View a markdown version of this page

Amazon Deep Learning Containers in Amazon Web Services China - Getting Started with Amazon Web Services China
Region availabilityHow Amazon Deep Learning Containers differsDocumentation
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Amazon Deep Learning Containers in Amazon Web Services China

Amazon Deep Learning Containers (Deep Learning Containers) are a set of Docker images for training and serving models in TensorFlow, TensorFlow 2, PyTorch, and MXNet. Deep Learning Containers provide optimized environments with TensorFlow and MXNet, Nvidia CUDA (for GPU instances), and Intel MKL (for CPU instances) libraries and are available in the Amazon Elastic Container Registry (Amazon ECR).

Region availability

Amazon Deep Learning Containers is available in the following regions in China:

  • China (Beijing) Region

  • China (Ningxia) Region

How Amazon Deep Learning Containers differs

The following differences apply to Amazon Deep Learning Containers:

  • Elastic Inference Deep Learning Container images are not available in the China (Beijing) and China (Ningxia) Regions.

  • Deep Learning Containers images are published to private Amazon ECR registries. The registry account differs by Region and image type. Use the following information to authenticate to the applicable registry and construct image URIs:

    • Registry accounts:

      • Deep Learning Containers framework images (for example, base and vLLM):

        • China (Beijing) Region - 727897471807.dkr.ecr.cn-north-1.amazonaws.com.cn

        • China (Ningxia) Region - 727897471807.dkr.ecr.cn-northwest-1.amazonaws.com.cn

      • sagemaker-scikit-learn and sagemaker-xgboost:

        • China (Beijing) Region - 450853457545.dkr.ecr.cn-north-1.amazonaws.com.cn

        • China (Ningxia) Region - 451049120500.dkr.ecr.cn-northwest-1.amazonaws.com.cn

    • ECR login - authenticate to the applicable account from the preceding list. For example, for the framework account in the China (Beijing) Region:

      aws ecr get-login-password --region cn-north-1 | docker login --username AWS --password-stdin 727897471807.dkr.ecr.cn-north-1.amazonaws.com.cn

    • Image URI format:

      <ACCOUNT>.dkr.ecr.<REGION>.amazonaws.com.cn/<REPOSITORY>:<TAG>

      For example, the sagemaker-xgboost repository with the tag 3.0-5 in the China (Beijing) Region:

      450853457545.dkr.ecr.cn-north-1.amazonaws.com.cn/sagemaker-xgboost:3.0-5

    • Available tags - only mutable tags are available in these Regions. A mutable tag can be repointed to a newer image when a version is patched or updated, so pulling the same tag at a later date may return different image content. Immutable tags, which pin to a fixed, unchanging build, are generally not published, apart from a few exceptions required by specific consumers (for example, the vLLM -v1.x SageMaker AI tags). To illustrate the mutable-tag scheme:

      • Ubuntu-based vLLM - available tags: 0.25.1-gpu-py312-cu130-ubuntu22.04-ec2, 0.25.1-gpu-py312-ec2. Immutable tags such as 0.25.1-gpu-py312-cu130-ubuntu22.04-ec2-v1.2-2026-07-20-21-30-05 and 0.25-gpu-py312-cu130-ubuntu22.04-ec2-v1 are not available.

      • AL2023-based vLLM - available tags: server-cuda-v2.1, server-cuda-v2, server-cuda. Immutable tags such as server-cuda-v2.1.3 are not available.

      • For the full, current list of tags in a repository, use the standard Amazon ECR CLI commands (aws ecr list-images / aws ecr describe-images) against the applicable Region.

    • Not all framework versions are actively patched. Versions follow the standard DLC Support Policy - use a currently supported version to ensure you continue receiving security patches.

Documentation