TensorFlow-Serving: Flexible, High-Performance ML Serving
This solves the problem of scalable and efficient ML model serving for organizations like Google, but it is incremental as it builds on existing serving concepts with optimizations.
The paper introduces TensorFlow-Serving, a system designed to serve machine learning models with flexibility and high performance, addressing the need for efficient model deployment and updates in production environments, and it is used in Google's deployments including a multi-tenant hosting service.
We describe TensorFlow-Serving, a system to serve machine learning models inside Google which is also available in the cloud and via open-source. It is extremely flexible in terms of the types of ML platforms it supports, and ways to integrate with systems that convey new models and updated versions from training to serving. At the same time, the core code paths around model lookup and inference have been carefully optimized to avoid performance pitfalls observed in naive implementations. Google uses it in many production deployments, including a multi-tenant model hosting service called TFS^2.