Skip to content

Files

Latest commit

bd8da7c · Feb 22, 2024

History

History
62 lines (49 loc) · 1.6 KB

README.md

File metadata and controls

62 lines (49 loc) · 1.6 KB

A Transformer Framework Based Translation Task

一个基于Transformer网络结构的文本翻译模型

论文Attention Is All You Need 基于PyTorch的实现

1. 环境准备

  • Python==3.12.2
  • PyTorch==2.2.0
  • torchtext==0.16.2

spacy model: de_core_news_sm 3.7.0 en-core-web-sm 3.7.1

2. 使用方法

2.1 训练

直接执行如下命令即可进行模型训练:

python train.py

训练过程:

Epoch: 2, Train loss: 5.685, Train acc: 0.240947
Epoch: 2, Train loss: 5.668, Train acc: 0.241493
Epoch: 2, Train loss: 5.714, Train acc: 0.224682
Epoch: 2, Train loss: 5.660, Train acc: 0.235888
Epoch: 2, Train loss: 5.584, Train acc: 0.242052
Epoch: 2, Train loss: 5.611, Train acc: 0.243428

学习率变化:

2.2 预测(inference)

直接运行如下命令即可:

python translate.py

示例结果:

德语Eine Gruppe von Menschen steht vor einem Iglu.
翻译A group of people standing in fraon of an igloo .
英语A group of people are facing an igloo.
=========
德语Ein Mann in einem blauen Hemd steht auf einer Leiter und putzt ein Fenster.
翻译A man in a blue shirt is standing on a ladder cleaning a window.
英语A man in a blue shirt is standing on a ladder cleaning a window.

3. 结果

bleu评测结果

val test_2016_flickr