借助 AWS Trainium 突破 AI 性能和成本瓶颈.pdf

《借助 AWS Trainium 突破 AI 性能和成本瓶颈.pdf》由会员分享，可在线阅读，更多相关《借助 AWS Trainium 突破 AI 性能和成本瓶颈.pdf（64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 2 0 1Breakthrough AI Performance and Cost Barriers with AWS TrainiumColin BraceVice President,Neuron Annapurna Labs,AWSJoe RowellFounding Engineer poolsideOria

2、n LeitersdorfChief ScientistDecart AI 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.What will be driving AI in 2026?2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.What will be driving AI in 20

3、26?2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.What will be driving AI in 2026?AGENTIC AI AGENTS IS A TECTONIC SHIFT FOR HOW WE BUILD,DEPLOY AND INTERACT WITH AI 40%of enterprise apps will be integrated with task-specific AI agents by the end of 2026,up from less than 5%todayE

4、xample:CodingGartner,2025Challenge:How to scale to meet the token generation demands of Agentic AI applications?2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Domain Specific Models build for specific applications,private data or personalized to the individual Lower Training Cost

5、Post-training rather than pre-training models can reduce the cost by up to 80%What will be driving AI in 2026?HIGHER DEMAND FOR DOMAIN SPECIFIC MODELS PRODUCED BY MORE EFFICIENT POST TRAINING TECHNIQUESChallenge:Post-Training techniques like RHLF require high performance training and inference 2025,

6、Amazon Web Services,Inc.or its affiliates.All rights reserved.High performance token generationYou will needLow latency token generationEasily accessible,cost-efficient acceleratorsAbility to scale to meet the demand 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.You will needAWS