Alphabet Inc (NASDAQ:GOOGL) Google have launched TensorFlow Serving today with the main purpose of making it easier to deploy machine-learning models. TensorFlow Serving is an open source project that aims at helping developers to easily take their machine learning models into production.
The software is available on GitHub and has been optimized for Google’s own TensorFlow machine learning library though Google noted that it can be extended to support other tools.
TensorFlow Serving software has been written in C++ language making it quite convenient for people to deploy their models when using open source tools. It has been optimized for high performance, which includes the ability to handle over 100,000 queries per second when operating on a 16-core Xeon machine. Google further notes that TensorFlow serving can utilize available GPU resource within a machine for the sole purpose of speeding up processing.
Projects like TensorFlow are commonly used for building machine-learning algorithms and training them for particular types of data inputs. TensorFlow has been designed to make these models usable during the production environment. Developers can train their models using TensorFlow and, later on, use TensorFlow Serving’s API to respond to input from a client.
TensorFlow though flexible has a high ability to boost the adoption of Google’s TensorFlow machine learning library. This will stream from great conveniences of using both in machine learning. Google notes that having a system like this means developers can take their models into production faster while also experimenting with different algorithms and models while still having a stable API and stable architecture in place.
Google’s software engineer Noah Fiedel noted in a blog post that TensorFlow serving makes the process of taking models into production faster and easier. It gives developers the chance of deploying new models and running experiments while maintaining their server architecture and API.
Startups have joined big companies the likes of Facebook Inc (NASDAQ:FB) and Google in embracing deep learning. The main reason behind this is the ability to help with speech recognition, image recognition, and natural language processing. Deep learning enables artificial training of neutral networks on large groups of data and later having them make inference about new information. TensorFlow serving software has been designed for the inference phase.