16.支持RISC-V加速的Ray框架赋能AI工作负载.pdf

《16.支持RISC-V加速的Ray框架赋能AI工作负载.pdf》由会员分享，可在线阅读，更多相关《16.支持RISC-V加速的Ray框架赋能AI工作负载.pdf（14页珍藏版）》请在三个皮匠报告上搜索。

1、EnableRISC-VAcceleratedRayforAIWorkloadsTiejunChenRISC-VInternationalAmbassador#RISCVSummit#RISCVEverywhereRayOverviewEnableRayonRISC-VldemosRayextensionslAccelerateRaylRemoteML/AIWhatslikelynextAgenda#RISCVSummit#RISCVEverywhereRayOverviewIBriefhttps:/ perform any model-specific optimizationsRatSer

2、ve#RISCVSummit#RISCVEverywhereRayOverviewRayServiceRayJobRayClusterKubeRayManagingRayclustersonKubernetes#RISCVSummit#RISCVEverywhereEnableRaytoRISC-VBasicrequirements|ChallengesML frameworks Pytorch,Tensorflow,etcBazel=6.5.0abseil-cpp(RDCYCLE)GolangDocker50+packages pyarrowrelocate-perl#RISCVSummit

3、#RISCVEverywhereEnableRaytoRISC-VDemoIRayTrainonRISC-V#RISCVSummit#RISCVEverywhereEnableRaytoRISC-VDemoIIRayServeonRISC-V#RISCVSummit#RISCVEverywhereRayextensionsAccelerateRayLeveragegraphcompliersInterposerComputationalGraphNativeFrameworkTVMBackendGlowBackendGlowRuntimeVendorBackendVendorRuntimeVe

4、ndorBackendVendorRuntimeML/AIApplicationTOPHALFVendorBackendVendorRuntimeTVMRuntimeRISC-V+AIHWAcceleratorsTOPHALFBOTTOMHALF(s)MLFrameworksComputationalGraphLow-levelLibs/DeviceDrivers#RISCVSummit#RISCVEverywhereRayextensionsRemoteML/AIIEmpowernativeMLframeworksHow-InterceptkeyAPIsatruntimee.gTensorf

5、lowTpile()4.model.fit(model,input,epoch)model.fit(model,input)ray.trainermapRayclusteronRISC-V+AIHWAccelerators#RISCVSummit#RISCVEverywhereRayextensionsRemoteML/AIIIEmpowerdatasciencetools#RISCVSummit#RISCVEverywhereGetintoRaycommunityGetonRaycluster(KubeRay)GetmorecaseslRISC-VhostwithAIhardwareacceleratorsWhatslikelynext?EmpowerAIeverywherewithRISC-V#RISCVSummit#RISCVEverywhereThankyou!Tiejun_Chttps:/