From my understanding publishing AI/ML models to GCP marketplace differs from traditional apps that I see that currently exist like Solr, ELK , Wordpress both from an operational perspective ( packing the app as a k8s app or whatever) and also the need for enhanced security that comes with the model.
My question is more on the architecture / best good practices in publishing AI/ML models to GCP Marketplace and if in fact marketplace is the preferred solution here. Imagine a company with access to large datasets partnering with another company that has a foundational model and wanting to monetize the model by having a well defined orchestration / setup to package the app and offer a 1-click deploy solution to their end customers with metering included.
The ideal way I see this working is the foundational model getting deployed in a Google controlled project if you will that is part of the marketplace deployment and the end customers deploying their training and inference infra in their own customer project and calling the foundational model for training and inference experiments.
What would be the ideal architecture for the above considering security is paramount
LinkedIn
Twitter
