人工智能经济学：在保护资金的同时实现价值最大化.pdf

《人工智能经济学：在保护资金的同时实现价值最大化.pdf》由会员分享，可在线阅读，更多相关《人工智能经济学：在保护资金的同时实现价值最大化.pdf（15页珍藏版）》请在三个皮匠报告上搜索。

1、W P S 3 0 5Sanjeev PulapakaPrincipal Solutions Architect Amazon Web ServicesSujatha DantuluriSenior Solutions ArchitectAmazon Web ServicesGenerative AI economics:Maximizing value to protect the public purse 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AgendaIntroduction Amazon

2、Bedrock pricing basicsCost optimization techniquesInference Prompts VectorsAgents 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Teacher modelStudent modelDistilled modelSynthetic dataModel distillationUsersFront-end appPrompt optimizationAmazon BedrockEmbedding modelS3 VectorsPu

3、rpose-built agent 1Supervisor agentStrandsPurpose-built agent 2CalculatorCacheCacheCachePass knowledgeSimpleComplexPrompt routingSonnetHaikuProcessed intentTool callingInstructionsDocumentsUser promptPrompt 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Amazon Bedrock pricing bas

4、icsKey conceptsExamplePricing Price is calculated on tokens and differs from models Charged for every input token processed and every output token generated750 words =1,000 tokensClaude 3.5 Haiku example:Input:$0.0008/K|Output:$0.004/KOctank Insurance needs to summarize 50 calls per minute during 8-

5、hour workdays Input:25K tokens per call transcript Output:1K token summary Processing:24,000 calls daily Claude 3.5 Haiku(PDX)Whats the estimated cost?2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Efficiently run model inference on large volumes of data while avoid throttling50%

6、lower priceWell suited for use cases to process substantial volumes of dataEnhanced security and operational featuresBatch Inference Run multiple inference requests asynchronously with much lower cost 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Model distillationTeacher model