keras · Issues· 209 open

rgb_to_hsv does not validate channel count for Keras Input with channels_first

IoU.result raises an axis error when axis is not the last dimension

With Torch backend `sparse_categorical_crossentropy` silently accepts out-of-range class indices on MPS

Torch backend: `binary_crossentropy` gives a finite loss for NaN predictions on MPS

Torch backend: `binary_crossentropy` with an integer target stops the Python process on MPS

SimpleRNN and RandomGrayscale drop seed from get_config

soft_shrink, sparse_plus and selu mangle integer inputs (numpy/tensorflow/openvino truncate, torch raises)

TF backend: conv2d fails on TF-CPU when using channel_first

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

torch pooling does two NHWC<->NCHW copies when one would do

tree.traverse can't be captured by torch.compile on the torch backend

[Feature] LinearAttention layer for Keras

test_eigh is silently skipped on the OpenVINO backend, and it fails when it runs

Inconsistent behavior of binary_crossentropy under torch backend

Lambda model save/load silently changes numeric constants on Windows