testSortDescending(new byte[] {1, 3, 1}, 0, 1, new byte[] {1, 3, 1});
    testSortDescending(
        new byte[] {GREATEST - 1, 1, GREATEST - 2, 2},
        1,
        3,
        new byte[] {GREATEST - 1, GREATEST - 2, 1, 2});
  }

  public void testNulls() {
    new NullPointerTester().testAllPublicStaticMethods(UnsignedBytes.class);
  }

            # end up building the type adapter from a model field annotation so we
            # need to ignore the warning:
            if shared.PYDANTIC_VERSION_MINOR_TUPLE >= (2, 12):
                from pydantic.warnings import UnsupportedFieldAttributeWarning

                warnings.simplefilter(
                    "ignore", category=UnsupportedFieldAttributeWarning
                )

    testSortDescending(new byte[] {1, 3, 1}, 0, 1, new byte[] {1, 3, 1});
    testSortDescending(
        new byte[] {GREATEST - 1, 1, GREATEST - 2, 2},
        1,
        3,
        new byte[] {GREATEST - 1, GREATEST - 2, 1, 2});
  }

  public void testNulls() {
    new NullPointerTester().testAllPublicStaticMethods(UnsignedBytes.class);
  }

  private static boolean isJava8() {

 * If you use masked losses with Keras the loss values may be different in TensorFlow `2.12` compared to previous versions.
 * In cases where the mask was previously ignored, you will now get an error if you pass a mask with an incompatible shape.

### Major Features and Improvements

`tf.keras`: