PyPI - keras-nightly - Versions diffs - 3.14.0.dev2026013104__tar.gz → 3.14.0.dev2026020104__tar.gz - Mend

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: keras-nightly
-Version: 3.14.0.dev2026013104
+Version: 3.14.0.dev2026020104
 Summary: Multi-backend Keras
 Author-email: Keras team <keras-users@googlegroups.com>
 License: Apache License 2.0

@@ -1,6 +1,7 @@
 import builtins
 import math
+import jax
 import jax.experimental.sparse as jax_sparse
 import jax.numpy as jnp
 from jax import export as jax_export
@@ -16,6 +17,18 @@ from keras.src.backend.jax.core import cast
 from keras.src.backend.jax.core import convert_to_tensor
+def _uses_cpu(x):
+    if hasattr(x, "device"):
+        device = x.device
+        if not isinstance(device, jax.Device):
+            # Array is sharded.
+            return False
+        return device.platform == "cpu"
+    else:
+        # This is a Tracer, not a concrete Array.
+        return jax.default_backend() == "cpu"
 def rot90(array, k=1, axes=(0, 1)):
     """Rotate an array by 90 degrees in the specified plane."""
     if array.ndim < 2:
@@ -402,11 +415,9 @@ def arctanh(x):
 def argmax(x, axis=None, keepdims=False):
-    from keras.src.testing.test_case import uses_cpu
     x = convert_to_tensor(x)
     dtype = standardize_dtype(x.dtype)
-    if "float" not in dtype or not uses_cpu() or x.ndim == 0:
+    if "float" not in dtype or x.ndim == 0 or not _uses_cpu(x):
         return jnp.argmax(x, axis=axis, keepdims=keepdims)
     # Fix the flush-to-zero (FTZ) issue based on this issue:
@@ -419,11 +430,9 @@ def argmax(x, axis=None, keepdims=False):
 def argmin(x, axis=None, keepdims=False):
-    from keras.src.testing.test_case import uses_cpu
     x = convert_to_tensor(x)
     dtype = standardize_dtype(x.dtype)
-    if "float" not in dtype or not uses_cpu() or x.ndim == 0:
+    if "float" not in dtype or x.ndim == 0 or not _uses_cpu(x):
         return jnp.argmin(x, axis=axis, keepdims=keepdims)
     # Fix the flush-to-zero (FTZ) issue based on this issue:

@@ -366,6 +366,10 @@ def arctan2(x1, x2):
     x1 = ov_opset.convert(x1, result_type)
     x2 = ov_opset.convert(x2, result_type)
+    nan_x1 = ov_opset.is_nan(x1)
+    nan_x2 = ov_opset.is_nan(x2)
+    nan_mask = ov_opset.logical_or(nan_x1, nan_x2)
     x = ov_opset.divide(x1, x2)
     y = ov_opset.atan(x)
@@ -375,12 +379,12 @@ def arctan2(x1, x2):
     neg_half_pi = ov_opset.constant(-float(np.pi / 2), ov_type)
     zero_const = ov_opset.constant(0.0, ov_type)
-    cond_x2_gt0 = ov_opset.greater(x2, zero_const).output(0)
-    cond_x2_lt0 = ov_opset.less(x2, zero_const).output(0)
+    cond_x2_gt0 = ov_opset.greater(x2, zero_const)
+    cond_x2_lt0 = ov_opset.less(x2, zero_const)
-    cond_x1_ge0 = ov_opset.greater_equal(x1, zero_const).output(0)
-    cond_x1_gt0 = ov_opset.greater(x1, zero_const).output(0)
-    cond_x1_eq0 = ov_opset.equal(x1, zero_const).output(0)
+    cond_x1_ge0 = ov_opset.greater_equal(x1, zero_const)
+    cond_x1_gt0 = ov_opset.greater(x1, zero_const)
+    cond_x1_eq0 = ov_opset.equal(x1, zero_const)
     out_x2_lt0 = ov_opset.select(
         cond_x1_ge0,
@@ -393,7 +397,11 @@ def arctan2(x1, x2):
     out_not_pos = ov_opset.select(cond_x2_lt0, out_x2_lt0, out_x2_zero)
-    final_out = ov_opset.select(cond_x2_gt0, y, out_not_pos)
+    value_out = ov_opset.select(cond_x2_gt0, y, out_not_pos)
+    # Generate NaN safely for all floating dtypes (including bf16)
+    nan_value = ov_opset.divide(zero_const, zero_const)
+    final_out = ov_opset.select(nan_mask, nan_value, value_out)
     return OpenVINOKerasTensor(final_out.output(0))

@@ -949,11 +949,9 @@ def argmax(x, axis=None, keepdims=False):
 def argmin(x, axis=None, keepdims=False):
-    from keras.src.testing.test_case import uses_cpu
     x = convert_to_tensor(x)
     dtype = standardize_dtype(x.dtype)
-    if "float" not in dtype or not uses_cpu() or x.ndim == 0:
+    if "float" not in dtype or x.ndim == 0:
         _x = x
         if axis is None:
             x = tf.reshape(x, [-1])

@@ -342,6 +342,10 @@ def resize(
             "Expected `size` to be a tuple of 2 integers. "
             f"Received: size={size}"
         )
+    if size[0] <= 0 or size[1] <= 0:
+        raise ValueError(
+            f"`size` must have positive height and width. Received: size={size}"
+        )
     if len(images.shape) < 3 or len(images.shape) > 4:
         raise ValueError(
             "Invalid images rank: expected rank 3 (single image) "

@@ -1,5 +1,6 @@
 from keras.src.testing.test_case import TestCase
 from keras.src.testing.test_case import jax_uses_gpu
+from keras.src.testing.test_case import jax_uses_tpu
 from keras.src.testing.test_case import tensorflow_uses_gpu
 from keras.src.testing.test_case import torch_uses_gpu
 from keras.src.testing.test_case import uses_gpu

@@ -650,6 +650,10 @@ def uses_gpu():
     return False
+def jax_uses_tpu():
+    return backend.backend() == "jax" and uses_tpu()
 def uses_tpu():
     # Condition used to skip tests when using the TPU
     try:
@@ -661,13 +665,6 @@ def uses_tpu():
     return False
-def uses_cpu():
-    devices = distribution.list_devices()
-    if any(d.startswith("cpu") for d in devices):
-        return True
-    return False
 def create_keras_tensors(input_shape, dtype, sparse, ragged):
     if isinstance(input_shape, dict):
         return {

@@ -1,7 +1,7 @@
 from keras.src.api_export import keras_export
 # Unique source of truth for the version number.
-__version__ = "3.14.0.dev2026013104"
+__version__ = "3.14.0.dev2026020104"
 @keras_export("keras.version")

keras-nightly 3.14.0.dev2026013104__tar.gz → 3.14.0.dev2026020104__tar.gz

keras-nightly 3.14.0.dev2026013104tar.gz → 3.14.0.dev2026020104tar.gz