# transformer

**Repository Path**: null_998_1909/transformer

## Basic Information

- **Project Name**: transformer
- **Description**: nlp 演示代码
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 1
- **Created**: 2024-05-07
- **Last Updated**: 2024-10-29

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

## 论文来源

[Attention is all you need](https://arxiv.org/abs/1706.03762)

## 代码参考

[哈弗 nlp](http://nlp.seas.harvard.edu/2018/04/03/attention.html)

## 代码解释
https://blog.csdn.net/weixin_44343282/article/details/124575684

## 项目结构

- data `源数据目录`
- log  `日志存放目录 （每次预测产生一个 log-timestamp.txt）`
- save `模型存放目录`
- model `模型目录`
    - attention.py
    - embedding.py
    - encoder.py
    - decoder.py
    - generator.py
    - sublayer.py
    - position_wise_feedforward.py
    - transformer.py
- lib  `损失函数、优化器等存放位置`
    - criterion.py `损失函数`
    - optimizer.py `优化器`
    - loss.py `优化器 + 损失函数封装类`
- evaluate.py `预测.py`
- train.py `训练.py`
- parser.py `参数.py`
- utils.py `工具类.py`
- run.py `入口文件.py`
- README.md `readme`


## 训练
`python3 run.py`

## 预测 (前提：训练过)
`python3 run.py --type evaluate`