Increase tolerance in LinearOperatorUnitaryTest.test_solve.

PiperOrigin-RevId: 453239614

Author: jburnim

tensorflow_probability/python/experimental/linalg/linear_operator_unitary_test.py

@@ -82,11 +82,13 @@ def test_solve(self):
     broadcasted_y = tf.broadcast_to(y, [7, 1, 13, 3])
     self.assertAllClose(
         self.evaluate(tf.linalg.solve(q, broadcasted_y)),
-        self.evaluate(operator.solve(y)))
+        self.evaluate(operator.solve(y)),
+        atol=1e-5, rtol=1e-5)
 
     self.assertAllClose(
         self.evaluate(tf.linalg.solve(q, broadcasted_y, adjoint=True)),
-        self.evaluate(operator.solve(y, adjoint=True)))
+        self.evaluate(operator.solve(y, adjoint=True)),
+        atol=1e-5, rtol=1e-5)
 
 
 if __name__ == '__main__':