Removes/relaxes dtype requirements from kernel in `multitask_gaussi…

…an_process_regression_model.py`

PiperOrigin-RevId: 715188555

tensorflow_probability/python/experimental/distributions/multitask_gaussian_process_regression_model.py

@@ -67,7 +67,8 @@ def _flattened_conditional_mean_fn_helper(
   observations = tf.convert_to_tensor(observations)
   if observation_index_points is not None:
     observation_index_points = nest_util.convert_to_nested_tensor(
-        observation_index_points, dtype=kernel.dtype, allow_packing=True)
+        observation_index_points, dtype_hint=kernel.dtype, allow_packing=True
+    )
 
   k_x_obs_linop = kernel.matrix_over_all_tasks(x, observation_index_points)
   if solve_on_observations is None:
@@ -296,12 +297,13 @@ def __init__(self,
 
       input_dtype = dtype_util.common_dtype(
           dict(
-              kernel=kernel,
               index_points=index_points,
               observation_index_points=observation_index_points,
           ),
           dtype_hint=nest_util.broadcast_structure(
-              kernel.feature_ndims, tf.float32))
+              kernel.feature_ndims, tf.float32
+          ),
+      )
 
       # If the input dtype is non-nested float, we infer a single dtype for the
       # input and the float parameters, which is also the dtype of the MTGP's
@@ -573,9 +575,11 @@ def precompute_regression_model(
     with tf.name_scope(name) as name:
       if tf.nest.is_nested(kernel.feature_ndims):
         input_dtype = dtype_util.common_dtype(
-            [kernel, index_points, observation_index_points],
+            [index_points, observation_index_points],
             dtype_hint=nest_util.broadcast_structure(
-                kernel.feature_ndims, tf.float32))
+                kernel.feature_ndims, tf.float32
+            ),
+        )
         dtype = dtype_util.common_dtype(
             [observations, observation_noise_variance,
              predictive_noise_variance], tf.float32)