Customize Training Task Docker images

Brief introduction

Xiaomi Cloud-ML supports the Bring Your Own Image feature, allowing users to import custom Docker images to initiate training tasks. This can solve the problem of training tasks that rely on non-Python libraries.

Note that user-supplied image addresses should be publicly accessible. If you download images and launch tasks in DockerHub, it takes a long time.

Usage example

When submitting a task, the user merely specifies the Docker image address.

cloudml jobs submit -n linear -m trainer.task -u fds://cloud-ml/linear/trainer-1.0.tar.gz -d cloudml/tensorflow:1.0.0

Parameter introduction

• -d indicates the Docker image address specified by the user. Note that it cannot be used simultaneously with the -F and -V parameters.