Placing model on bfloat16 on CPU make it freeze/hang · Issue https://github.com/pytorch/pytorch/issues/75458