diff --git a/tensorflow_probability/python/bijectors/bijector.py b/tensorflow_probability/python/bijectors/bijector.py
index d30183fcec..dd10b51ad5 100644
--- a/tensorflow_probability/python/bijectors/bijector.py
+++ b/tensorflow_probability/python/bijectors/bijector.py
@@ -44,6 +44,7 @@
 ]
 
 JAX_MODE = False
+NUMPY_MODE = False
 SKIP_DTYPE_CHECKS = False
 
 # Singleton object representing "no value", in cases where "None" is meaningful.
@@ -1914,11 +1915,12 @@ def __str__(self):
       maybe_dtype = ''
     if self.forward_min_event_ndims == self.inverse_min_event_ndims:
       maybe_min_ndims = ', min_event_ndims={}'.format(
-          self.forward_min_event_ndims)
+          _unwrap_event_ndims(self.forward_min_event_ndims))
     else:
       maybe_min_ndims = (
           ', forward_min_event_ndims={}, inverse_min_event_ndims={}'.format(
-              self.forward_min_event_ndims, self.inverse_min_event_ndims))
+              _unwrap_event_ndims(self.forward_min_event_ndims),
+              _unwrap_event_ndims(self.inverse_min_event_ndims)))
     maybe_min_ndims = maybe_min_ndims.replace('\'', '')
     return ('tfp.bijectors.{type_name}('
             '"{self_name}"'
@@ -1949,8 +1951,10 @@ def __repr__(self):
                 type_name=type(self).__name__,
                 self_name=self.name or '<unknown>',
                 batch_shape=batch_shape_str,
-                forward_min_event_ndims=self.forward_min_event_ndims,
-                inverse_min_event_ndims=self.inverse_min_event_ndims,
+                forward_min_event_ndims=_unwrap_event_ndims(
+                    self.forward_min_event_ndims),
+                inverse_min_event_ndims=_unwrap_event_ndims(
+                    self.inverse_min_event_ndims),
                 dtype_x=_str_dtype(self.inverse_dtype()),
                 dtype_y=_str_dtype(self.forward_dtype())))
 
@@ -2026,7 +2030,7 @@ def check_valid_ndims(ndims, validate=True):
   assertions = []
 
   shape = ps.shape(ndims)
-  if not tf.is_tensor(shape):
+  if not tf.is_tensor(shape) or NUMPY_MODE or JAX_MODE:
     if shape.tolist():
       raise ValueError('Expected scalar, saw shape {}.'.format(shape))
   elif validate:
@@ -2390,3 +2394,7 @@ def _str_tensorshape(x):
   if tensorshape_util.rank(x) is None:
     return '?'
   return str(tensorshape_util.as_list(x)).replace('None', '?')
+
+
+def _unwrap_event_ndims(ndims):
+  return nest.map_structure(int, ndims)
diff --git a/tensorflow_probability/python/bijectors/sinh_arcsinh.py b/tensorflow_probability/python/bijectors/sinh_arcsinh.py
index 643c2be459..306307dcbc 100644
--- a/tensorflow_probability/python/bijectors/sinh_arcsinh.py
+++ b/tensorflow_probability/python/bijectors/sinh_arcsinh.py
@@ -99,6 +99,7 @@ def __init__(self,
           forward_min_event_ndims=0,
           validate_args=validate_args,
           parameters=parameters,
+          dtype=dtype,
           name=name)
 
   @classmethod
diff --git a/tensorflow_probability/python/distributions/joint_distribution_coroutine_test.py b/tensorflow_probability/python/distributions/joint_distribution_coroutine_test.py
index 93759cc814..35b2385509 100644
--- a/tensorflow_probability/python/distributions/joint_distribution_coroutine_test.py
+++ b/tensorflow_probability/python/distributions/joint_distribution_coroutine_test.py
@@ -989,7 +989,7 @@ def _get_support_bijectors(dists, xs=None, ys=None):
     event_ndims = [0, 1, 0]
     fldj = joint_bijector.forward_log_det_jacobian(xs, event_ndims)
     fldj_fd = _finite_difference_ldj(bijectors, 'forward', xs, delta=0.01)
-    self.assertAllClose(self.evaluate(fldj), self.evaluate(fldj_fd), rtol=1e-5)
+    self.assertAllClose(self.evaluate(fldj), self.evaluate(fldj_fd), rtol=2e-5)
 
     # Test inverse log det Jacobian via finite differences.
     ildj = joint_bijector.inverse_log_det_jacobian(ys, event_ndims)
diff --git a/tensorflow_probability/python/distributions/markov_chain_test.py b/tensorflow_probability/python/distributions/markov_chain_test.py
index a28b8a697c..35c8d9a263 100644
--- a/tensorflow_probability/python/distributions/markov_chain_test.py
+++ b/tensorflow_probability/python/distributions/markov_chain_test.py
@@ -106,7 +106,7 @@ def test_log_prob_matches_linear_gaussian_ssm(self):
         num_steps=7)
 
     x = markov_chain.sample(5, seed=seed)
-    self.assertAllClose(lgssm.log_prob(x), markov_chain.log_prob(x), rtol=1e-4)
+    self.assertAllClose(lgssm.log_prob(x), markov_chain.log_prob(x), rtol=5e-4)
 
   @test_util.numpy_disable_test_missing_functionality(
       'JointDistributionNamedAutoBatched')
diff --git a/tensorflow_probability/python/internal/backend/numpy/numpy_math.py b/tensorflow_probability/python/internal/backend/numpy/numpy_math.py
index 2037d40e7e..7ecc67639f 100644
--- a/tensorflow_probability/python/internal/backend/numpy/numpy_math.py
+++ b/tensorflow_probability/python/internal/backend/numpy/numpy_math.py
@@ -300,7 +300,7 @@ def _invert_permutation(x, name=None):
 def _l2_normalize(x, axis=None, epsilon=1e-12, name=None):  # pylint: disable=unused-argument
   x = _convert_to_tensor(x)
   norm = np.linalg.norm(x, ord=2, axis=_astuple(axis), keepdims=True)
-  norm = np.maximum(norm, np.sqrt(epsilon))
+  norm = np.maximum(norm, np.sqrt(np.asarray(epsilon, dtype=norm.dtype)))
   return x / norm