AI × AEC · CURATED INSIGHT
Deploying NVIDIA Nemotron 3.5 Super VL on Amazon EKS and SageMaker HyperPod
AWS outlines deployment architectures for NVIDIA's 120B multimodal model Nemotron 3.5 Super VL across Amazon EKS, EKS Auto Mode, and SageMaker HyperPod EKS.
AWS detailed how to deploy NVIDIA Nemotron 3.5 Super VL on Amazon EKS and SageMaker HyperPod EKS. Nemotron 3.5 Super VL is an open multimodal mixture-of-experts model distilled from Nemotron 3 Ultra, featuring 120B total parameters and activating 12B parameters per token. The architecture supports text, image, document, and video inputs, handling context lengths up to 1M tokens.
Key takeaways
- Nemotron 3.5 Super VL pairs a hybrid Mamba-2 and Transformer MoE architecture with a RADIO vision encoder, activating 12B of 120B parameters per token.
- AWS validated deployment patterns across EKS and SageMaker HyperPod using NVIDIA Dynamo and vLLM on instances with B200 and H200 GPUs, offering reference templates via aws-do-eks.
Source：AWS Machine Learning Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human.