github.com/sentienttechnologies/studio-go-runner@v0.0.0-20201118202441-6d21f2ced8ee/docs/slides/2018-Sentient-DAI-platform.html (about) 1 <!doctype html> 2 <html> 3 <head> 4 <meta charset="utf-8"> 5 6 <title>Studio.ml Cloud Neutral Platform - 2018</title> 7 8 <meta name="description" content="The 2018 mission for Sentients DAI Platform"> 9 <meta name="author" content="karl Mutch"> 10 11 <meta name="apple-mobile-web-app-capable" content="yes"> 12 <meta name="apple-mobile-web-app-status-bar-style" content="black-translucent"> 13 14 <meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no"> 15 16 <link rel="stylesheet" href="css/reveal.css"> 17 <link rel="stylesheet" href="css/theme/black.css" id="theme"> 18 19 <!-- Theme used for syntax highlighting of code --> 20 <link rel="stylesheet" href="lib/css/zenburn.css"> 21 22 <!-- Printing and PDF exports --> 23 <script> 24 var link = document.createElement( 'link' ); 25 link.rel = 'stylesheet'; 26 link.type = 'text/css'; 27 link.href = window.location.search.match( /print-pdf/gi ) ? 'css/print/pdf.css' : 'css/print/paper.css'; 28 document.getElementsByTagName( 'head' )[0].appendChild( link ); 29 </script> 30 </head> 31 32 <body> 33 <div class="reveal"> 34 <div class="slides"> 35 <section data-transition="convex"> 36 <h3>Common Platform Roadmap - 2018</h3> 37 <br> <br> 38 <h5>2 Solution Types</h5> 39 <ul style="font-size: 66%"> 40 <li>Solutions for research (StudioML)</li> 41 <li>Solutions for applications (ENN Services)</li> 42 </ul> 43 </section> 44 <section data-transition="convex"> 45 <h5>Existing services</h5> 46 <ul style="font-size: 66%"> 47 <li>Evolution or Revolution ? Blog - evolutionary optimization (Research)</li> 48 <li>LEAF cloud deployed service(s)</li> 49 <li>Smaller clusters of dedicated services</li> 50 </ul> 51 <br> <br> 52 <h5>Existing research</h5> 53 <ul style="font-size: 66%"> 54 <li>StudioML orchestrates python experiments in research and product settings</li> 55 <br> 56 <li>StudioML runners reproducing AWS ENN experiment results on Azure clusters</li> 57 <li>Experimenters tune, test and improve the training performance of multiple experiments</li> 58 <li>Typical 16,800 GPU hours per experiment, clusters of NVIDIA P40 machines</li> 59 <br> 60 <li>AWS, Azure, Workstation Kubernetes, and standalone</li> 61 </ul> 62 </section> 63 <!-- 64 <section data-transition="convex"> 65 <h5>Services</h5> 66 <ul style="font-size: 66%"> 67 <li>Ramp to rich service meshes was bumpy, but now we know how</li> 68 <li>Used for both the Blog website and LEAF</li> 69 <li>Platform cloud neutral, also services, unless vendor APIs used</li> 70 </ul> 71 <br> <br> 72 <h5>Research</h5> 73 <ul style="font-size: 66%"> 74 <li>Cross platform support works, limited migration relatively painless</li> 75 <li>Azure ease of use was on par with AWS</li> 76 <li>Hybrid cloud services work</li> 77 </ul> 78 </section> 79 --> 80 <section data-transition="convex"> 81 <h5>Motivation for Cloud Agnostic StudioML</h5> 82 <br> 83 <body style="font-size: 66%">Diverse & growing list of deployment configurations and business demands</body> 84 <br><br> 85 <ul style="font-size: 66%"> 86 <li>StudioML is worthwhile and useful</li> 87 <li>StudioML has multi-cloud capability now</li> 88 </ul> 89 <br> <br> 90 <h4>Downside</h4> 91 <ul style="font-size: 66%"> 92 <li>linear development cost for each deployment type</li> 93 <li>testing cost/complexity slope</li> 94 <li>not so adaptable/flexible to business changes</li> 95 </ul> 96 </section> 97 <section data-transition="convex"> 98 <h3>Why now</h3> 99 <br> 100 <ul style="font-size: 66%"> 101 <li>recent use cases, both informed and posed challenges</li> 102 <li>Need to establish StudioMLs raison d'être as a part of the Platform</li> 103 </ul> 104 <br> <br> 105 <ul style="font-size: 66%"> 106 <li>Fully broken out software stack, using services</li> 107 <li>Both Partner and Sentient IP can be deployed into walled gardens</li> 108 <li>Infrastructure must be either Cloud Neutral, or Azure, and be secured</li> 109 <li>3rd Party software widely used</li> 110 <li>The need to support long running jobs</li> 111 </ul> 112 </section> 113 <section data-transition="convex"> 114 <h4>Cloud Neutral StudioML</h4> 115 <ul style="font-size: 66%"> 116 <li>Encourage StudioML to gain increased independence, community governance for pypi for example</li> 117 <li>Fix https://github.com/studioml/studio/community</li> 118 </ul> 119 <br> <br> 120 <h4>Decouple us from technology choices made for the entire StudioML audience</h4> 121 <br> 122 <ul style="font-size: 66%"> 123 <li>provide one technology stack that works for organizations, align to LEAF/ENN services platform</li> 124 <li>encourage the StudioML community to deal with use cases for individuals</li> 125 <li>limit future Sentient StudioML work to cloud agnostic use cases</li> 126 </ul> 127 </section> 128 <section data-transition="convex"> 129 <h3>Whats in it for the community</h3> 130 <br> 131 <ul style="font-size: 66%"> 132 <li>keep focus on use cases that are common</li> 133 <li>small skill set needed for researchers to do AI at scale</li> 134 <li>a single reference solution, avoids make it work everywhere, time and $</li> 135 <li>attack cost of deployment, and ownership</li> 136 </ul> 137 <br> <br> 138 <h3>Whats in it for Sentient</h3> 139 <br> 140 <ul style="font-size: 66%"> 141 <li>private infrastructure use cases for govt etc</li> 142 </ul> 143 </section> 144 </div> 145 </div> 146 147 <script src="lib/js/head.min.js"></script> 148 <script src="js/reveal.js"></script> 149 150 <script> 151 // More info about config & dependencies: 152 // - https://github.com/hakimel/reveal.js#configuration 153 // - https://github.com/hakimel/reveal.js#dependencies 154 Reveal.initialize({ 155 dependencies: [ 156 { src: 'plugin/markdown/marked.js' }, 157 { src: 'plugin/markdown/markdown.js' }, 158 { src: 'plugin/notes/notes.js', async: true }, 159 { src: 'plugin/highlight/highlight.js', async: true, callback: function() { hljs.initHighlightingOnLoad(); } } 160 ] 161 }); 162 </script> 163 </body> 164 </html>