jonsafari / nmt-list

A list of Neural MT implementations

Neural Machine Translation Implementations

Contributions

Feel free add to this via a pull request, with each section alphabetically ordered.

Project Stats

• 23 project names include nmt
• 6 project names include seq2seq
• 2 project names include dl4mt
• 3 projects are derived from Kyunghyun Cho's tutorial at the 2015 DL4MT winter school

The Lists

Caffe2-based NMT

Software	Who	Comments
seq2seq	Deepak, Yury

Chainer-based NMT

Software	Who	Comments
attention_is_all_you_need	Sosuke	impements the Transformer
chainn	Philip
KyotoNMT	Fabien

Custom C/C++ NMT

Software	Who	Comments
EUREKA-MangoNMT	Jiajun
Mantidae	Vu	A successor of Mantis
Marian	Marcin, Tomasz	Compatible with Nematus models
N3LP	Kazuma, Akiko
Zoph_RNN	Barret	Supports multi-GPU parallelism

DyNet-based NMT

Software	Who	Comments
dynmt-py	Roee	Python
lamtram	Graham	C++
mantis	Trevor	C++
NMTKit	Yusuke	C++
Transformer-DyNet	Vu	C++; implements the Transformer
xnmt	NeuLab	Python

Keras-based NMT

Software	Who	Comments
Keras seq2seq	Fariz
NMT-Keras	Álvaro

MXNet-based NMT

Software	Who	Comments
MXNMT	magic282
sockeye	AWS Labs

PyTorch-based NMT

Software	Who	Comments
attention-is-all-you-need-pytorch	Yu-Hsiang
fairseq-py	Myle & Facebook Research
Joey NMT	Heidelberg University & University of Amsterdam	minimalist, beginner-friendly
NJUNMT-pytorch	Haoran Wei & Zaixiang Zheng
nmtpytorch	Ozan
nonauto-nmt	Salesforce Research
OpenNMT-py	Harvard, Systran
pytorch_nmt	Pengcheng
seq2seq.pytorch	Elad
UNdreaMT	Mikel	Unsupervised

TensorFlow-based NMT

Software	Who	Comments
byteNet-tensorflow	Paarth
bytenet_translation	Kyubyong
Neural Monkey	ÚFAL
NJUNMT-tf	ZhaoChengqi
NMT	Thang, Eugene, Rui
OpenNMT-tf	Guillaume
OpenSeq2Seq	Kuchaiev
RNNsearch	Playinf, Xiamen University
seq2seq	Alexandre
Tensor2Tensor	Google	Implements many models
tf-seq2seq	Denny	See also tutorial

Theano-based NMT

Note: Theano will be discontinued

Software	Who	Comments
DCNMT	SwordYork
dl4mt-tutorial	Cho, Orhan
dl4mt-c2c	Jason, Cho
HNMT	Robert
Nematus	Rico, Cho, Orhan
neuralmt	Raphael
NMT	Zhaopeng	Adds coverage and context gate
nmtpy	LIUM	refactored dl4mt-tutorial
SGNMT	Cambridge	Syntactically-guided NMT
THUMT	Tsinghua University

Torch-based NMT

Software	Who	Comments
fairseq	Facebook	Novel CNN+attention
nmt-android	Yoon, Sasha
NPMT	Po-Sen, Fairseq team	Phrase-based NMT
OpenNMT	Harvard, Systran
seq2seq-attn	Yoon	Superseded by OpenNMT, except for character-level stuff

Other

Software	Who	Backend	Comments
nmt.hybrid	Thang	Matlab	also word-based
textsum	Google	TensorFlow	text summarization demo using seq2seq+attn

nmt-list's Issues

maybe add https://github.com/facebookresearch/fairseq-py to pytorch list?

https://github.com/facebookresearch/fairseq-py is the currently supported version of this source base..I believe
:-)
d

caffe2 seq2seq link does not exist

https://github.com/caffe2/caffe2/blob/master/caffe2/python/examples/seq2seq.py
this link generated 404

I found
https://github.com/caffe2/caffe2/tree/master/caffe2/python/models/seq2seq
I found
https://github.com/caffe2/caffe2/blob/master/caffe2/python/models/seq2seq/translate.py
but going up one level on this url (to seq2seq)
changes the url to
https://github.com/caffe2/caffe2/tree/master/caffe2/python/models/seq2seq
not sure exactly how this works..but the source suggests it is a translator..but I found no instructions on configuration on invocation.

Texar: a general-purpose text generation toolkit based on TF

The list is awesome. It'd be great if you can add Texar to the list :)

Texar is an open source toolkit aiming to support a broad set of machine learning especially text generation tasks, such as machine translation, dialog, summarization, content manipulation, language modeling, and more. Texar is based on TensorFlow, and is designed to be highly modularized, versatile, and extensible, to support arbitrary model architectures and algorithmic paradigms:


• encoder(s) to decoder(s), sequential- and self-attentions, memory, hierarchical models, classifiers, ...

• maximum likelihood learning, reinforcement learning, adversarial learning, probabilistic modeling, ...

More details:

• Website: https://texar.io
• GitHub: https://github.com/asyml/texar
• Examples: https://github.com/asyml/texar/blob/master/examples
• Documentation: https://texar.readthedocs.io/
• Tech report: https://arxiv.org/pdf/1809.00794.pdf

It is good to mention the source language and target language of translation?

The list should include https://github.com/tensorflow/nmt

as this is good for comparing to sockeye and fairseq

RNNsearch is based on tensorflow..not theano

RNNsearch is based on tensorflow..not theano
see
https://github.com/XMUNLP/RNNsearch/blob/master/model/nmt.py
:-)

s/MT/NLP?

Much thanks for the helpful list

Some items are not MT, for example Google's textsum is not machine translation but text summarisation.

Should the list description be widened or should those items go in a separate list?