keras-team/keras v3.1.0

Keras 3.1.0

on GitHub

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