当前位置:首页 > 报告详情

自定义 Amazon Nova 模型以增强工具调用.pdf

上传人: 明**** 编号:1013653 2025-12-21 18页 325.65KB

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 0Customize Amazon Nova models for enhanced tool callingAnupam DewanSr.Solutions Architect,GenAI,AWSHarsh AsnaniWW Specialist SA-GenAI,AWS 2025,Amazon Web S

2、ervices,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 AIImproving Agent performance via model customization on SageMaker AI 2025,Amazon Web

3、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.Model customization:Progression of options5LEVEL OF CUSTOMIZATION/LEVEL OF EFFORT Fine-tuning(S

4、FT)Model distillation Direct preferenceoptimization(DPO)Reinforcement learning(RL)Continued pre-training(CPT)POST-TRAININGPRE-TRAININGCUSTOMIZING A MODEL Prompt engineering(e.g.,few shot prompting,chain of thought prompting)Retrieval augmentedgeneration(RAG)OPTIMIZING AN FM 2025,Amazon Web Services,

5、Inc.or its affiliates.All rights reserved.Fully managed training jobsFully-managed resilient infrastructure for large-scale and cost-effective trainingAmazon SageMaker HyperPodResilient and self-orchestration infrastructure for maximum resource control Purpose-built infrastructure for FM training 6

6、2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.RIV 2025 Nova Customization Launches 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Amazon Nova 2LiteS E C O N D G E N E R A T I O N M O D E L SE

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。
根据报告的内容,全文主要内容概括如下: - **模型定制化**:介绍了模型定制化的不同层次和关键考虑因素,包括微调、模型蒸馏、直接偏好优化、强化学习和持续预训练。 - **Amazon Nova定制能力**:在Amazon SageMaker AI上引入了针对Amazon Nova的定制能力,支持从预训练到后训练的整个模型训练阶段。 - **Amazon Nova模型系列**:介绍了Amazon Nova的不同版本,包括Nova 2Lite、Nova 2Pro、Nova 2Omni和Nova Frontier,以及Nova Lite、Nova Pro和Nova Micro等第一代模型。 - **Nova Forge优势**:Nova Forge允许用户访问模型开发的各个阶段,使用新的Nova模型,进行强化学习,并利用内置的AI工具包实施安全防护。 - **定制化技术**:详细列出了在Amazon SageMaker AI上可用的定制化技术,如监督微调、直接偏好优化、近端策略优化、持续预训练和知识蒸馏。 - **案例架构**:提供了一个使用合成数据生成器、基础模型、监督微调、强化微调的案例架构,展示了如何通过模型定制化提高工具预测性能。 关键点: - 模型定制化包括微调、模型蒸馏等。 - Amazon Nova支持全阶段的模型训练定制。 - Nova Forge提供早期访问和强化学习功能。 - Amazon SageMaker AI支持多种定制化技术。 - 案例展示了如何通过定制化提高工具性能。
提升AI效率" "SageMaker AI助力模型定制" "Amazon Nova定制,AI新高度"
客服
商务合作
小程序
服务号
折叠