Industrial-grade implementation of seq2seq algorithm based on Pytorch, integrated beam search algorithm.
Project description
seq2seq
Industrial-grade implementation of seq2seq algorithm based on Pytorch, integrated beam search algorithm.
seq2seq is based on other excellent open source projects, this project has the following highlights:
- easy to train, predict and deploy;
- lightweight implementation;
- multitasking support (including dialogue generation and machine translation).
Model description
- Encoder: Bidirectional GRU
- Decoder: GRU with Attention Mechanism
- Bahdanau Attention: Neural Machine Translation by Jointly Learning to Align and Translate
- Luong Attention: Effective Approaches to Attention-based Neural Machine Translation
- Diversity Promoting Beam Search: A Simple, Fast Diverse Decoding Algorithm for Neural Generation
Install
seq2seq is dependent on PyTorch. Two ways to install:
Install seq2seq from Pypi:
pip install seq2seq-pytorch
Install seq2seq from the Github source:
git clone https://github.com/Chiang97912/seq2seq.git
cd seq2seq
python setup.py install
Usage
Train
from seq2seq.model import Seq2Seq
sources = ['...']
targets = ['...']
model = Seq2Seq('seq2seq-model', embed_size=256, hidden_size=512, lang4src='en', lang4tgt='en', device='cuda:0')
model.fit(sources, targets, epochs=20, batch_size=64)
Test
from seq2seq.model import Seq2Seq
model = Seq2Seq('seq2seq-model')
outputs = model.generate('...', beam_size=3, method='greedy')
print(outputs)
Dependencies
python
version 3.6pyTorch
version 1.9.0torchtext
version 0.3.1numpy
version 1.19.5nltk
version 3.5jieba
version 0.42.1
References
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
seq2seq-pytorch-0.1.2.tar.gz
(9.3 kB
view hashes)
Built Distribution
Close
Hashes for seq2seq_pytorch-0.1.2-py2.py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 82df2014028960e6dad306103f6e8f70dc71e0d88892c9427ef358b2fb549d7f |
|
MD5 | 278f38e856f6b6f3d0e63cfbc10a6b6b |
|
BLAKE2b-256 | 7cf912a41d3dc7da8729249f5d843928e3d9e419d92ae86ce8f61ddf1fcec016 |