Keras 3.1.1

This is a minor bugfix release over 3.1.0.

What's Changed

• Unwrap variable values in all stateless calls. by @hertschuh in #19287
• Fix draw_seed causing device discrepancy issue during torch's symbolic execution by @KhawajaAbaid in #19289
• Fix TestCase.run_layer_test for multi-output layers by @shkarupa-alex in #19293
• Sine docstring by @grasskin in #19295
• Fix keras.ops.softmax for the tensorflow backend by @tirthasheshpatel in #19300
• Fix mixed precision check in TestCase.run_layer_test: compare with output_spec dtype instead of hardcoded float16 by @shkarupa-alex in #19297
• ArrayDataAdapter no longer converts to NumPy and supports sparse tens… by @hertschuh in #19298
• add token to codecov by @haifeng-jin in #19312
• Add Tensorflow support for variable scatter_update in optimizers. by @hertschuh in #19313
• Replace dm-tree with optree by @james77777778 in #19306
• downgrade codecov to v3 by @haifeng-jin in #19319
• Allow tensors in tf.Datasets to have different dimensions. by @hertschuh in #19318
• update codecov setting by @haifeng-jin in #19320
• Set dtype policy for uint8 by @sampathweb in #19327
• Use Value dim shape for Attention compute_output_shape by @sampathweb in #19284

New Contributors

• @tirthasheshpatel made their first contribution in #19300

Full Changelog: v3.1.0...v3.1.1

Keras 3.1.0

New features

• Add support for int8 inference. Just call model.quantize("int8") to do an in-place conversion of a bfloat16 or float32 model to an int8 model. Note that only Dense and EinsumDense layers will be converted (this covers LLMs and all Transformers in general). We may add more supported layers over time.
• Add keras.config.set_backend(backend) utility to reload a different backend.
• Add keras.layers.MelSpectrogram layer for turning raw audio data into Mel spectrogram representation.
• Add keras.ops.custom_gradient decorator (only for JAX and TensorFlow).
• Add keras.ops.image.crop_images.
• Add pad_to_aspect_ratio argument to image_dataset_from_directory.
• Add keras.random.binomial and keras.random.beta functions.
• Enable keras.ops.einsum to run with int8 x int8 inputs and int32 output.
• Add verbose argument in all dataset-creation utilities.

Notable fixes

• Fix Functional model slicing
• Fix for TF XLA compilation error for SpectralNormalization
• Refactor axis logic across all backends and add support for multiple axes in expand_dims and squeeze

New Contributors

• @mykolaskrynnyk made their first contribution in #19190
• @chicham made their first contribution in #19201
• @joycebrum made their first contribution in #19214
• @EtiNL made their first contribution in #19228

Full Changelog: v3.0.5...v3.1.0

Keras 3.0.5

This release brings many bug fixes and performance improvements, new linear algebra ops, and sparse tensor support for the JAX backend.

Highlights

• Add support for sparse tensors with the JAX backend.
• Add support for saving/loading in bfloat16.
• Add linear algebra ops in keras.ops.linalg.
• Support nested structures in while_loop op.
• Add erfinv op.
• Add normalize op.
• Add support for IterableDataset to TorchDataLoaderAdapter.

New Contributors

• @frazane made their first contribution in #19107
• @SamanehSaadat made their first contribution in #19111
• @sitamgithub-MSIT made their first contribution in #19142
• @timotheeMM made their first contribution in #19169

Full Changelog: v3.0.4...v3.0.5

Keras 3.0.4

This is a minor release with improvements to the LoRA API required by the next release of KerasNLP.

Full Changelog: v3.0.3...v3.0.4

Keras 3.0.3 release

This is a minor Keras release.

What's Changed

• Add built-in LoRA (low-rank adaptation) API to all relevant layers (Dense, EinsumDense, Embedding).
• Add SwapEMAWeights callback to make it easier to evaluate model metrics using EMA weights during training.
• All DataAdapters now create a native iterator for each backend, improving performance.
• Add built-in prefetching for JAX, improving performance.
• The bfloat16 dtype is now allowed in the global set_dtype configuration utility.
• Bug fixes and performance improvements.

New Contributors

• @kiraksi made their first contribution in #18977
• @dugujiujian1999 made their first contribution in #19010
• @neo-alex made their first contribution in #18997
• @anas-rz made their first contribution in #19057

Full Changelog: v3.0.2...v3.0.3

Keras 3.0.2

Breaking changes

There are no known breaking changes in this release compared to 3.0.1.

API changes

• Add keras.random.binomial and keras.random.beta RNG functions.
• Add masking support to BatchNormalization.
• Add keras.losses.CTC (loss function for sequence-to-sequence tasks) as well as the lower-level operation keras.ops.ctc_loss.
• Add ops.random.alpha_dropout and layers.AlphaDropout.
• Add gradient accumulation support for all backends, and enable optimizer EMA for JAX and torch

Full Changelog: v3.0.1...v3.0.2

Keras 3.0.1

This is a minor release focused on bug fixes and performance improvements.

What's Changed

• Bug fixes and performance improvements.
• Add stop_evaluating and stop_predicting model attributes for callbacks, similar to stop_training.
• Add keras.device() scope for managing device placement in a multi-backend way.
• Support dict items in PyDataset.
• Add hard_swish activation and op.
• Fix cuDNN LSTM performance on TensorFlow backend.
• Add a force_download arg to get_file to force cache invalidation.

Full Changelog: v3.0.0...v3.0.1

Keras 3.0.0

Major updates

See the release announcement for a detailed list of major changes. Main highlights compared to Keras 2 are:

• Keras can now be run on top of JAX, PyTorch, TensorFlow, and even NumPy (note that the NumPy backend is inference-only).
• New low-level keras.ops API for building cross-framework components.
• New large-scale model distribution keras.distribution based on JAX.
• New stateless API for layers, models, optimizers, and metrics.

Breaking changes

See this thread for a complete list of breaking changes, as well as the Keras 3 migration guide.

Keras Release 2.15.0

What's Changed

• Typofixes for StringLookup documentation by @cw118 in #18333
• Fix ModelCheckpoint trained-on batch counting when using steps_per_execution>1 by @jasnyj in #17632
• Fix legacy optimizer handling in compile_from_config(). by @nkovela1 in #18492
• Remove options arg from ModelCheckpoint callback for Keras V3 saving, streamline ModelCheckpoint saving flow. Parameterize associated tests. by @nkovela1 in #18545
• Use TENSORFLOW_VERSION when available during pip_build script by @sampathweb in #18739

New Contributors

• @cw118 made their first contribution in #18333
• @jasnyj made their first contribution in #17632

Full Changelog: v2.14.0...v2.15.0

Keras Release 2.14.0

What's Changed

• [keras/layers/normalization] Standardise docstring usage of "Default to" by @SamuelMarks in #17965
• Update Python ver to 3.9 in Dockerfile by @sampathweb in #18076
• [keras/saving/legacy/saved_model] Standardise docstring usage of "Default to" by @SamuelMarks in #17978
• [keras/metrics] Standardise docstring usage of "Default to" by @SamuelMarks in #17972
• Update example losses to bce- metrics/confusion_metrics.py by @Frightera in #18045
• [keras/layers/regularization] Standardise docstring usage of "Default to" by @SamuelMarks in #17968
• [keras/applications/efficientnet.py] Standardise docstring usage of "Default to" by @SamuelMarks in #17758
• [keras/models] Standardise docstring usage of "Default to" by @SamuelMarks in #17974
• [keras/mixed_precision] Standardise docstring usage of "Default to" by @SamuelMarks in #17973
• Update indentation level - losses.py by @Frightera in #18043
• Remove github user sushreebarsa from assignees. by @shmishra99 in #18058
• [keras/preprocessing] Standardise docstring usage of "Default to" by @SamuelMarks in #17977
• [keras/testing_infra] Standardise docstring usage of "Default to" by @SamuelMarks in #17979
• [keras/benchmarks/benchmark_util.py] Use var rather than string literal for is None checks on measure_performance by @SamuelMarks in #17980
• Some cleanup // Optimizers by @Frightera in #18124
• [keras/layers/preprocessing] fix comments in RandomWidth, change to 'horizontally' instead of 'vertically' by @qibolee in #18113
• [keras/applications] Standardise docstring usage of "Default to" by @SamuelMarks in #17954
• [keras/estimator] Standardise docstring usage of "Default to" by @SamuelMarks in #17957
• Fix CategoricalFocalCE documentation by @Frightera in #18144
• [keras/utils] Standardise docstring usage of "Default to" by @SamuelMarks in #17953
• [keras/layers/pooling] Standardise docstring usage of "Default to" by @SamuelMarks in #17966
• use io_utils.print_msg in datset_utils by @pedrobrs in #18054
• Update indentation level - activations.py by @Frightera in #18036
• [keras/legacy_tf_layers] Standardise docstring usage of "Default to" by @SamuelMarks in #17971
• [keras/layers/convolutional] Standardise docstring usage of "Default to" by @SamuelMarks in #17963
• [keras/engine] Standardise docstring usage of "Default to" by @SamuelMarks in #17956
• [keras/feature_column] Standardise docstring usage of "Default to" by @SamuelMarks in #17958
• Fix markdown rendering issue by @sachinprasadhs in #18205
• Consistently use "pickleable" instead of "picklable" by @NicoWeio in #18140
• Use prefetch() after batching // image_dataset.py by @Frightera in #18160
• Update Argument padding for zero_padding1d.py by @SuryanarayanaY in #18223
• RGB image data is not grayscale image data by @misterrioes in #18133
• updated read.me by @VaishnaviMudaliar in #18226
• Fixed the typo in policy.py by @tilakrayal in #18233
• Update example losses @ probabilistic_metrics.py by @Frightera in #18234
• Fixed "reset_state" of R2Score metric by @pdyakov in #18251
• Ignore hidden folders for image_dataset_from_directory by @sachinprasadhs in #18177
• Improve error message for input data to fit. by @tomrtk in #18042
• refactor: _log_epoch_metrics() by @arjun-234 in #18274
• Refactor test cases to improve unit test quality by @freddiewanah in #18303
• typo in disable_interactie_logging by @ganeshiva in #18314
• Removes Python 3.8 support by @sampathweb in #18332
• Update requirements.txt by @qlzh727 in #18382

New Contributors

• @sampathweb made their first contribution in #18076
• @shmishra99 made their first contribution in #18058
• @qibolee made their first contribution in #18113
• @pedrobrs made their first contribution in #18054
• @NicoWeio made their first contribution in #18140
• @SuryanarayanaY made their first contribution in #18223
• @misterrioes made their first contribution in #18133
• @VaishnaviMudaliar made their first contribution in #18226
• @pdyakov made their first contribution in #18251
• @tomrtk made their first contribution in #18042
• @arjun-234 made their first contribution in #18274
• @freddiewanah made their first contribution in #18303
• @ganeshiva made their first contribution in #18314

Full Changelog: v2.13.1...v2.14.0