Closed cocaer closed 5 years ago
I am wondering if the final result in paper uses beam search ? If used, what's the beam size? @glample
No, it uses greedy decoding.
We tried to use beam search 5 or 12 in https://github.com/facebookresearch/XLM which you may want to have a look at since that code gives better results than in this repo.
I am wondering if the final result in paper uses beam search ? If used, what's the beam size? @glample