利用多智能体编排实现LLM微调自动化[重复].pdf

《利用多智能体编排实现LLM微调自动化[重复].pdf》由会员分享，可在线阅读，更多相关《利用多智能体编排实现LLM微调自动化[重复].pdf（68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.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Sharlina KeshavaShe/herApplied Science Manager,Custom ModelAmazon Web ServicesHuan SongHe/himSenio

2、r Applied ScientistAmazon Web ServicesAutomating LLM Fine-tuning with Multi-Agent OrchestrationA I M 4 2 0-R 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.-Challenges with Fine-tuning Language Models-Multi-Agent Architecture and Automating the Fine-tuning Pipeline-Improving Fine

3、-tuning Peformance with Specialized Agents-Future Work-ConclusionsAgenda 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.4The AWS Generative AI Innovation Center(GenAIIC)aims to help AWS customers accelerate the ad

4、option of Generative AI and take it into production.We are a team of AI scientists and strategists with extensive experience in solving various business problems across industries.AWS Generative AI Innovation Center 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web S

5、ervices,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Challenges with Fine-tuning Language Models 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Optimization areasModels should meet operational requirements,but also d

6、eliver optimal performance and efficiency by balancing three factorsAccuracy Set an accuracy benchmark that aligns with your production criteria123CostConsider which of the models selected for accuracy are the most economicalLatency Choose an efficient model by evaluating on the basis of processing