《3322 - 企业私有人工智能的模型即服务模式.pdf》由会员分享,可在线阅读,更多相关《3322 - 企业私有人工智能的模型即服务模式.pdf(101页珍藏版)》请在三个皮匠报告上搜索。
1、How to become the hero of your AI storyA Model-as-a-Service pattern for Private AI in the EnterpriseErwan Granger,Director,AI Customer Adoption and Innovation,AI BU,Red HatJennifer Vargas,Senior Principal Product Marketing Manager,AI BU,Red Hat1Agenda2What well discuss today Chapter 1:An origin stor
2、y Chapter 2:A coder emerges Chapter 3:All for AI,and AI for all!Chapter 4:The accountant Chapter 5:Epilogue3Optional section marker or titleModel-as-a-Service Chapter 1:An origin story Chapter 2:A coder emerges Chapter 3:All for AI,and AI for all!Chapter 4:The accountant Chapter 5:Epilogue4Model-as-
3、a-ServiceMaaS pronounced Mazz5Model-as-a-ServiceMaaS pronounced MazzMass6Model-as-a-ServiceMaaS pronounced MazzMassMass7Model-as-a-ServiceMaaS pronounced MazzMass8How it all started 9How it all started 10How it all started 11How it all started 12And soon enough 13GPU-Powered Nodes!14Alls well that e
4、nds well?15Alls well that ends well?16Alls well that ends well?17Alls well that ends well?18Alls well that ends well?In OpenShift AI:19And so quickly enough 20And so quickly enough 21And so quickly enough 22And so quickly enough 23And so quickly enough 24And so quickly enough 25Because,well 26No goo
5、d deed goes unpunished27No good deed goes unpunished28No good deed goes unpunished29No good deed goes unpunished30No good deed goes unpunished31No good deed goes unpunished32No good deed goes unpunished33No good deed goes unpunished34No good deed goes unpunished35No good deed goes unpunished36Time f
6、or a new strategy37Time for a new strategy38Ideas discarded along the wayIdea:Why not?Again,19K RH employees.Even 1 GPU/Employee is problematic19K RH employees What should new Max be?Increase Max on Auto-ScalingSet quotas in OpenShift to prevent GPUs overuseUse MIG to split whole GPUs in smaller sli