Skip to content
AWS Logo AWS Logo

AWS Deep Learning Containers

Pre-built Docker images for running AI/ML workloads on AWS.

Tested for performance and patched for security vulnerabilities.


What are DLCs?

AWS Deep Learning Containers (DLCs) are Docker images pre-configured with deep learning frameworks and tools. AWS builds, tests, and security-patches them so you can focus on your workload instead of environment setup.

Each image includes a framework (e.g. vLLM, PyTorch, Ray), its dependencies, and optimized libraries — ready to run on AWS compute services. All DLC images are provided at no cost — you only pay for the compute resources you use.

Getting Started

It's easy to get started. For example, to run a large language model server:

docker pull public.ecr.aws/deep-learning-containers/vllm:server-cuda
docker run --gpus all -p 8000:8000 \
  public.ecr.aws/deep-learning-containers/vllm:server-cuda \
  --model openai/gpt-oss-20b

Query the server:

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "openai/gpt-oss-20b", "messages": [{"role": "user", "content": "Hello!"}]}'

LLM serving is just one example. DLCs cover a range of AI/ML workloads — explore guides by use case:

  • Serve Large Language Models


    Deploy large language models with vLLM, SGLang, or Ray LLM on EC2, EKS, or Amazon SageMaker AI.

    vLLM Guide · SGLang Guide · Ray LLM Guide

  • Serve Multimodal Models


    Serve TTS, image generation, video generation, and omni-chat models with vLLM-Omni.

    vLLM-Omni Guide

  • Transcribe Speech


    Transcribe, align, and diarize audio with WhisperX on EC2 or Amazon SageMaker AI.

    WhisperX Guide

  • Serve GGUF Models on CPU or GPU


    Serve quantized GGUF models with llama.cpp on x86 CPU, NVIDIA GPU, or Graviton (ARM64) — EC2 or Amazon SageMaker AI.

    llama.cpp Guide

  • Answer Decision Questions


    Serve strands-decider models, which answer yes/no, choice, and score questions with calibrated probabilities, on Amazon SageMaker AI.

    Strands Decider Guide

  • Disaggregated LLM Serving


    Run NVIDIA Dynamo with vLLM, SGLang, or TensorRT-LLM, prefill and decode on separate GPU nodes over EFA, on Amazon SageMaker HyperPod.

    NVIDIA Dynamo Guide

  • Serve ML Models


    Deploy any ML model — NLP, vision, audio, and tabular — with Ray Serve on EC2 or Amazon SageMaker AI, or TensorFlow SavedModels with TensorFlow Serving on Amazon SageMaker AI.

    Ray Serve Guide · TensorFlow Inference Guide

  • Train Models


    Run distributed training with PyTorch, TensorFlow, or Ray Train on GPU or CPU, with EFA and NCCL pre-installed for multi-node workloads.

    PyTorch Guide · TensorFlow Training Guide · Ray Train Guide

  • Run GPU Functions on Lambda


    Build GPU-accelerated Lambda functions with CUDA, CuPy, PyTorch, vLLM, or SGLang on Lambda Managed Instances.

    Lambda GPU Guide

  • Build Your Own Image


    Use the lightweight Base images (CUDA + Python on Amazon Linux 2023) as the FROM for your custom AI/ML container.

    Base Guide

For step-by-step walkthroughs on EKS, SageMaker, and more, see our blog posts and tutorials. You can also subscribe to release notifications to stay up to date with new images.

Looking for something else? Let us know on GitHub.


GitHub · ECR Gallery · Support Policy · Security