Earlier this month Google's TensorFlow 1.0 was announced and dominated the sessions at the TensorFlow Dev Summit 2017. It's amazing to see how far the TensorFlow environment has come since its initial release a little over a year ago. If you missed the #TFdevsummit it's not too late, check out the replay here => https://goo.gl/0sySqI. As the TensorFlow ecosystem continues to grow, we are bringing these features to PowerAI to give Deep Learning developers the highest-performance and most cost-effective turnkey system.
Building on our relationship with Google as a founding member of the OpenPOWER foundation and Google’s focus on building an open innovative data center ecosystem, we have worked closely with Google by contributing code back so this release is enabled for POWER from the start and we’re able to launch a PowerAI release of TensorFlow 1.0 concurrently with its announcement. It was exciting for us to participate in this event, with our collaboration on POWER enablement for TensorFlow prominently featured during the developer summit.
With release 3.3 of PowerAI, developers have access to a complete TensorFlow environment that includes:
- TensorFlow 1.0 is the newest release for TensorFlow in PowerAI. TF 1.0 includes many enhancements to TensorFlow functions and APIs. For those looking for TF 0.12 model compatibility, PowerAI will also include the TF 0.12 environment. In addition, Google publishes tools to upgrade TF models to the new TF 1.0 APIs.
- Bazel 0.4 is the TensorFlow build environment and is used to build many TensorFlow models.
- Tensorboard provides a powerful visualization of Deep Learning data sets and training results in TensorFlow.
- TensorFlow Debugger (tfdbg) provides a debug tool for TensorFlow models and training.
- XLA (Accelerated Linear Algebra) is a new experimental feature that provides a just-in-time compiler to accelerate TensorFlow models by translating TensorFlow graphs to native code. Using XLA, users can configure TensorFlow to generate high-performance native Open POWER code to take advantage of the advanced features found in POWER8™. This feature is still considered to be an alpha release by the TensorFlow developers. We have enabled this feature in PowerAI so that developers can become familiar with the newest, exciting features available in the TensorFlow world.
TensorFlow offers interfaces to a variety of programming environments. A popular environment for building and using TensorFlow models is Python. Ubuntu for POWER8 includes a complete Python environment that supports TensorFlow models written in Python. Install the following and more Python features with the Python installer pip to build TensorFlow models in Python:
- NumPy provides a Linear Algebra environment that serves as the foundation for Deep Learning environments in Python.
- Keras offers a high-level API for building TensorFlow (and Theano) models in Python.
- Jupyter provides an interactive notebook environment for Deep Learning developers and data scientists to prototype new Deep Learning use cases.
Get started with PowerAI to develop cognitive applications on Power, and now, you can accelerate deep learning with the power of IBM POWER8 and NVIDIA GPUs! Share how you are unleashing the power of deep learning to transform the future of computing in the comments section.
Dr. Michael Gschwind is Chief Engineer for Machine Learning and Deep Learning for IBM Systems where he leads the development of hardware/software integrated products for cognitive computing. During his career, Dr. Gschwind has been a technical leader for IBM’s key transformational initiatives, leading the development of the OpenPOWER Hardware Architecture as well as the software interfaces of the OpenPOWER Software Ecosystem. In previous assignments, he was a chief architect for Blue Gene, POWER8, POWER7, and Cell BE. Dr. Gschwind is a Fellow of the IEEE, an IBM Master Inventor and a Member of the IBM Academy of Technology.