We are happy to announce the release of SynapseAI version 1.0.0. In addition to many improvements and bug fixes, the new version includes the following.
We added support for TensorFlow version 2.5.0. We further simplified the code needed for initializing TensorFlow on Gaudi. Extended support for TensorFlow operators will provide even better acceleration. This release also introduces support for tf.keras.mixed_precision for Gaudi.
TensorBoard graph visualization is now supported for models trained with Keras or Estimator.
Also, we added 7 new TensorFlow reference topologies, including Unet3D, CycleGan, T5-base, DenseNet, Transformer, EfficientDet, RetinaNet and SegNet. These are available in the Habana Model References GitHub repository.
We continued making improvements to our PyTorch operator coverage and Lazy mode implementation. With this release, we are dropping support for TorchScript graph mode; only eager mode and lazy mode will be supported going forward.
We have added support for PyTorch version 1.8.1.
You can find more information on our release notes page.