If -define=tf_api_version=2 flag is passed in, then bazel will build TensorFlow API version 2.0.

In tf eager execution, allow re-entering a GradientTape context.There was previously a bug in the bijector_impl where the _reduce_jacobian_det_over_event does not handle scalar ILDJ implementations properly.Support negative positions for tf.substr.Add model explainability to Boosted Trees.Add searchsorted (ie lower/upper_bound) op.Add functionality to SubSlice a tensor.Add option to calculate string length in Unicode characters.Allow subslicing Tensors with a single dimension.This is a bugfix these ops were never meant to support integers. Remove integer types from tf.nn.softplus and tf.nn.softsign OpDefs.Instead of jemalloc, revert back to using system malloc since it simplifies build and has comparable performance.Deprecate, and replace by tf.data.experimental.Catching a ValueError exception thrown by tf.contrib.get_signature_def_by_key should be replaced by catching a KeyError exception. Replace any calls to tf.contrib.get_signature_def_by_key(metagraph_def, signature_def_key) with meta_graph_def.signature_def.Add support for num_parallel_calls to tf.All C++ code moves to the tensorflow::data namespace.New tf.() API allows users to create finite windows of input dataset when combined with the tf.() API, this allows users to implement customized batching.New tf.() API allows users to reduce a finite dataset to a single element using a user-provided reduce function.tf.data users can now represent, get, and set options of TensorFlow input pipelines using tf.data.Options(), tf.(), and tf._options() respectively.Ignite Dataset added to contrib/ignite that allows to work with Apache Ignite.TensorFlow binaries are built with XLA support linked in by default.Keras models now support evaluating with a tf.data.Dataset.Keras models can now be directly exported to the SavedModel format( tf.contrib.saved_model.save_keras_model()) and used with Tensorflow Serving.

Release 1.12.0 Major Features and Improvements

Offline resampler as good as hqplayer