It’s only been a few days since we witnessed the release of Tensorflow 1.6.0, and now the first release candidate of Tensorflow 1.7.0 is already here!
There are quite a few major features and improvements in this new release candidate. However, no breaking changes are unveiled as such. With Tensorflow 1.7.0-rc0, TensorBoard Debugger Plugin, the graphical user interface (GUI) of TensorFlow Debugger (tfdbg), is now in alpha. Also, Eager mode is moving out of contrib.
Other additional major features include:
- EGraph rewrites emulating fixed-point quantization compatible with TensorFlow Lite are now supported by new tf.contrib.quantize package.
- Easily customize gradient computation available with tf.custom_gradient.
- New tf.contrib.data.SqlDataset provides an experimental support for reading a sqlite database as a Dataset
- Distributed Mutex / CriticalSection added to tf.contrib.framework.CriticalSection.
- Better text processing with tf.regex_replace.
- Easy, efficient sequence input with tf.contrib.data.bucket_by_sequence_length
Apart from these, there is a myriad of bug fixes and small changes. Some of these include:
Unlock access to the largest independent learning library in Tech for FREE!
Get unlimited access to 7500+ expert-authored eBooks and video courses covering every tech area you can think of.
Renews at $19.99/month. Cancel anytime
- MaxPoolGradGrad support is added for Accelerated Linear Algebra (XLA). CSE pass from Tensorflow is now disabled.
- tf.py_func now reports the full stack trace if an exception occurs.
- TPUClusterResolver now integrated with GKE's integration for Cloud TPUs.
- A new library added for statistical testing of samplers.
- Helpers added to stream data from the GCE VM to a Cloud TPU.
- ClusterResolvers are integrated with TPUEstimator.
- Metropolis_hastings interface unified with HMC kernel.
- LIBXSMM convolutions moved to a separate --define flag so that they are disabled by default.
- MomentumOptimizer lambda fixed.
- tfp.layers boilerplate reduced via programmable docstrings.
- auc_with_confidence_intervals, a method for computing the AUC and confidence interval with linearithmic time complexity added.
- regression_head now accepts customized link function, to satisfy the usage that user can define their own link function if the array_ops.identity does not meet the requirement.
- initialized_value and initial_value behaviors fixed for ResourceVariables created from VariableDef protos.
- TensorSpec added to represent the specification of Tensors.
- Constant folding pass is now deterministic.
To know about other bug-fixes and changes visit the Tensorflow 1.7.0-rc0 Github Repo.