engblogs

Efficiently deploying large language models on devices with limited memory using flash memory optimization techniques.

Proposing a method for aligning large language models with human expectations through self-rewarding contrastive prompt distillation.

Analysis of a cross-regional network performance issue uncovers the root cause in a Linux kernel upgrade affecting TCP receive window size calculations

Automate the deployment of an end-to-end RAG solution using Knowledge Bases for Amazon Bedrock and AWS CloudFormation

Insights on 3DS transactions analysis in the US reveal surprising findings that highlight differences in authentication behaviors between regions

Automate software delivery from code commit to production release on Google Cloud using Gitlab CI/CD and Cloud Deploy.

Catalog, query, and search audio programs efficiently using Amazon Transcribe and Knowledge Bases for Amazon Bedrock

Accelerate large language model inference with speculative decoding on AWS Inferentia2

An open source benchmark suite, DCPerf, for hyperscale compute applications aimed at improving hardware and software optimization and platform design.

Building a robust RoCE network for large-scale distributed AI training workloads at Meta

A new generation of African talent leverages cutting-edge AI for scientific challenges through the AI for Science Master’s program at AIMS

Introducing Friend Streak, a social motivation feature for learning together with friends on Duolingo.