当前位置:首页 > 报告详情

大模型时代的搜推广应用和实践.pdf

上传人: 2*** 编号:149661 2023-12-18 26页 28.63MB

1、LLM Era on Cloud Innovation!#$%&()*+,-./01234567Offline+Online Architecture!AI!#$%+AI&(!)*+,-./0Engineering!1234*!FeatureStore!EasyRec56783!9:+;?!-.ABCDEFAI-AI-PAI-ECS/ACKCPUGPURDMACPFSOSSNASODPSOpenSearchPAI-RECPAI-UG&-DashScope-AIPaaS-DataWorksPAI-DSWPAI-DesignerPAI-iTAGMaxComputeHologresFlinkPAI-

2、DLCEMRElasticSearchDLFPAI-EASPAI-BladeEPLTorchAccDataSetAccDataHubAIFeatureStoreAIAIODPS&AI&-AIPaaS-DataWorksMaxComputeHologresFlinkEMRElasticSearchDLFDataHubAIPAI-DSWPAI-DesignerPAI-iTAGPAI-DLCPAI-EASPAI-BladeEPLTorchAccDataSetAccFeatureStoreODPS&AI-OpenSearchPAI-RECPAI-UG&-AIPaaSMaxComputeHologres

3、FlinkEMRElasticSearchDLFDataHubAIAI&AI-OpenSearchPAI-RECPAI-UGPAI-DSWPAI-DesignerPAI-iTAGPAI-DLCPAI-EASPAI-BladeEPLTorchAccDataSetAccFeatureStoreAIPaaSMaxComputeHologresFlinkEMRElasticSearchDLFDataHub&AIOpenSearchPAI-RECPAI-UG&-DashScope-PAI-DSWPAI-DesignerPAI-iTAGPAI-DLCPAI-EASPAI-BladeEPLTorchAccD

4、ataSetAccFeatureStorePAI-REC-Useri2iCate2iItem/User/ItemItemUserGDBT/FM/LRDeepFM/DSSM/ESSMWord2vec/BertItem EmbeddingDBMTL/FM/DeepFM(+query)App/Web+ABAIEasyRecPAI-FeatureStore?AI?embedding?80%?PAI-FeatureStore(FS)OfflineStore(MaxCompute)OfflineStore(Hologres/GraphCompute/TableStore)?meta?ProjectJoin

5、 ID?User/Item?Label?TrainSet?User/Item?FS SDK?User/Item Feature EntityUser/Item FeatureViewLable?ModelFeature?EasyRec SOTAEasyRechttps:/ LossPairWise LossAUCAccuracyMSEHitPateComputingPlatformMaxComputeDLCDSWFlinkEMR-DataScienceAutoML-HPOAuto FeatureDist-EvaluateFeature SelectionExportEarly StopDist

6、illEasyRec EstimatorTraining with EvalossMaxComputeHDFSHiveDataHubKafkaFeatureGenerationDeepModelsUser_id:stringage:integerPriece:floatID FeatureRaw FeatureSeq FeatureDeepFMMultiTowerDINOptimizerTensorboardAdamOptimizerAdaGradAdaDeltaFG(MatchFeature)inferenceHPOnum epochposition

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
本文主要介绍了PAI-FeatureStore的架构和功能,以及其在OfflineStore(MaxCompute)和DeepModels中的应用。PAI-FeatureStore提供了特征工程和模型训练的平台,支持多种计算引擎,如MaxCompute、Hologres、GraphCompute和TableStore。它还提供了自动特征分布评估、特征选择和早期停止等优化功能。在DeepModels方面,文章提到了多种深度学习模型,如DIN、DSSM和MINDRank,以及它们在推荐系统中的应用。此外,文章还简要介绍了PAI-REC和PAI-EAS等产品的特点和应用场景。
"PAI-FeatureStore如何提高模型训练效率?" "PAI-REC中的Embedding技术如何优化推荐效果?" "LLM在PAI-EAS中的应用场景有哪些?"
客服
商务合作
小程序
服务号
折叠