TensorFlow 1.8.0 is released, this release has made a lot of feature changes and bug fixes.
Main features and performance improvements:
tf.contrib.distribute.MirroredStrategy() can now be passed to tf.estimator.RunConfig() to run Estimator models on multiple GPUs on a single machine.
Added tf.contrib.data.prefetch_to_device() to support prefetching GPU memory.
Added gradient boosted trees: BoostedTreesClassifier and BoostedTreesRegressor.
Added third-generation pipeline configuration for cloud TPUs for improved performance and availability.
tf.contrib.bayesflow moved to its own repo.
Added tf.contrib. {proto,rpc} to allow generic prototype parsing and RPC communication.
bug fix:
tf.data
:Added tf.contrib.data.prefetch_to_device to prefetch dataset elements to GPU memory.
Adds tf.contrib.data.AUTOTUNE, which allows the tf.data runtime to automatically adjust the prefetch buffer size based on your system and environment.
Add tf.contrib.data.make_csv_dataset to build a dataset of CSV files.
Eager Execution:
Automatic device placement is enabled.
tf.GradientTape has been moved out of contrib.
tf.keras
:Added fashion mnist dataset
New data preprocessing functions:
image/random_brightness
,sequence/TimeseriesGenerator
, andtext/hashing_trick
See the announcement for details .
download link: