: $1,196 + $0 = $1,196
But if you are using an Enterprise edition when it becomes generally available, it will have a machine learning surcharge (for training and inferencing). When this is enabled the calculator will be updated to give you an estimate of surcharge based on the amount of time you run the training and inferencing.
So, in the current scenario the charge for your training depends on the amount of time you use the VM. If you use compute instance type you need to ensure to stop or delete it when not required, If compute cluster type is used you can set the minimum number of nodes to 0 to avoid billing when there is no training.
We hope this helps.