We are pleased to announce the release of TensorFlow for SX-Aurora. This TF supports Vector Engin in SX-Aurora as a computing device. We have implemented some kernels for VE. Such kenrels are offloaded to VE for acceleration.
We have also released:
- keras includes small modification for VE,
- vetfkernel includes implemetation of kernels for VE, and
- vednn is Vector Engine DNN Library.
You can pip install prebuild packages to start to use TF on SX-Aurora.
These packages were created with python3.6. You should setup virtualenv before installing packages. See README_ve.md for details.
Let’s join us!
We are releasing TF to encourage people to use SX-Aurora for machine learning. The Vector Engine is more general high performance processor than deep learning acclerators such as GPU and TPU. We hope variety of machine learning algorithm will be tried and acclerated on SX-Aurora.
Current TF for SX-Aurora is just a start point. For example, all kernels in TF are not ported to VE (Such kernels are run on CPU). We welcome not only users but also develpers to improve our TF.