From the source

Mixture-of-Experts kernel engineering on Amazon SageMaker AI

Optimizing MoE kernels on Hopper GPUs via matrix transposition

Amazon describes kernel optimizations for fused Mixture-of-Experts models on SageMaker AI, focusing on a Swap-AB technique that transposes matrices to improve WGMMA utilization on NVIDIA Hopper GPUs during the decode phase.

From the source
By programmatically flipping the matrix multiplication before it reaches the PTX assembly layer, we realigned the software to match the exact topography of the Hopper silicon.
aws.amazon.com

Most covered

OpenAI GPT-6 and Intelligent UI for everyone
SiliconANGLE: “With GPT-6, ChatGPT’s responses are much more visually appealing and interactive”
Reporting · 33 outlets
OpenAI· 7 Oct
GPT-6 and Intelligent UI for everyone
Meta Muse Wonder: A Frontier Model For Personal & Professional AI Agents
SiliconANGLE: “Meta releases Muse Wonder AI models built to run personal and professional agents”
Reporting · 29 outlets
Meta· 12 Oct
Muse Wonder: A Frontier Model For Personal & Professional AI Agents
Anthropic Introducing Claude Haiku 5.5
Android Authority: “Claude Haiku 5.5 is here: The fastest, most affordable Anthropic model yet”
Reporting · 20 outlets
Anthropic· 7 Oct
Introducing Claude Haiku 5.5
Microsoft Deep dive into Azure Maia 300 architecture
Axios: “Every major Nvidia customer is now a competitor as Microsoft reveals new Maia chip”
Reporting · 14 outlets
Image from azure.microsoft.com
Microsoft· 12 Oct
Deep dive into Azure Maia 300 architecture
Google DeepMind EmbeddingGemma 2: an open, lightweight multimodal embedding model
SiliconANGLE: “Google expands EmbeddingGemma beyond text to images, audio and video”
Reporting · 10 outlets
Google DeepMind· 6 Oct
EmbeddingGemma 2: an open, lightweight multimodal embedding model

Other stories

Other stories

Other stories

Mixture-of-Experts kernel engineering on Amazon SageMaker AI

From the source

Other stories

Natural language security queries at Orca Security using Amazon Bedrock

How ShipStation Global built an AI IT help desk on Amazon Bedrock AgentCore

Build long-horizon agents with Grok 4.7 on Amazon Bedrock

Accessing OpenAI GPT models on Amazon Bedrock with global cross-Region inference in Thailand, Malaysia, Singapore, Indonesia, and Taiwan

Deploy decision models with llama.cpp on Amazon SageMaker AI

Run multi-node inference workloads using Ray LLM Deep Learning Containers