Inconsistent behavior tfp.distributions.Bernoulli tf.float vs. np.float

Since the Bernoulli distribution is implemented for real numbers, I was testing a bit to see what it will do. There, I found an interesting behavior difference between tf.floats, python floats and numpy floats.
This test was done with the following versions:
numpy = 1.17.4
tensorflow = 1.14.0
tensorflow_probability = 0.7.0

Both behaviors make sense, but an inconsistency between numpy floats and tf/python floats seems weird.

I used the following code to run the experiments:

import numpy as np
import tensorflow as tf
import tensorflow_probability as tfp
import matplotlib.pyplot as plt

prob = np.array([0.9])
bern = tfp.distributions.Bernoulli(probs=prob)

x = np.arange(-2, 2, 0.01)
numpy_64_x = np.array(x, dtype=np.float64)
numpy_32_x = np.array(x, dtype=np.float32)
tf_64_x = tf.constant(x, dtype=tf.float64)
python_x = [float(el) for el in x]

sess = tf.Session()
sess.run(tf.global_variables_initializer())

numpy_64_y = sess.run(bern.prob(numpy_64_x))
numpy_32_y = sess.run(bern.prob(numpy_32_x))
tf_64_y = sess.run(bern.prob(tf_64_x))
python_y = sess.run(bern.prob(python_x))

# Just for plotting
lw = 5
plt.plot(x, numpy_64_y, label="np.float64", lw=lw)
plt.plot(x, numpy_32_y, label="np.float32", lw=lw, ls=":", zorder=10)
plt.plot(x, python_y, label="python float", lw=lw)
plt.plot(x, tf_64_y, label="tf.float64", lw=lw, ls=":", zorder=20)
ax = plt.gca()
ax.legend()
plt.show()

PS: tf.float64 had the same behavior as tf.float32
Same behavior for tf.contrib.distributions.Bernoulli

Start by running the supplied Python example with the listed NumPy, TensorFlow, and TensorFlow Probability versions, then inspect Bernoulli.prob handling for NumPy, Python, and TensorFlow float inputs. Done means the intended behavior is established and the inconsistent result is either corrected or documented with a regression check for the affected input types.