Google Cloud Trigger a job task which uses GPU using an http request and no GKE

Question

Using Google Cloud, is there a way to trigger a task of a job using GPUs?

My use case is that I need to run Deep Learning models in images using GPUs, but they are sporadic, and they are normally called by a user who selects an image to use the model on and then a compute resource fetches the item, runs the model and saves results.

This is not a duplicate of this question, I don't want to have a bloated possibly expensive solution, I don't need Kubernetes, no GKE etc.

If anything, this is a better resource: Starting and stopping a Compute VM

Where a single VM is turned on when needed.

But I am wondering what is the best way to cover this use case where a GPU is needed only for a short time, only a few times per year.

Can someone please give me any pointers on how to do this?

I have read cloud.google.com/batch/docs/create-run-job-gpus but as I mention it only explains how to schedule it.