AttributeError: 'MirroredStrategy' object has no attribute 'experimental_run_v2'

[lylyjy]

@tf.function() def train(self, data, log_images=False):

strategy.run

self._strategy.experimental_run_v2(self._train, args=(data, log_images))

[danijar]

You're using a newer version of TF than is listed in the readme. The method is called run() now.

That said, I highly recommend giving DreamerV2 a try. The code base also supports Gaussian latents with a flag.