Google steps up machine learning production with launch of TensorFlow Serving

Google launched a new open-sourced machine learning software called TensorFlow Service on 16 February. The software allows artificial intelligence (AI) developers to put their machine learning models into production more effectively.

Google's new machine learning software also provides AI developers with the option of running experiments and deploying new models, while retaining the "same server architecture and APIs". Machine learning is an integral part of AI and has powered various Google products like expanding search abilities to powering speech recognition features on Google's app.

Google software engineer Noah Fiedel said in a blog: "While decades of experience have enabled the software industry to establish best practices for building and supporting products, doing so for services based upon machine learning introduces new and interesting challenges. Today, we announce the release of TensorFlow Serving designed to address some of these challenges. TensorFlow Serving is a high performance, open source serving system for machine learning models, designed for production environments and optimized for TensorFlow."

According to Google, TensorFlow Serving has been designed not only to make machine learning model production faster and easier for AI developers, but also to enable them to test out the effectiveness of their models by conducting experiments. Google's software allows developers to perform experiments, using various algorithm programs and models, without having to switch APIs or even severs.

The move is reflective of Google's deep involvement with AI. Given that TensorFlow Serving has also been designed to enhance Google's own TensorFlow machine learning library, it is likely to push Google AI forward and improve its abilities by collecting more data and even identifying new talent. Since the software is written primarily in the most commonly used coding language – C++, instead of Google's own Go, it is highly probable that it will attract more developers to try it out with their machine learning models.

Google says that TensorFlow Serving has been designed to perform at optimum levels and boasts of handling over "100,000 queries per second (QPS) per score" on a 16 core Xeon machine. TensorFlow Serving code and tutorials are available on Github.