加速智能发展——释放实时人工智能的力量.pdf

《加速智能发展——释放实时人工智能的力量.pdf》由会员分享，可在线阅读，更多相关《加速智能发展——释放实时人工智能的力量.pdf（28页珍藏版）》请在三个皮匠报告上搜索。

1、Accelerating Intelligence:Unlocking the Power of Realtime AIBoyuan HuangProduct Director of Alibaba Cloud PAIBoyuan HuangProduct Director of Alibaba Cloud PAIProduct Director at Alibaba Cloud Intelligent Group Head of Alibaba Clouds Artificial Intelligence Platform PAI,Big Data Platform DataWorks,an

2、d AI Search Products(OpenSearch&Elasticsearch)15 years of experience in technology and products in the fields of big data and AI,previously worked at Alibaba Groups Advertising,Search/Recommendation teams;Microsoft Advertising Technology Center,Search Technology Center.IntroThe Evolving AI Landscape

3、&Alibaba Cloud AI infrastructure01Key Trends of AI and InfraEvolving Trends of AIFocus of AI InfraPlatform for AI at a GlanceModel building platformAI Workflow Scheduling(PAIFlow)Interactive Dev(JupyterLab,WebIDE,Copilot)Visual Dev(200+algorithms,Data Warehouse Connection）FeatureStoreAutoMLScenarios

4、 best practices(LLM,SD,RAG,Search)Computing IaaS(Lingjun,ECS/EGS/ECI,ACK/ASI/ACS;MaxCompute,EMR,Flink）Computing Resource ManagmentResource Group,Resource QuotaAI Assets ManagementDataset,Feature,Image,Model,CodeAI WorkspaceUser Role,Security Config,Computing Resource BonddingModel Training ServiceMo

5、del Inference ServiceAutomatic Fault-Tolerant Training Engine(AIMaster)Fast recovery training(EasyCKPT)Large-scale RLHF training framework(ChatLearn)Distributed Training Acceleration Engine(TorchAcc，DeepSpeed，Megatron)Automatic Health Check(SanityCheck)Elastic Inference Service(Load sensing,Schedule

6、d elasticity,Elastic resource pool)Inference optimization(Blade)Service orchestration and management(Service Grouping,AB Test,Automated pressure testing)One-click model deployment(EAS minimalist mode,scenario-based)Support for heterogeneous service load(real-time inference,asynchronous inference,off