I tried to change the cuddn version, but it didn’t work. It should be a matter of memory, as follows:

tensorflow：

config = tf.ConfigProto()
config.gpu_options.allow_growth = True
with tf.Session(config=config) as session:

Hard

config = tf.ConfigProto()
config.gpu_options.allow_growth = True
keras.backend.tensorflow_backend.set_session(tf.Session(config=config))

My environment is cuddn7.6 + cuda10.0 + Python 3.6 2080ti, so it can run