PyTorch Models

学習済みモデル

Available models trained using OpenNMT.

Benchmarks

This page benchmarks training results of open-source NMT systems with generated models of OpenNMT and other systems.

English-> German (WMT)

Who/When Corpus Prep Training Tool Training Parameters Server Details Training Time/Memory Translation Parameters Scores Model
2018/03/15
Baseline
WMT OpenNMT 6 layers, LSTM 512, BPE, Transformer     here BLEU Score: WMT14 26.89 WMT17: 28.09 here

German->English

Who/When Corpus Prep Training Tool Training Parameters Server Details Training Time/Memory Translation Parameters Scores Model
2018/02/11
Baseline
IWSLT ‘14 DE-EN OpenNMT d4ab35a 2 layers, LSTM 500, WE 500, encoder_type brnn input feed
20 epochs
Trained on 1 GPU TITAN X     BLEU Score: 30.33 203MB here

English Summarization

Who/When Corpus Prep Training Tool Training Parameters Server Details Training Time/Memory Translation Parameters Scores Model
2018/02/11
Baseline
Gigaword Standard OpenNMT d4ab35a 2 layers, LSTM 500, WE 500, input feed
20 epochs
Trained on 1 GPU TITAN X     Gigaword F-Score R1: 33.60 R2: 16.29 RL: 31.45 331MB here
2018/02/22
Baseline
Gigaword Standard OpenNMT 338b3b1 2 layers, LSTM 500, WE 500, input feed, copy_attn, reuse_copy_attn
20 epochs
Trained on 1 GPU TITAN X   replace_unk Gigaword F-Score R1: 35.51 R2: 17.35 RL: 33.17 331MB here
Who/When Corpus Prep Training Tool Training Parameters Server Details Training Time/Memory Translation Parameters Scores Model
2018/03/20
CNN/Daily Mail OpenNMT Transfomer 6x512 Trained on 1 GPU TITAN X   here Gigaword F-Score R1: R2: RL: 1.1GB here
2018/03/20
CNN/Daily Mail OpenNMT 1 layers BiLSTM 512 Trained on 1 GPU TITAN X     Gigaword F-Score R1: 39.12 R2: 17.35 RL: 36.12 900MB here

Chinese Summarization

Who/When Corpus Prep Training Tool Training Parameters Server Details Training Time/Memory Translation Parameters Scores Model
playma 2018/02/25 LCSTS
src_vocab_size 8000, tgt_vocab_size 8000, src_seq_length 400, tgt_seq_length 30, src_seq_length_trunc 400, tgt_seq_length_trunc 100
OpenNMT 338b3b1 1 layer, LSTM 300, WE 500, encoder_type brnn, input feed
AdaGrad, adagrad_accumulator_init 0.1, learning_rate 0.15
30 epochs
      Gigaword F-Score R1: 35.67 R2: 23.06 RL: 33.14 99MB here

Dialog System

Who/When Corpus Prep Training Tool Training Parameters Server Details Training Time/Memory Translation Parameters Scores Model
2018/02/22
Baseline
Opensubtitles OpenNMT 338b3b1 2 layers, LSTM 500, WE 500, input feed, dropout 0.2, global_attention mlp, start_decay_at 7
13 epochs
Trained on 1 GPU TITAN X     TBD 355MB here