Comment by tomtom1337

Is it possible to "pretend" that one's float16 datatype is in fact bfloat16? Since they occupy the same about of memory?