search

ictnlp / DST

DST is a Decoder-only simultaneous machine translation model, which can conduct policy decision and translation concurrently
MIT License
6 stars 1 forks source link

pretrained model #1

Open KeyBoarder8 opened 3 months ago

KeyBoarder8 commented 3 months ago

Thanks for open source. Some questions: Should pretrained model in phase 1(Full-Sentence Pre-training) be a finetuned model for phase 2(Simultaneous Translation Fine-Tuning). For example, use a parameter “--restore-file” in fairseq to train the model phrase 2?

ShoutaoGuo commented 3 weeks ago

Yes, the model of the first state is the begining of the second state