Release 0.7. (FR-EN)
I’m trying to get non-translated words in output but for some reason it doesn’t work. Examples:
SENT 16: 2 Histoire
PRED 16: 2
PRED SCORE: -0.21
SENT 17: 2.1 Origines
PRED 17: 2.1
PRED SCORE: -5.22
By the way I’m using trained model from Release 0.5 but via 0.7 version, is that can be problem?
Does it have back compatibility?
This is related to your previous post:
Does the training data contain short sequences like this?
Also, we don’t need NMT for these examples.