Search results

tree walks still run on single-tensor calls whose structure is already known

Status: Open.

#23294 In keras-team/keras;

· pctablet505 opened on Jul 19, 2026

_clear_losses walks every sublayer even when no losses exist

Status: Open.

#23292 In keras-team/keras;

· pctablet505 opened on Jul 19, 2026

Variable value/dtype accessors redo derivations on every weight read

Status: Open.

#23293 In keras-team/keras;

· pctablet505 opened on Jul 19, 2026

Functional models rebuild static call state on every forward pass

Status: Open.

#23291 In keras-team/keras;

· pctablet505 opened on Jul 19, 2026

the layer name scope opens twice per call

Status: Open.

#23290 In keras-team/keras;

· pctablet505 opened on Jul 19, 2026

Race condition in PyDataset on_epoch_begin

type:Bug

Status: Open.

#23283 In keras-team/keras;

· mr-field opened on Jul 17, 2026

TorchTrainer takes the slow loss/batch-size path every step (and crashes GPU fit with a device bug)

type:Bug

Status: Open.

#23282 In keras-team/keras;

· pctablet505 opened on Jul 17, 2026

_dict_to_ordered_dict does a full pytree traversal even for a plain leaf tensor

type:Bug

Status: Open.

#23281 In keras-team/keras;

· pctablet505 opened on Jul 17, 2026

Layer.__call__ re-assigns self.built/self._called even when already set

type:Bug

Status: Open.

#23280 In keras-team/keras;

· pctablet505 opened on Jul 17, 2026

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

type:Bug

Status: Open.

#23279 In keras-team/keras;

· pctablet505 opened on Jul 17, 2026

slice() coerces dynamic-shape indices to Python int and specializes shapes

type:Bug

Status: Open.

#23278 In keras-team/keras;

· pctablet505 opened on Jul 17, 2026

MultiHeadAttention doesn't use scaled_dot_product_attention on torch (also breaks torch.compile)

type:Bug

Status: Open.

#23277 In keras-team/keras;

· pctablet505 opened on Jul 17, 2026