Hello,
so, something doesn't add up.
I defined translation_maxlen to be 100.
Now, I was going through the beam_search_sampler code, and it seems that sentences that are longer than 100 are still being translated. I know that in the training phase, sentences that are longer than maxlen are not being processed at all. Isn't it the same case for the inference phase with translation_maxlen? I am specifically talking about source language sentences that are longer than translation_maxlen.
Thanks!
--translation_maxlen controls the Maximum length of translation output sentence. There's currently no mechanism to exclude long source sentences from translation.
Hello, so, something doesn't add up. I defined translation_maxlen to be 100. Now, I was going through the beam_search_sampler code, and it seems that sentences that are longer than 100 are still being translated. I know that in the training phase, sentences that are longer than maxlen are not being processed at all. Isn't it the same case for the inference phase with translation_maxlen? I am specifically talking about source language sentences that are longer than translation_maxlen. Thanks!