README-zh-Hans.md 9.28 KB
Newer Older
binmakeswell's avatar
binmakeswell committed
1
# Colossal-AI
2
<div id="top" align="center">
binmakeswell's avatar
binmakeswell committed
3

4
5
   [![logo](https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/Colossal-AI_logo.png)](https://www.colossalai.org/)

binmakeswell's avatar
binmakeswell committed
6
   Colossal-AI: 一个面向大模型时代的通用深度学习系统
binmakeswell's avatar
binmakeswell committed
7
8
9

   <h3> <a href="https://arxiv.org/abs/2110.14883"> 论文 </a> | 
   <a href="https://www.colossalai.org/"> 文档 </a> | 
10
   <a href="https://github.com/hpcaitech/ColossalAI-Examples"> 例程 </a> |   
binmakeswell's avatar
binmakeswell committed
11
   <a href="https://github.com/hpcaitech/ColossalAI/discussions"> 论坛 </a> | 
12
   <a href="https://medium.com/@hpcaitech"> 博客 </a></h3>
binmakeswell's avatar
binmakeswell committed
13

Frank Lee's avatar
Frank Lee committed
14
   [![Build](https://github.com/hpcaitech/ColossalAI/actions/workflows/build.yml/badge.svg)](https://github.com/hpcaitech/ColossalAI/actions/workflows/build.yml)
binmakeswell's avatar
binmakeswell committed
15
   [![Documentation](https://readthedocs.org/projects/colossalai/badge/?version=latest)](https://colossalai.readthedocs.io/en/latest/?badge=latest)
16
   [![CodeFactor](https://www.codefactor.io/repository/github/hpcaitech/colossalai/badge)](https://www.codefactor.io/repository/github/hpcaitech/colossalai)
Frank Lee's avatar
Frank Lee committed
17
   [![HuggingFace badge](https://img.shields.io/badge/%F0%9F%A4%97HuggingFace-Join-yellow)](https://huggingface.co/hpcai-tech)
binmakeswell's avatar
binmakeswell committed
18
   [![slack badge](https://img.shields.io/badge/Slack-join-blueviolet?logo=slack&amp)](https://join.slack.com/t/colossalaiworkspace/shared_invite/zt-z7b26eeb-CBp7jouvu~r0~lcFzX832w)
19
   [![WeChat badge](https://img.shields.io/badge/微信-加入-green?logo=wechat&amp)](https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/WeChat.png)
binmakeswell's avatar
binmakeswell committed
20
21

   | [English](README.md) | [中文](README-zh-Hans.md) |
22

binmakeswell's avatar
binmakeswell committed
23
</div>
24
25
26
27


## 目录
<ul>
binmakeswell's avatar
binmakeswell committed
28
 <li><a href="#为何选择-Colossal-AI">为何选择 Colossal-AI</a> </li>
29
30
 <li><a href="#特点">特点</a> </li>
 <li>
binmakeswell's avatar
binmakeswell committed
31
   <a href="#并行训练样例展示">并行训练样例展示</a> 
32
33
34
35
36
   <ul>
     <li><a href="#ViT">ViT</a></li>
     <li><a href="#GPT-3">GPT-3</a></li>
     <li><a href="#GPT-2">GPT-2</a></li>
     <li><a href="#BERT">BERT</a></li>
binmakeswell's avatar
binmakeswell committed
37
     <li><a href="#PaLM">PaLM</a></li>
38
39
   </ul>
 </li>
40
<li>
binmakeswell's avatar
binmakeswell committed
41
   <a href="#单GPU训练样例展示">单GPU训练样例展示</a> 
42
43
44
45
46
   <ul>
     <li><a href="#GPT-2-Single">GPT-2</a></li>
     <li><a href="#PaLM-Single">PaLM</a></li>
   </ul>
 </li>
binmakeswell's avatar
binmakeswell committed
47
48
49
50
51
52
<li>
   <a href="#推理样例展示">推理样例展示</a> 
   <ul>
     <li><a href="#GPT-3-Inference">GPT-3</a></li>
   </ul>
 </li>
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
 <li>
   <a href="#安装">安装</a>
   <ul>
     <li><a href="#PyPI">PyPI</a></li>
     <li><a href="#从源代码安装">从源代码安装</a></li>
   </ul>
 </li>
 <li><a href="#使用-Docker">使用 Docker</a></li>
 <li><a href="#社区">社区</a></li>
 <li><a href="#做出贡献">做出贡献</a></li>
 <li><a href="#快速预览">快速预览</a></li>
   <ul>
     <li><a href="#几行代码开启分布式训练">几行代码开启分布式训练</a></li>
     <li><a href="#构建一个简单的2维并行模型">构建一个简单的2维并行模型</a></li>
   </ul>
 <li><a href="#引用我们">引用我们</a></li>
</ul>
binmakeswell's avatar
binmakeswell committed
70

binmakeswell's avatar
binmakeswell committed
71
72
73
74
75
76
77
78
79
80
81
## 为何选择 Colossal-AI
<div align="center">
   <a href="https://youtu.be/KnXSfjqkKN0">
   <img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/JamesDemmel_Colossal-AI.png" width="600" />
   </a>

   James Demmel 教授 (加州大学伯克利分校): Colossal-AI 让分布式训练高效、易用、可扩展。
</div>

<p align="right">(<a href="#top">返回顶端</a>)</p>

binmakeswell's avatar
binmakeswell committed
82
83
## 特点

binmakeswell's avatar
binmakeswell committed
84
Colossal-AI 为您提供了一系列并行组件。我们的目标是让您的分布式 AI 模型像构建普通的单 GPU 模型一样简单。我们提供的友好工具可以让您在几行代码内快速开始分布式训练和推理。
binmakeswell's avatar
binmakeswell committed
85

binmakeswell's avatar
binmakeswell committed
86
87
88
89
90
- 并行化策略
  - 数据并行
  - 流水线并行
  - 1维, [2维](https://arxiv.org/abs/2104.05343), [2.5维](https://arxiv.org/abs/2105.14500), [3维](https://arxiv.org/abs/2105.14450) 张量并行
  - [序列并行](https://arxiv.org/abs/2105.13120)
binmakeswell's avatar
binmakeswell committed
91
  - [零冗余优化器 (ZeRO)](https://arxiv.org/abs/1910.02054)
binmakeswell's avatar
binmakeswell committed
92
93
94
95
- 异构内存管理
  - [PatrickStar](https://arxiv.org/abs/2108.05818)
- 使用友好
  - 基于参数文件的并行化
binmakeswell's avatar
binmakeswell committed
96
97
- 推理
  - [Energon-AI](https://github.com/hpcaitech/EnergonAI)
98
99
<p align="right">(<a href="#top">返回顶端</a>)</p>

binmakeswell's avatar
binmakeswell committed
100
## 并行训练样例展示
binmakeswell's avatar
binmakeswell committed
101
### ViT
binmakeswell's avatar
binmakeswell committed
102
<p align="center">
103
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/ViT.png" width="450" />
binmakeswell's avatar
binmakeswell committed
104
</p>
105

106
- 14倍批大小和5倍训练速度(张量并行=64)
binmakeswell's avatar
binmakeswell committed
107

108
### GPT-3
binmakeswell's avatar
binmakeswell committed
109
<p align="center">
110
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/GPT3.png" width=700/>
binmakeswell's avatar
binmakeswell committed
111
</p>
binmakeswell's avatar
binmakeswell committed
112

113
- 释放 50% GPU 资源占用, 或 10.7% 加速
114
115

### GPT-2
116
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/GPT2.png" width=800/>
117

binmakeswell's avatar
binmakeswell committed
118
- 降低11倍 GPU 显存占用,或超线性扩展(张量并行)
119

Sze-qq's avatar
Sze-qq committed
120
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/(updated)GPT-2.png" width=800>
121

binmakeswell's avatar
binmakeswell committed
122
- 用相同的硬件训练24倍大的模型
Sze-qq's avatar
Sze-qq committed
123
- 超3倍的吞吐量 
binmakeswell's avatar
binmakeswell committed
124
125

### BERT
126
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/BERT.png" width=800/>
binmakeswell's avatar
binmakeswell committed
127

128
- 2倍训练速度,或1.5倍序列长度
binmakeswell's avatar
binmakeswell committed
129

binmakeswell's avatar
binmakeswell committed
130
131
132
### PaLM
- [PaLM-colossalai](https://github.com/hpcaitech/PaLM-colossalai): 可扩展的谷歌 Pathways Language Model ([PaLM](https://ai.googleblog.com/2022/04/pathways-language-model-palm-scaling-to.html)) 实现。

binmakeswell's avatar
binmakeswell committed
133
134
请访问我们的[文档和教程](https://www.colossalai.org/)以了解详情。

135
<p align="right">(<a href="#top">返回顶端</a>)</p>
binmakeswell's avatar
binmakeswell committed
136

binmakeswell's avatar
binmakeswell committed
137
## 单GPU训练样例展示
binmakeswell's avatar
binmakeswell committed
138

139
140
141
142
### GPT-2
<p id="GPT-2-Single" align="center">
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/GPT2-GPU1.png" width=450/>
</p>
binmakeswell's avatar
binmakeswell committed
143

binmakeswell's avatar
binmakeswell committed
144
- 用相同的硬件训练20倍大的模型
binmakeswell's avatar
binmakeswell committed
145

146
147
148
149
### PaLM
<p id="PaLM-Single" align="center">
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/PaLM-GPU1.png" width=450/>
</p>
binmakeswell's avatar
binmakeswell committed
150

binmakeswell's avatar
binmakeswell committed
151
152
153
154
155
156
157
158
159
160
161
162
163
- 用相同的硬件训练34倍大的模型

<p align="right">(<a href="#top">back to top</a>)</p>


## 推理样例展示

### GPT-3
<p id="GPT-3-Inference" align="center">
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/inference_GPT-3.jpg" width=800/>
</p>

- [Energon-AI](https://github.com/hpcaitech/EnergonAI) :用相同的硬件推理加速50%
164
165
166

<p align="right">(<a href="#top">back to top</a>)</p>

binmakeswell's avatar
binmakeswell committed
167

168
169
170
## 安装

### 从官方安装
binmakeswell's avatar
binmakeswell committed
171

binmakeswell's avatar
binmakeswell committed
172
您可以访问我们[下载](https://www.colossalai.org/download)页面来安装Colossal-AI,在这个页面上发布的版本都预编译了CUDA扩展。
binmakeswell's avatar
binmakeswell committed
173

174
175
176
### 从源安装

> 此文档将与版本库的主分支保持一致。如果您遇到任何问题,欢迎给我们提 issue :)
binmakeswell's avatar
binmakeswell committed
177
178
179
180

```shell
git clone https://github.com/hpcaitech/ColossalAI.git
cd ColossalAI
181
182

# install dependency
binmakeswell's avatar
binmakeswell committed
183
184
pip install -r requirements/requirements.txt

185
# install colossalai
binmakeswell's avatar
binmakeswell committed
186
187
188
pip install .
```

189
如果您不想安装和启用 CUDA 内核融合(使用融合优化器时强制安装):
binmakeswell's avatar
binmakeswell committed
190
191

```shell
192
NO_CUDA_EXT=1 pip install .
binmakeswell's avatar
binmakeswell committed
193
194
```

195
196
<p align="right">(<a href="#top">返回顶端</a>)</p>

binmakeswell's avatar
binmakeswell committed
197
198
## 使用 Docker

binmakeswell's avatar
binmakeswell committed
199
运行以下命令从我们提供的 docker 文件中建立 docker 镜像。
binmakeswell's avatar
binmakeswell committed
200

201
202
203
> 在Dockerfile里编译Colossal-AI需要有GPU支持,您需要将Nvidia Docker Runtime设置为默认的Runtime。更多信息可以点击[这里](https://stackoverflow.com/questions/59691207/docker-build-with-nvidia-runtime)。
> 我们推荐从[项目主页](https://www.colossalai.org)直接下载Colossal-AI.

binmakeswell's avatar
binmakeswell committed
204
205
206
207
208
```bash
cd ColossalAI
docker build -t colossalai ./docker
```

binmakeswell's avatar
binmakeswell committed
209
运行以下命令从以交互式启动 docker 镜像.
binmakeswell's avatar
binmakeswell committed
210
211
212
213
214

```bash
docker run -ti --gpus all --rm --ipc=host colossalai bash
```

215
<p align="right">(<a href="#top">返回顶端</a>)</p>
binmakeswell's avatar
binmakeswell committed
216
217
218
219

## 社区
欢迎通过[论坛](https://github.com/hpcaitech/ColossalAI/discussions),
[Slack](https://join.slack.com/t/colossalaiworkspace/shared_invite/zt-z7b26eeb-CBp7jouvu~r0~lcFzX832w),
binmakeswell's avatar
binmakeswell committed
220
[微信](https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/WeChat.png "qrcode")加入 Colossal-AI 社区,与我们分享你的建议和问题。
binmakeswell's avatar
binmakeswell committed
221
222


binmakeswell's avatar
binmakeswell committed
223
224
225
226
## 做出贡献

欢迎为该项目做出贡献,请参阅[贡献指南](./CONTRIBUTING.md)

binmakeswell's avatar
binmakeswell committed
227
228
229
230
231
232
真诚感谢所有贡献者!

<a href="https://github.com/hpcaitech/ColossalAI/graphs/contributors"><img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/contributor_avatar.png" width="800px"></a>

*贡献者头像的展示顺序是随机的。*

233
<p align="right">(<a href="#top">返回顶端</a>)</p>
binmakeswell's avatar
binmakeswell committed
234
235
236

## 快速预览

237
### 几行代码开启分布式训练
binmakeswell's avatar
binmakeswell committed
238
239

```python
240
241
242
parallel = dict(
    pipeline=2,
    tensor=dict(mode='2.5d', depth = 1, size=4)
binmakeswell's avatar
binmakeswell committed
243
244
245
)
```

246
### 几行代码开启异构训练
binmakeswell's avatar
binmakeswell committed
247
248

```python
249
250
251
252
253
254
255
256
zero = dict(
    model_config=dict(
        tensor_placement_policy='auto',
        shard_strategy=TensorShardStrategy(),
        reuse_fp16_shard=True
    ),
    optimizer_config=dict(initial_scale=2**5, gpu_margin_mem_ratio=0.2)
)
binmakeswell's avatar
binmakeswell committed
257
258
```

259
<p align="right">(<a href="#top">返回顶端</a>)</p>
binmakeswell's avatar
binmakeswell committed
260

261
## 引用我们
binmakeswell's avatar
binmakeswell committed
262
263
264
265
266
267
268
269
270

```
@article{bian2021colossal,
  title={Colossal-AI: A Unified Deep Learning System For Large-Scale Parallel Training},
  author={Bian, Zhengda and Liu, Hongxin and Wang, Boxiang and Huang, Haichen and Li, Yongbin and Wang, Chuanrui and Cui, Fan and You, Yang},
  journal={arXiv preprint arXiv:2110.14883},
  year={2021}
}
```
271

Jie Zhu's avatar
Jie Zhu committed
272
<p align="right">(<a href="#top">返回顶端</a>)</p>