Wiss. Rechnen » Tensorflow

Caution: this page contains information about the OMNI cluster! This software is not available on the HoRUS cluster.

Tensorflow is a software library for machine learning. It is mostly developed by Google, however it is available as free and open-source software under the Apache license. The user interface is written in Python and usually operated in the form of Jupyter notebooks.

Caution: currently, Jupyter is not yet available on the OMNI cluster due to technical problems. We will notify you when Jupyter is usable.

On the cluster, Tensorflow version 2.2.0 will be available.

The learning portal of Tensorflow, which contains documentation and tutorials, can be found here.

Using Tensorflow

On the cluster, Tensorflow will be available as part of the Jupyter installation. You therefore need to use Tensorflow from a Jupyter notebook with the corresponding Tensorflow kernel. Getting onto our Jupyter portal and more information on notebooks and kernels can be found on our Jupyter page. When selecting the kernel, you need to use “Tensorflow 2 via SLURM”.

Tip: the kernel selection dialog appears automatically when you create a new notebook (for example when you select “File”->“New”->“Notebook” in the JupyterLab interface). For an existing notebook, you can select the kernel by clicking the kernel name in the upper right. You have further control over the kernel via the “Kernel” menu and the tab “Kernel Sessions” in the left side bar.

Tensorflow kernels are currently launched on the GPU nodes, see also our GPU page. That also means that they are launched with the default parameters for that queue (1 GPU node, 12 hours runtime). We reserve the right to modify this as we collect experience with Jupyter, Tensorflow and their usage on the cluster.

Aktualisiert um 16:11 am 25. February 2021 von Antonia Vitt