What is Tarantella?

Tarantella is an open-source, distributed Deep Learning
framework built on top of TensorFlow 2, providing
scalable Deep Neural Network training on CPU and GPU
compute clusters.

Tarantella is easy-to-use, allows to re-use existing
TensorFlow 2/Keras models, and does not require any
knowledge of parallel computing.

SCALABLE DNN TRAINING

Tarantella allows you to speed up your AI workflows by providing
scalable Deep Neural Network
training on multi-GPU and multi-no de
systems.

USER FRIENDLY

Tarantella comes with a simple, minimalistic API that abstracts away any parallel computing details. It provides a rich technical documentation and tutorials to quickly get started.

EASILY INTEGRABLE

Tarantella supports the full Keras API of TensorFlow 2 and lets you easily integrate Tarantella in your existing workflows.

REPRODUCIBLE RESULTS

Tarantella takes automatic care of the distribution of data and computation in such a way that serial results are reproduced.

OPEN-SOURCE

Tarantella is a community-driven,
open-source framework that builds
on top of TensorFlow 2.

PLATFORM- and VENDOR-
INDEPENDENT

Tarantella supports CPU and GPU
clusters, independently of the
hardware type and vendor.

Speed Up Your DNN Training

Transformer (big) on NVidia V100 GPUs

Tarantella provides strong scalability using data parallelism on models like ResNet-50 and Transformers. It reaches speed-ups of up to 50x on GPU and CPU clusters.

ResNet-50 on NVidia V100 GPUs

ResNet-50 on Intel Skylake CPUs

Enable Tarantella in
Two Simple Steps

Tarantella integrates seamlessly into your existing TensorFlow 2/ Keras model. You only need to add three lines of code

Tarantella provides an easy-to-use command line interface. Run your DNN through Tarantella to start distributed training

Tarantella integrates seamlessly into your existing TensorFlow 2/ Keras model. You only need to add three lines of code


Tarantella provides an
easy-to-use command
line interface. Run your
DNN through Tarantella
to start distributed training

Getting Started

To get you started quickly, download Tarantella from github, check out the installation guidelines, and
the tutorials and documentation.

If you want to contribute to Tarantella, have a look at feature requests, bug reports and the
contributing guidelines.

About us

Tarantella is developed at the Competence Center for High Performance Computing, which is part of the Fraunhofer
Institute for Industrial Mathematics ITWM. In close cooperation with industrial and academic partners,
the Competence Center for High Performance Computing develops solutions for the efficient use of increasingly
more complex processors and parallel computers. Our focus lies particularly in the fields of HPC tools, such as
parallel filesystems and scalable parallel programming, seismic and visualization, Deep Learning tools and
applications, as well as hardware-software co-design and Green by IT.

Peter Labus, Ph.D.

Principal Engineer

Alexandra Carpen-Amarie, Ph.D.

Research Engineer

Martin Kühn, Ph.D.

Research Engineer

TARANTELLA IS DEVELOPED WITHIN THE BMBF-FUNDED PROJECT "HIGH PERFORMANCE DEEP LEARNING FRAMEWORK"

CONTRIBUTORS:

SPONSORS:

Diese E-Mail-Adresse ist vor Spambots geschützt! Zur Anzeige muss JavaScript eingeschaltet sein! | Imprint | Github | Docs