1 Release published by 1 person

• v3.11.3 Keras 3.11.3

  published Aug 22, 2025

15 Pull requests merged by 11 people

• Remove the usage of JAX_DEFAULT_DTYPE_BITS and the tests for 64-bit dtypes.

  #21604 merged Aug 22, 2025

• Implement hypot function in keras.ops

  #21606 merged Aug 22, 2025

• Version bump to 3.11.3

  #21607 merged Aug 21, 2025

• Fix support for optional inputs in model.fit

  #21548 merged Aug 21, 2025

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

  #21343 merged Aug 21, 2025

• feat(quantization): Add GPTQ n-bit quantization support

  #21551 merged Aug 21, 2025

• Refactor TFDataLayer to be more generic for Grain.

  #21598 merged Aug 21, 2025

• Fix GRU with return_state=True on tf backend with cuda

  #21603 merged Aug 21, 2025

• Propagate safe_mode flag to legacy h5 loading code.

  #21602 merged Aug 20, 2025

• Fix shape mismatch in Keras Attention layer during masking

  #21595 merged Aug 20, 2025

• Add Grain support to image_dataset_from_directory and text_dataset_from_directory

  #21593 merged Aug 19, 2025

• Make vectorized_map op serializable.

  #21597 merged Aug 19, 2025

• [OpenVINO backend] fix ops.slice

  #21580 merged Aug 18, 2025

• Changed the specification of MeanMetricWrapper

  #21569 merged Aug 18, 2025

• Fix numpy tests for window ops, angle, identity, eye.

  #21596 merged Aug 18, 2025

7 Pull requests opened by 5 people

• Add meshgrid op support in Keras OpenVINO backend

  #21600 opened Aug 20, 2025

• Ignore regex fallback for quantized policies in DTypePolicyMap

  #21608 opened Aug 22, 2025

• Add TF support for mixed optional inputs (None/tensor) from generator in model.fit/evaluate/predict

  #21609 opened Aug 22, 2025

• Implements production GPTQ quantization

  #21610 opened Aug 23, 2025

• Fix issue with concatenate, masking and symbolic inputs

  #21611 opened Aug 23, 2025

• Add dtype tests for `ops.image.*`.

  #21612 opened Aug 24, 2025

• Add linspace and logspace implementations in OpenVINO NumPy backend

  #21613 opened Aug 24, 2025

2 Issues closed by 2 people

• Attention layer issue broadcasting mask

  #21483 closed Aug 20, 2025

• Implement DINOv3: Self-Supervised Vision Transformer as a Keras 3 Example

  #21591 closed Aug 18, 2025

3 Issues opened by 3 people

• Compatibility issue between `add_loss` and `TimeDistributed` when using JAX backend

  #21605 opened Aug 21, 2025

• TF = 2.18, Keras= 3.10 Unrecognized keyword arguments passed to MultiHeadAttention: {'return_attention_scores': True}

  #21601 opened Aug 20, 2025

• Input and Output layer name inconsistency with Keras' sequential model

  #21599 opened Aug 19, 2025

64 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.

• Proposed fix for issue #21519: Reshape layer does not handle -1 shape…

  #21568 commented on Aug 24, 2025 • 6 new comments

• Add Distillation API to Keras

  #21572 commented on Aug 20, 2025 • 2 new comments

• [Keras 3 OpenVINO Backend]: Support numpy.logspace operation #30114

  #21253 commented on Aug 23, 2025 • 0 new comments

• Fix stateful RNN to raise ValueError on batch size mismatch

  #21249 commented on Aug 23, 2025 • 0 new comments

• [OpenVINO Backend] Support for numpy.flip

  #21230 commented on Aug 20, 2025 • 0 new comments

• implementation of logaddexp in openvino backend (issue #29489)

  #21216 commented on Aug 20, 2025 • 0 new comments

• OpenVINO: linspace

  #21209 commented on Aug 23, 2025 • 0 new comments

• Support numpy.prod operation

  #21188 commented on Aug 20, 2025 • 0 new comments

• [OpenVino BackEnd] support np.diagnol for ov BackEnd

  #21156 commented on Aug 20, 2025 • 0 new comments

• Resolve: [Good First Issue][Keras 3 OpenVINO Backend]: Support numpy.linspace operation #29485

  #21140 commented on Aug 20, 2025 • 0 new comments

• Add support for numpy.flip in OpenVINO backend

  #21132 commented on Aug 20, 2025 • 0 new comments

• [OpenVINO Backend] Support: Implemented numpy.digitize for keras ov backend

  #21087 commented on Aug 20, 2025 • 0 new comments

• [OpenVINO Backend] Support: Implemented numpy.diagonal for keras ov backend

  #21086 commented on Aug 20, 2025 • 0 new comments

• Updated numpy.flip using openvino opset

  #21057 commented on Aug 20, 2025 • 0 new comments

• [Documentation] Updated Categorical Focal Crossentropy Loss Docstring

  #21044 commented on Aug 23, 2025 • 0 new comments

• Add numpy.logaddexp to openvino backend

  #21043 commented on Aug 20, 2025 • 0 new comments

• 20954. adding a some changes line no.423

  #20991 commented on Aug 23, 2025 • 0 new comments

• This PR implements the numpy.logspace function for the OpenVINO backend as requested in issue #30114.

  #21266 commented on Aug 20, 2025 • 0 new comments

• Implement numpy.prod method for OpenVino backend

  #21282 commented on Aug 20, 2025 • 0 new comments

• use native dumpy conversion when available

  #21288 commented on Aug 23, 2025 • 0 new comments

• numpy.meshgrid implementation

  #21292 commented on Aug 20, 2025 • 0 new comments

• Added openvino backend support for numpy.median (Issue #30115)

  #21379 commented on Aug 20, 2025 • 0 new comments

• Openvino backend improvements

  #21381 commented on Aug 20, 2025 • 0 new comments

• Fix(OpenVINO): Ensure numpy.prod passes tests and handles boolean inp…

  #21403 commented on Aug 20, 2025 • 0 new comments

• [Keras 3 OpenVINO Backend]: Support numpy.prod operation

  #21405 commented on Aug 20, 2025 • 0 new comments

• [Keras 3 OpenVINO Backend]: Support numpy.flip operation

  #21410 commented on Aug 20, 2025 • 0 new comments

• Running TPU tests on linux-x86-ct6e-44-1tpu

  #21425 commented on Aug 22, 2025 • 0 new comments

• Fix TypeError in keras.ops.searchsorted by using shape instead of len

  #21472 commented on Aug 25, 2025 • 0 new comments

• [OpenVINO backend] Support numpy.logaddexp

  #21522 commented on Aug 20, 2025 • 0 new comments

• [OpenVINO backend] Support numpy.prod

  #21567 commented on Aug 20, 2025 • 0 new comments

• [OpenVINO backend] Support numpy.diagonal issue 29115

  #21584 commented on Aug 23, 2025 • 0 new comments

• My fix branch

  #21585 commented on Aug 23, 2025 • 0 new comments

• MeanIoU and its child classes throwing error when used with `ignore_class`

  #19524 commented on Aug 18, 2025 • 0 new comments

• Backward compatibility issues: arbitrary inputs and overreliance on custom layers

  #19314 commented on Aug 19, 2025 • 0 new comments

• Reshape layer does not handle -1 shape infor dynamically

  #21519 commented on Aug 19, 2025 • 0 new comments

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

  #21301 commented on Aug 20, 2025 • 0 new comments

• Mixed precision on AMD GPUs results in lower performance

  #21520 commented on Aug 20, 2025 • 0 new comments

• Integrate Grain into Keras

  #21479 commented on Aug 20, 2025 • 0 new comments

• Feature Extraction Losses

  #20839 commented on Aug 20, 2025 • 0 new comments

• RandAugment incorrect behavior in tf graph mode

  #21169 commented on Aug 20, 2025 • 0 new comments

• How does Keras MultiHeadAttention handle input feature dimension vs heads and reshape the output?

  #21219 commented on Aug 20, 2025 • 0 new comments

• 🚀OpenVINO Backend: GFIs for Further Development🚀

  #20910 commented on Aug 20, 2025 • 0 new comments

• Model.quantize() increases memory allocated.

  #21518 commented on Aug 20, 2025 • 0 new comments

• BatchNormalization fails after Concatenation of masked Embeddings

  #21581 commented on Aug 21, 2025 • 0 new comments

• LSTM layer doesn't learn with TextVectorization output (padding ?)

  #20898 commented on Aug 21, 2025 • 0 new comments

• does Functional Model support "string" dtype for end-to-end pipelines?

  #18410 commented on Aug 21, 2025 • 0 new comments

• `Function.operations` include operations used to compute inputs

  #18647 commented on Aug 21, 2025 • 0 new comments

• OpenVINOKerasTensor does not implement the full expected API of KerasTensor

  #21544 commented on Aug 21, 2025 • 0 new comments

• Custom call context args are not supported for the common methods for model calling (e.g. fit, predict, evaluate, ...)

  #21530 commented on Aug 22, 2025 • 0 new comments

• Keras 3.7 Broke My Code

  #20568 commented on Aug 22, 2025 • 0 new comments

• Training performance degradation after switching from Keras 2 mode to Keras 3 using Tensorflow

  #20283 commented on Aug 22, 2025 • 0 new comments

• Is their a way to create custom tensor-like types in Keras (similar to torch.Tensor or tf.experimental.ExtensionType)

  #21574 commented on Aug 22, 2025 • 0 new comments

• Keras stuck on set_random_seed with PyTorch backend on Metal (Applied Silicon)

  #21542 commented on Aug 22, 2025 • 0 new comments

• Multi-backend support for timeseries_dataset_from_array()

  #21536 commented on Aug 23, 2025 • 0 new comments

• Modified example in depthwise_conv1d file

  #20268 commented on Aug 23, 2025 • 0 new comments

• Fix gradient tape issue in test_gradient_tape_issue.py

  #20277 commented on Aug 23, 2025 • 0 new comments

• Added description for keras.constraints.serialize

  #20628 commented on Aug 23, 2025 • 0 new comments

• Fix typos in string_lookup

  #20667 commented on Aug 23, 2025 • 0 new comments

• Fix to issue #20616

  #20683 commented on Aug 23, 2025 • 0 new comments

• Fix typos in additive_attention

  #20711 commented on Aug 23, 2025 • 0 new comments

• fix tracking issue with lazy filled attributes

  #20783 commented on Aug 23, 2025 • 0 new comments

• Perceptual Similarity loss

  #20844 commented on Aug 23, 2025 • 0 new comments

• [Documentation] Added missing docstrings to 92 function(s) in examples.

  #20908 commented on Aug 23, 2025 • 0 new comments

• [OpenVINO Backend] Support np.diag

  #20967 commented on Aug 20, 2025 • 0 new comments