Explore distributed TensorFlow implementation on DC/OS in this conference talk. Learn about the challenges of running distributed TensorFlow on personal infrastructure and discover an open-source framework designed to simplify machine learning with large models and datasets. Gain insights into TensorFlow on Mesos and DC/OS, and witness a live demonstration of the framework. Delve into topics such as machine intelligence, open-source TensorFlow, DC/OS architecture, deep learning overview, and the training phase. Examine the demo setup, visualization, and cluster overview, including DC/OS Catalog and TensorFlow tools. Understand CPU and GPU allocation, beta TensorFlow, deployment phase, and developer workflow. Investigate the challenges of moving to distributed systems, cluster specifications, failure handling, and manual configuration. Learn about high-level service definition, deployer responsibilities, DC/OS Secrets, runtime configuration, and framework types. Conclude with special thanks, questions, and links for further discussion.
Overview
Syllabus
Overview
Introduction to TensorFlow
What is Machine Intelligence
What is TensorFlow
Open Source TensorFlow
What is DCOS
Highlevel picture of DCOS
What are DCOS
Deep Learning Overview
Training Phase
Recap
Demo Overview
Demo Setup
Demo Visualization
Summary
Demo
Cluster Overview
DCOS Catalog
DCOS Tensorflow Tools
DCOS Tensorflow Packages
Storage Bucket
CPUGPU Allocation
Beta Tensorflow
Deployment Phase
Developer Workflow
Moving to Distributed
The Real Challenge
What DCOS Does
Cluster Spec
Dealing with failures
Manual configuration
Highlevel service definition
Deployer responsibilities
Clean Read
DCOS Secrets
Runtime Configuration Dictionary
Single Framework
Meta Framework
Flow
Special Thanks
Questions Links
Discussion
Taught by
Linux Foundation
