Commit 1566cfb9 authored by Peng-Jen Chen's avatar Peng-Jen Chen Committed by Facebook Github Bot
Browse files

Fix multilingual translation bug for to-many case

Summary:
The logic for adding decoder side language token was wrongly implemented.
The way we inject the language token is by replacing the eos symbol with language token symbol. However, the parameter for source / target eos symbol was not set correctly.

Reviewed By: tangyuq

Differential Revision: D17129108

fbshipit-source-id: 6fae385b787370656fd7ca7ab74e6bb91fe5463b
parent 1f0f7cd8
...@@ -214,8 +214,9 @@ class MultilingualTranslationTask(FairseqTask): ...@@ -214,8 +214,9 @@ class MultilingualTranslationTask(FairseqTask):
) )
return self.alter_dataset_langtok( return self.alter_dataset_langtok(
langpair_dataset, langpair_dataset,
src_eos=self.dicts[tgt].eos(), src_eos=self.dicts[src].eos(),
src_lang=src, src_lang=src,
tgt_eos=self.dicts[tgt].eos(),
tgt_lang=tgt, tgt_lang=tgt,
) )
...@@ -239,6 +240,7 @@ class MultilingualTranslationTask(FairseqTask): ...@@ -239,6 +240,7 @@ class MultilingualTranslationTask(FairseqTask):
), ),
src_eos=self.source_dictionary.eos(), src_eos=self.source_dictionary.eos(),
src_lang=self.args.source_lang, src_lang=self.args.source_lang,
tgt_eos=self.target_dictionary.eos(),
tgt_lang=self.args.target_lang, tgt_lang=self.args.target_lang,
), ),
)]), )]),
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment