PyPI - keras-nightly - Versions diffs - 3.14.0.dev2026011604__tar.gz → 3.14.0.dev2026011704__tar.gz - Mend

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: keras-nightly
-Version: 3.14.0.dev2026011604
+Version: 3.14.0.dev2026011704
 Summary: Multi-backend Keras
 Author-email: Keras team <keras-users@googlegroups.com>
 License: Apache License 2.0

@@ -539,11 +539,21 @@ def _do_lstm_arguments_support_cudnn(
 def _has_fully_masked_sequence(mask):
-    # Cudnn kernel will error out if the input sequence contains any
-    # fully masked data. We walk around this issue by rerouting the computation
-    # to standard kernel, until the issue on cudnn side has been fixed.  For a
-    # fully masked sequence, it will contain all Falses. To make it easy to
-    # check, we inverse the boolean, check if any of the sequence has all True.
+    """Check if input sequence contains any fully masked data.
+    cuDNN kernel will error out if the input sequence contains any fully masked
+    data. We work around this issue by rerouting the computation to the
+    standard kernel until the issue on the cuDNN side has been fixed. For a
+    fully masked sequence, it will contain all `False` values. To make it easy
+    to check, we invert the boolean and check if any of the sequences has all
+    `True` values.
+    Args:
+        mask: The mask tensor.
+    Returns:
+        A boolean tensor, `True` if the mask contains a fully masked sequence.
+    """
     return tf.reduce_any(
         tf.reduce_all(tf.logical_not(tf.cast(mask, dtype="bool")), axis=1)
     )
@@ -900,8 +910,8 @@ def _cudnn_lstm(
     if tf.sysconfig.get_build_info()["is_rocm_build"]:
         # ROCm MIOpen's weight sequence for LSTM is different from both
-        # canonical and Cudnn format
-        # MIOpen: [i, f, o, c] Cudnn/Canonical: [i, f, c, o]
+        # canonical and cuDNN format
+        # MIOpen: [i, f, o, c] cuDNN/Canonical: [i, f, c, o]
         # i is input gate weights.
         # f is forget gate weights.
         # o is output gate weights.

@@ -413,11 +413,21 @@ def _is_sequence_right_padded(mask):
 def _has_fully_masked_sequence(mask):
-    # Cudnn kernel will error out if the input sequence contains any
-    # fully masked data. We walk around this issue by rerouting the computation
-    # to standard kernel, until the issue on cudnn side has been fixed.  For a
-    # fully masked sequence, it will contain all Falses. To make it easy to
-    # check, we inverse the boolean, check if any of the sequence has all True.
+    """Check if input sequence contains any fully masked data.
+    cuDNN kernel will error out if the input sequence contains any fully masked
+    data. We work around this issue by rerouting the computation to the
+    standard kernel until the issue on the cuDNN side has been fixed. For a
+    fully masked sequence, it will contain all `False` values. To make it easy
+    to check, we invert the boolean and check if any of the sequences has all
+    `True` values.
+    Args:
+        mask: The mask tensor.
+    Returns:
+        A boolean tensor, `True` if the mask contains a fully masked sequence.
+    """
     return torch.any(torch.all(~mask, dim=1))
@@ -447,8 +457,8 @@ def _compute_sequence_length_from_mask(mask, batch_first):
     The masking tensor is a 2D boolean tensor with shape [batch, timestep]. For
     any timestep that should be masked, the corresponding field will be False.
     Consider the following example:
-      a = [[True, True, False, False]
-           [True, True, True, False]]
+        a = [[True, True, False, False]
+             [True, True, True, False]]
     It is a (2, 4) tensor, and the corresponding sequence length result should
     be 1D tensor with value [2, 3]. Note that the masking tensor must be right
     padded that could be checked by, e.g., `is_sequence_right_padded()`.
@@ -467,12 +477,19 @@ def _compute_sequence_length_from_mask(mask, batch_first):
 def prepare_lstm_weights(lstm, kernel, recurrent_kernel, bias, device):
-    """Copies kernel and recurrent kernel weights in the Pytorch format
+    """Copies kernel and recurrent kernel weights into the PyTorch format.
     We split the kernel and recurrent kernel weights, create associated
-    torch tensors adapted to be in line with the Cudnn optimization.
-    After we have copied the weights, we ensure the paramters are on
-    the same device and memory layout is optimized for Cudnn.
+    torch tensors adapted to be in line with the cuDNN optimization.
+    After we have copied the weights, we ensure the parameters are on
+    the same device and memory layout is optimized for cuDNN.
+    Args:
+        lstm: The PyTorch LSTM layer to prepare weights for.
+        kernel: The kernel weights tensor.
+        recurrent_kernel: The recurrent kernel weights tensor.
+        bias: The bias tensor.
+        device: The device to place the tensors on.
     """
     lstm = lstm.to(device)

@@ -1,7 +1,7 @@
 from keras.src.api_export import keras_export
 # Unique source of truth for the version number.
-__version__ = "3.14.0.dev2026011604"
+__version__ = "3.14.0.dev2026011704"
 @keras_export("keras.version")

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: keras-nightly
-Version: 3.14.0.dev2026011604
+Version: 3.14.0.dev2026011704
 Summary: Multi-backend Keras
 Author-email: Keras team <keras-users@googlegroups.com>
 License: Apache License 2.0

keras-nightly 3.14.0.dev2026011604__tar.gz → 3.14.0.dev2026011704__tar.gz

keras-nightly 3.14.0.dev2026011604tar.gz → 3.14.0.dev2026011704tar.gz