TensorFlow

Last updated 4 months ago

Clusterone supports a variety of different versions of TensorFlow, including the most recent releases. Below we outline how you can choose TensorFlow when creating jobs through the Matrix, as well as the command line.

Do not install TensorFlow as a Python package through a package manager. Including TensorFlow in a requirements file can lead to inconsistencies in the runtime environment and your job can fail. Please select the Docker image with the TensorFlow version you want to use when creating the job.

From the Matrix

In the Environment step of the job creation wizard, select a Tensorflow-based Docker image with the version you want to use.

From the command line

Use just create job distributed or just create job single with the option: --docker-image <tensorflow-image-name> where the image name is one of the following.

Example:

just create job distributed --project <project> \
--datasets <dataset> --docker-image tensorflow-1.9.0-cpu-py36