Highlights

• Support for BigDL 0.7.2 and Spark 2.4; see the download page for all the supported versions.
• Initial OpenVINO support for the model serving API, which can use OpenVINO toolkit to accelerate the inference speed for the TensorFlow models on Analytics Zoo. please refer to the related document and example for more details.
• Initial Persistent Memory support for distributed deep learning training, which can leverage Intel Optane DC Persistent Memory to cache large training data set; please refer to the related document for more details.
• Various new features, including additional built-in models (such as sequence-to-sequence model and unsupervised time series anomaly detection model), learning rate schedule in Adam, Spark ML vector support in nnframes, TensorFlow Keras model support in TFOptimizer, etc.