Set Environment Images

Determined launches workloads using Docker containers. By default, workloads execute inside a Determined-provided container that includes common deep learning libraries and frameworks. The default containers can be found on the Determined Docker Hub with tags for each Determined version:

• Default containers for CPU and GPU training

• Containers for TensorFlow CPU and GPU training

By default, Determined will use the tag corresponding to your cluster’s version. To specify a different image from this default, update your job configuration to include:

environment:
  image:
    cpu: # full CPU image path, e.g., determined/tensorflow-ngc:<tag>
    gpu: # full GPU image path, e.g., determined/tensorflow-ngc:<tag>

If one of the images above contain your required libraries, there is no additional environment preparation needed.

If you need to add additional customization to the training environment, review the Customize Your Environment page.

For details on using AMD ROCm-enabled images, including our ROCm 6.1 images with DeepSpeed support for MI300x users, see our AMD ROCm Support documentation.