生成式 AI 的下一步是什么？.pdf

《生成式 AI 的下一步是什么？.pdf》由會員分享，可在線閱讀，更多相關《生成式 AI 的下一步是什么？.pdf（25頁珍藏版）》請在三個皮匠報告上搜索。

1、AI Hardware&SystemsaiandsystemsWhats Next In Generative AI?Future Looking Trends In LLM DesignBaskar SridharanVice President,AI/ML Services&InfrastructureAI Hardware&SystemsaiandsystemsGenerative AIAI Hardware&SystemsaiandsystemsThe year of POCsWhat does this mean for my business?What is a foundatio

2、n model?What is a large language model?What is generative AI?Do I need to become a prompt engineer?Is this secure?How do I choose a model?Where do I get started?Which models should we try out?What is FM?AI Hardware&SystemsaiandsystemsThe year of production How do I prioritize my projects?How can we

3、move faster?How do I make this real?How can I lower my costs?How I can I scale this?Which models should I use?How do I manage risks?Should I train my own model?What customization method should I use?AI Hardware&SystemsaiandsystemsNew NFL stats for the 2025 seasonFeatures for each defenderData points

4、 for every NFL gameHistorical NFL data for trainingAI Hardware&SystemsaiandsystemsAI Hardware&Systemsaiandsystemsof 2024 Forbes AI 50 run on AWS96%of all AI/ML unicorns run on AWS AI Hardware&SystemsaiandsystemsGenerative AI stackEC2 Capacity BlocksNeuronUltraClustersNitroGPUsInferentiaTrainiumSageM

5、akerEFAAmazon BedrockGuardrails|Agents|Studio|Customization|Custom Model Import|Amazon ModelsAmazon QAWS App StudioAI Hardware&SystemsaiandsystemsGenerative AI powered by FMs1Pretrained on vast amounts of unstructured dataContain large number of parameters that make them capable of learning complex

6、concepts2Customize FMs using your data for domain specific tasks43Can be applied in a wide range of contextsAI Hardware&SystemsaiandsystemsEC2 Capacity BlocksNeuronUltraClustersNitroGPUsInferentiaTrainiumSageMakerEFAAmazon BedrockGuardrails|Agents|Studio|Customization|Custom Model Import|Amazon Mode

7、lsAmazon QAWS App StudioGenerative AI stackAI Hardware&SystemsaiandsystemsLarge scale FM training challengesInfrastructurestabilityClusters provision&managementStrategies to optimize training performanceGenerative AI inference challengesExpensive GPUs and acceleratorsPerformance lag with user experi

8、ence impactComplex optimizations require months of timeAI Hardware&SystemsaiandsystemsN E WAmazon SageMakerRemove the heavy lifting to scale across thousands of AI accelerators A fully resilient infrastructure purpose-built for foundation model development Optimize utilization of clusters compute,me

9、mory,and network resources between training and inference workloads Elastic Kubernetes Service(EKS)support for HyperPodAI Hardware&SystemsaiandsystemsN E WAmazon SageMakerSpeculative decoding,compilation,and quantizationFully managed across Studio,Jumpstart,SDKs,and AWS command-line interfaceInferen

10、ce Optimization ToolkitAI Hardware&SystemsaiandsystemsEC2 Capacity BlocksNeuronUltraClustersNitroGPUsInferentiaTrainiumSageMakerEFAAmazon BedrockGuardrails|Agents|Studio|Customization|Custom Model Import|Amazon ModelsAmazon QAWS App StudioGenerative AI stackAI Hardware&SystemsaiandsystemsS T A B L E

11、 D I F F U S I O N X L 1.0&3 L A R G ES T A B L E I M A G E U L T R AS T A B L E I M A G E C O R EC O M M A N D R+C O M M A N D RC O M M A N D E M B E DJ U R A S S I C-2J A M B A-I N S T R U C TL L A M A 3.1T I T A N:T E X T,L I T E,E X P R E S ST E X T P R E M I E R EI M A G E G E N E R A T O RE M

12、B E D D I N G S V 2M U L T I M O D A L E M B E D D I N G SC L A U D E 3.5 S O N N E TC L A U D E 3 H A I K U,S O N N E T,O P U S M I S T R A L 7 BM I X T R A L 8 x 7 BM I S T R A L L A R G EM I S T R A L S M A L LC U S T O M M O D E L I M P O R T Leverage your customized models on Amazon BedrockBroa

13、dest selection of modelsAmazon Bedrock AI Hardware&SystemsaiandsystemsNo one model to rule them allAI Hardware&SystemsaiandsystemsEnterprises are deploying models from multiple model providers3%34%41%22%N U M B E R O F L L M P R O V I D E R S U S E D1234AI Hardware&SystemsaiandsystemsAI Hardware&Sys

14、temsaiandsystemsWord filtersTopic filtersHarmful content filtersPII filtersSecurityPrompt injectionDetect and block hallucinationsAmazon Bedrock GuardrailsImplement safeguards customized to your application requirements and responsible AI policiesAI Hardware&SystemsaiandsystemsEC2 Capacity BlocksNeu

15、ronUltraClustersNitroGPUsInferentiaTrainiumSageMakerEFAAmazon BedrockGuardrails|Agents|Studio|Customization|Custom Model Import|Amazon ModelsAmazon QAWS App StudioGenerative AI stackAI Hardware&Systemsaiandsystemsin SageMaker StudioN E WAmazon QTailored,step-by-step recommendations inside your SageM

16、aker Studio notebooksBuild ML models using natural languageAI Hardware&SystemsaiandsystemsAmazon Qin SageMaker Studio1Product guidance and supportCode generation23TroubleshootingAI Hardware&SystemsaiandsystemsThere has never been a better time to be a builderAI Hardware&SystemsaiandsystemsAI Hardware&SystemsaiandsystemsWhat will you build today?