How Cloud Functions for Machine Learning Can Become True

DeepMux recently announced GPU-powered serverless functions to make your MLOps (DevOps for ML) easier. Let’s talk about Cloud Functions and their use in Machine Learning!

Modern applications face higher workloads and stricter SLA. It means that more and more effort is needed to be made to meet the production requirements. Serverless approach and Cloud Functions in particular allow you to abstract low-level design and focus on business logic.

Literally Cloud Function is a piece of code or, should I say a piece of logic, that is placed to serve at the specific endpoint or triggered by changes in data. Cloud Functions are a part of serverless approach.

They are executed and fully managed by a cloud provider on its own infrastructure. In other words, it is an instrument that allows you to avoid tons of routine work and resources keeping your code operational: maintaining, scaling and provisioning it.

Cloud Functions have “pay-per-use” payment model that may lead to reduction in costs by raising billing granularity comparing to instance groups.

Cloud Functions are clearly popular and frequently used in production. Data processing, microservices, IoT and small applications and existing SaaS integrations.

But despite their well-known benefits, its usage in Machine Learning is not so widespread. Well, they may be used for pre-processing or post-processing data used by ML models but not so often for running them.

Common Cloud Function vendors such as AWS, Google Cloud, MS Azure provide only a CPU for their execution. When it comes to running heavy models e.g. computer vision or deep learning GPU is slightly faster and way more cost-effective.

DeepMux launched GPU-powered Cloud Functions specially to run ML including heavy deep learning and especially computer vision models.

In the second part of this article, I am going to create web application using deepfake and deploy it within minutes and no worries about Kubeflow or Flask, using only DeepMux Cloud Functions. Stay tuned and subscribe! :) Feedback and comments are strongly appreciated!

In case if you want to start learning DeepMux with yourself, here is a quick start tutorial. Enjoy! :)