tutorial
Folders and files
Name | Name | Last commit date | ||
---|---|---|---|---|
parent directory.. | ||||
Simple scripts for training and using EUREKA-MangoNMT --train a Chinese to English NMT in spoken language. The validation perplexity with attention and feed-input should be around 22.84, 24.63 without feed-input ./train_ch2en.sh --given source language sentences, generate the output sequence using beam search. NOTE: odd lines in test data are useless for decoding, they can be blank lines ./generate_ch2en.sh lstm.s2s.z2e.decoder.best lstm.s2s.z2e.encoder.best lstm.s2s.z2e.encoder_reverse.best --given source language sentences, generate the output sequence using beam search and similar training sentences to fine-tune the NMT model (One Sentence One Model for Neural Machine Translation) ./adapt_generate_ch2en.sh lstm.s2s.z2e.decoder.best lstm.s2s.z2e.encoder.best lstm.s2s.z2e.encoder_reverse.best ### More informaiton can refer the command options in codes!