13 Pull requests merged by 8 people

• Allow ops_test.py to run with the api folder collapsed at the root.

  #21360 merged Jun 6, 2025

• Make dtype a static parameter of OnesLike and ZerosLike.

  #21358 merged Jun 5, 2025

• [OpenVINO Backend] support comparison ops (>, <, <=, >=, ==, !=) for OpenVINOKerasTensor

  #21348 merged Jun 4, 2025

• [OpenVINO Backend] support where operation

  #21354 merged Jun 4, 2025

• [OpenVINO Backend] enable core tests

  #21353 merged Jun 4, 2025

• [OpenVINO Backend] support __rpow__ for OpenVINOKerasTensor

  #21347 merged Jun 4, 2025

• Fix - Set torch to None when import fails in conftest

  #21357 merged Jun 4, 2025

• Fix callback

  #21349 merged Jun 4, 2025

• Fix comment

  #21340 merged Jun 2, 2025

• Make ops consistent and test ops consistency.

  #21334 merged Jun 2, 2025

• Ensures consistent serialization for Functional model

  #21341 merged Jun 2, 2025

• Modified InputSpec of SpectralNormalization layer

  #21335 merged Jun 1, 2025

• Bump the github-actions group with 2 updates

  #21339 merged Jun 1, 2025

8 Pull requests opened by 5 people

• [*.py] Use f-strings and `os.path.join` throughout

  #21343 opened Jun 2, 2025

• [keras/src/print.py] Implement backend-specialised `print` function

  #21344 opened Jun 2, 2025

• Modified compute_output_shape function to handle broadcasting behavior in layers.Rescaling

  #21351 opened Jun 4, 2025

• Fix rms_scaling in LayerNormalization

  #21352 opened Jun 4, 2025

• [OpenVINO Backend] update __getitem__

  #21359 opened Jun 5, 2025

• [OpenVINO Backend] update slice method

  #21361 opened Jun 6, 2025

• [OpenVINO Backend] support ops.slice_update

  #21362 opened Jun 6, 2025

• Bump torch from 2.6.0 to 2.7.1

  #21363 opened Jun 7, 2025

12 Issues closed by 4 people

• SavedModel format doesn't support masking in Keras 3

  #21133 closed Jun 7, 2025

• Sharing weights across layers in keras 3 [feature request]

  #18821 closed Jun 6, 2025

• Keras predict/predict_on_batch giving different answers than predict_step/__call__()

  #19403 closed Jun 6, 2025

• Potential bug in legacy h5 weights loading.

  #19694 closed Jun 6, 2025

• Attention layer does not accept output of previous layers in functional API

  #20318 closed Jun 6, 2025

• Issue error you're encountering stems from an issue with using masking in the Embedding layer for POS Tagging Model

  #20773 closed Jun 6, 2025

• Error: ``model`` must be a callable, a Keras Model instance or None

  #20893 closed Jun 6, 2025

• sample_weight with multiple, named losses

  #20992 closed Jun 6, 2025

• evaluate() shows incorrect metric values in progress bar vs returned results

  #21301 closed Jun 5, 2025

• Ensured torch import is properly handled

  #21004 closed Jun 4, 2025

• keras

  #21345 closed Jun 3, 2025

• SpectralNormalization layer fails on multiple batch dimensions

  #21320 closed Jun 1, 2025

5 Issues opened by 5 people

• FlaxLayer with NNX

  #21364 opened Jun 7, 2025

• MultipleChoice Task with Transfer Learning example web page display error

  #21356 opened Jun 4, 2025

• Documentation: better info about class_weight and/or DataAdapters

  #21355 opened Jun 4, 2025

• Cannot properly save the model using TorchModuleWrapper and load the model

  #21350 opened Jun 4, 2025

• Data-Augmentation layers can't be exported to ONNX

  #21346 opened Jun 3, 2025

20 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Keras <> NNX integration

  #21252 commented on Jun 7, 2025 • 35 new comments

• Re-apply Fixed issue with dot_product_attention when using TPU. #21254 after addressing cuDNN/FlashAttention API updates

  #21333 commented on Jun 7, 2025 • 4 new comments

• [OpenVINO backend] Update `ndim` behavior

  #21317 commented on Jun 6, 2025 • 1 new comment

• Add memory usage monitor callback

  #21245 commented on Jun 7, 2025 • 0 new comments

• MLX Backend

  #19571 commented on Jun 8, 2025 • 0 new comments

• Stateful metrics produce incorrect output in distributed context

  #21054 commented on Jun 7, 2025 • 0 new comments

• layers.GRU returns wrong shaped output with GPU

  #20173 commented on Jun 6, 2025 • 0 new comments

• Cannot generate heatmaps using Gradcam library

  #20475 commented on Jun 6, 2025 • 0 new comments

• [Distribution] Standardize the device API.

  #18522 commented on Jun 5, 2025 • 0 new comments

• TimeDistributed layer computational time

  #21179 commented on Jun 5, 2025 • 0 new comments

• Error exporting model with multiple outputs using model.export when specifying input_signature

  #21258 commented on Jun 5, 2025 • 0 new comments

• keras.utils.timeseries_dataset_from_array bugs 1

  #21306 commented on Jun 5, 2025 • 0 new comments

• TF+XLA+Mixed Precision: Keras fails to compute gradients

  #20931 commented on Jun 4, 2025 • 0 new comments

• LayerNormalization with rms_scaling documentation is different from implementation

  #21234 commented on Jun 4, 2025 • 0 new comments

• keras.utils.timeseries_dataset_from_array example bugs

  #21305 commented on Jun 4, 2025 • 0 new comments

• Distributed training with tensorflow doesn't work with Keras 3.7 and above

  #21172 commented on Jun 3, 2025 • 0 new comments

• remat : The custom_gradient decorator currently supports keywords arguments only when eager execution is enabled.

  #21207 commented on Jun 3, 2025 • 0 new comments

• tf.keras.layers.Embedding() somehow passing a float to randrange

  #21285 commented on Jun 3, 2025 • 0 new comments

• Hierarchical classification

  #21274 commented on Jun 3, 2025 • 0 new comments

• Rescaling layer defaults to "float32" if dtype not specified - input is uint8

  #18464 commented on Jun 2, 2025 • 0 new comments