在 Amazon SageMaker AI 中自定义 Amazon Nova 以满足您的业务需求.pdf

《在 Amazon SageMaker AI 中自定义 Amazon Nova 以满足您的业务需求.pdf》由会员分享，可在线阅读，更多相关《在 Amazon SageMaker AI 中自定义 Amazon Nova 以满足您的业务需求.pdf（19页珍藏版）》请在三个皮匠报告上搜索。

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.A I M 3 8 5Customize Amazon Nova in Amazon SageMaker AI for your business needsAnupam DewanSr.Solutions Architect,GenAI,AmazonMona MonaSr Gen AI Specialist Solutions

2、 Architect,AWS 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AgendaWhat is model customizationKey considerations for model customization and training Introducing customization capabilities for Amazon Nova on Amazon SageMaker AIChoosing the right customization technique 2025,Amaz

3、on Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Model customization 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Improved RelevanceEnhanced GenerationQualityIncrease AccuracyDomain Customization 2025,A

4、mazon Web Services,Inc.or its affiliates.All rights reserved.Retrieval Augmented Generation(RAG)In context learningAugmenting prompts with context retrieved from large domain-specific databasesUsing an FM with few/no additional promptsSupervised Fine-Tuning(SFT)Continued pre trainingFine-tuningAlter

5、s model weights/parametersIn-Context Learning Does not alter model weights/parametersPre trainingPrompt EngineeringDesign prompts to make the outputs accurate,relevant to adapt model for your needs Parameter efficient fine tuning(PEFT)Post TrainingModel customization:Progression of optionsPreference

6、 alignmentDirect Preference Optimization(DPO)Proximal Policy Optimization(PP0)RFT(Reinforcement fine tuning)Low Rank Adaptation(LoRA)2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Key considerations for model cust