Right. Another option is running the model locally from a web page, served from the project source directory, which is still a lot easier than setting up a platform specific GPU accelerated ML dev environment.