Choose a container image

This page helps you choose which container image to use.

Choose a container image type

Deep Learning Containers supports each framework version based on a schedule to minimize security vulnerabilities. Review the Deep Learning Containers framework support policy to understand the implications of the end-of-support and end-of-availability dates.

Each container image provides a Python 3 environment and includes the selected data science framework (such as PyTorch or TensorFlow), Conda, the NVIDIA stack for GPU images (CUDA, cuDNN, NCCL2), and many other supporting packages and tools. To find the appropriate container image, see the tables below.

Base versions

TensorFlow versions

PyTorch versions

Hugging Face container images

This section lists the available Hugging Face Deep Learning Containers.

Text Generation Inference containers

Text Embeddings Inference containers

PyTorch Inference containers

PyTorch training containers

Model Garden container images

This section lists the available Model Garden container images.

vLLM inference containers

Experimental image families

Image families that aren't listed here are experimental. Experimental image families are supported on a best-effort basis, and don't receive refreshes on every new release of the framework.

Listing all available versions

If you need a specific framework or CUDA version, search the complete list of available container images. To list all available Deep Learning Containers images, use the following command in the Google Cloud CLI with your preferred terminal or in Cloud Shell.

gcloud container images list --repository="us-docker.pkg.dev/deeplearning-platform-release/gcr.io"

Using locally

Deep Learning Containers can be pulled and used locally. To do so, see Getting started with a local deep learning container.

What's next

• Read the Deep Learning Containers overview to learn more about what is pre-installed on container images.
• Get started with Deep Learning Containers by walking through the How-to guides, which provide instructions on how to build and push deep learning container images.