Releases · keras-team/keras

Please see https://github.com/tensorflow/tensorflow/blob/r2.8/RELEASE.md for Keras release notes.

tf.keras:
- Preprocessing Layers
  - Added a tf.keras.layers.experimental.preprocessing.HashedCrossing
    layer which applies the hashing trick to the concatenation of crossed
    scalar inputs. This provides a stateless way to try adding feature crosses
    of integer or string data to a model.
  - Removed keras.layers.experimental.preprocessing.CategoryCrossing. Users
    should migrate to the HashedCrossing layer or use
    tf.sparse.cross/tf.ragged.cross directly.
  - Added additional standardize and split modes to TextVectorization.
    - standardize="lower" will lowercase inputs.
    - standardize="string_punctuation" will remove all puncuation.
    - split="character" will split on every unicode character.
  - Added an output_mode argument to the Discretization and Hashing
    layers with the same semantics as other preprocessing layers. All
    categorical preprocessing layers now support output_mode.
  - All preprocessing layer output will follow the compute dtype of a
    tf.keras.mixed_precision.Policy, unless constructed with
    output_mode="int" in which case output will be tf.int64.
    The output type of any preprocessing layer can be controlled individually
    by passing a dtype argument to the layer.
- tf.random.Generator for keras initializers and all RNG code.
  - Added 3 new APIs for enable/disable/check the usage of
    tf.random.Generator in keras backend, which will be the new backend for
    all the RNG in Keras. We plan to switch on the new code path by default in
    tf 2.8, and the behavior change will likely to cause some breakage on user
    side (eg if the test is checking against some golden nubmer). These 3 APIs
    will allow user to disable and switch back to legacy behavior if they
    prefer. In future (eg tf 2.10), we expect to totally remove the legacy
    code path (stateful random Ops), and these 3 APIs will be removed as well.
- tf.keras.callbacks.experimental.BackupAndRestore is now available as
  tf.keras.callbacks.BackupAndRestore. The experimental endpoint is
  deprecated and will be removed in a future release.
- tf.keras.experimental.SidecarEvaluator is now available as
  tf.keras.utils.SidecarEvaluator. The experimental endpoint is
  deprecated and will be removed in a future release.
- Metrics update and collection logic in default Model.train_step() is now
  customizable via overriding Model.compute_metrics().
- Losses computation logic in default Model.train_step() is now
  customizable via overriding Model.compute_loss().
- jit_compile added to Model.compile() on an opt-in basis to compile the
  model's training step with XLA. Note that
  jit_compile=True may not necessarily work for all models.

What's Changed

Cleanup legacy Keras files by @qlzh727 in #14256
Sync OSS keras to head. by @qlzh727 in #14300
Update build script for GPU build. by @copybara-service in #14336
Move the LossReduction class from tf to Keras. by @copybara-service in #14362
Update keras API generate script. by @copybara-service in #14418
Adding extra target that are needed by PIP package dependency. by @copybara-service in #14421
Add test related targets to PIP package list. by @copybara-service in #14427
Sync OSS keras to head. by @copybara-service in #14428
Update visibility setting for keras/tests to enable PIP package testing. by @copybara-service in #14429
Remove items from PIP_EXCLUDED_FILES which is needed with testing PIP. by @copybara-service in #14431
Split bins into num_bins and bin_boundaries arguments for discretization by @copybara-service in #14507
Update pbtxt to use _PRFER_OSS_KERAS=1. by @copybara-service in #14519
Sync OSS keras to head. by @copybara-service in #14572
Sync OSS keras to head. by @copybara-service in #14614
Cleanup the bazelrc and remove unrelated items to keras. by @copybara-service in #14616
Sync OSS keras to head. by @copybara-service in #14624
Remove object metadata when saving SavedModel. by @copybara-service in #14697
Fix static shape inference for Resizing layer. by @copybara-service in #14712
Make TextVectorization work with list input. by @copybara-service in #14711
Remove deprecated methods of Sequential model. by @copybara-service in #14714
Improve Model docstrings by @copybara-service in #14726
Add migration doc for legacy_tf_layers/core.py. by @copybara-service in #14740
PR #43417: Fixes #42872: map_to_outputs_names always returns a copy by @copybara-service in #14755
Rename the keras.py to keras_lib.py to resolve the name conflict during OSS test. by @copybara-service in #14778
Switch to tf.io.gfile for validating vocabulary files. by @copybara-service in #14788
Avoid serializing generated thresholds for AUC metrics. by @copybara-service in #14789
Fix data_utils.py when name ends with .tar.gz by @copybara-service in #14777
Fix lookup layer oov token check when num_oov_indices > len(vocabulary tokens) by @copybara-service in #14793
Update callbacks.py by @jvishnuvardhan in #14760
Fix keras metric.result_state when the metric variables are sharded variable. by @copybara-service in #14790
Fix typos in CONTRIBUTING.md by @amogh7joshi in #14642
Fixed ragged sample weights by @DavideWalder in #14804
Pin the protobuf version to 3.9.2 which is same as the TF. by @copybara-service in #14835
Make variable scope shim regularizer adding check for attribute presence instead of instance class by @copybara-service in #14837
Add missing license header for leakr check. by @copybara-service in #14840
Fix TextVectorization with output_sequence_length on unknown input shapes by @copybara-service in #14832
Add more explicit error message for instance type checking of optimizer. by @copybara-service in #14846
Set aggregation for variable when using PS Strategy for aggregating variables when running multi-gpu tests. by @copybara-service in #14845
Remove unnecessary reshape layer in MobileNet architecture by @copybara-service in #14854
Removes caching of the convolution tf.nn.convolution op. While this provided some performance benefits, it also produced some surprising behavior for users in eager mode. by @copybara-service in #14855
Output int64 by default from Discretization by @copybara-service in #14841
add patterns to .gitignore by @haifeng-jin in #14861
Clarify documentation of DepthwiseConv2D by @vinhill in #14817
add DepthwiseConv1D layer by @fsx950223 in #14863
Make model summary wrap by @Llamrei in #14865
Update the link in Estimator by @hirobf10 in #14901
Fix int given for float args by @SamuelMarks in #14900
Fix RNN, StackedRNNCells with nested state_size, output_size TypeError issues by @Ending2015a in #14905
Fix the use of imagenet_utils.preprocess_input within a Lambda layer with mixed precision by @anth2o in #14917
Fix docstrings in MultiHeadAttention layer call argument return_attention_scores. by @guillesanbri in #14920
Check if layer has _metrics_lock attribute by @DanBmh in #14903
Make keras.Model picklable by @adriangb in #14748
Fix typo in docs by @seanmor5 in #14946
use getter setter by @fsx950223 in #14948
Close _SESSION.session in clear_session by @sfreilich in #14414
Fix keras nightly PIP package build. by @copybara-service in #14957
Fix EarlyStopping stop at fisrt epoch when patience=0 ; add auc to au… by @DachuanZhao in...