Closed lordtt13 closed 4 years ago
Thanks for your work. However, I want the language model to be integrated into the decoding algorithm to take full advantage of the information the neural network provides (like word beam search does). Therefore not integrating this PR into master.
The autocorrect is simply meant to augment final predictions while not disturbing the decoding algorithm the neural network provides.
Autocorrect Added and model and earlystopping reduced