Abstract: In this workshop, we will optimize, deploy, and scale various Tensorflow Models in a distributed, hybrid-cloud production environment. We will use 100% open source tools including Tensorflow, Jupyter Notebook, Docker, Kubernetes, Prometheus, Grafana, and NetflixOSS.First, we will focus on post-training model optimization techniques such as batch normalization folding, weight rounding, and 8-bit quantization. Next, we will explore Tensorflow’s XLA JIT and AOT static compiler features. Last, we will serve, load-test, and monitor our optimized models in a highly-tuned Tensorflow Serving and NetflixOSS Microservice runtime environment.All code and Docker Images are 100% open source. Links to GitHub and DockerHub are available at http://pipeline.io.
Bio: Chris Fregly is a Research Scientist at PipelineIO - a Machine Learning and Artificial Intelligence Startup in San Francisco. Chris is an Apache Spark Contributor, Netflix Open Source Committer, Founder of Advanced Spark and TensorFlow Meetup, and Author of the upcoming O'Reilly Video Series, "High Performance Distributed Tensorflow in Production."Previously, Chris was a Distributed Systems Engineer at Netflix, Data Solutions Engineer at Databricks, and a Founding Member of the IBM Spark Technology Center in San Francisco.