For sure but the big conflict here or the elephant in the room: it makes it a lot harder to trace/follow a models thinking.
I'm not sure though if it wouldn't be possible to have a debug flag and copy&paste the state and do a quick debug output on an embedding layer just for that.
I'm not sure though if it wouldn't be possible to have a debug flag and copy&paste the state and do a quick debug output on an embedding layer just for that.