README.md 3.1 KB
Newer Older
fengzch-das's avatar
fengzch-das committed
1
# <div align="center"><strong>PyTorch-Encoding</strong></div>
Hang Zhang's avatar
Hang Zhang committed
2

fengzch-das's avatar
fengzch-das committed
3
4
## 安装
组件支持组合
5

fengzch-das's avatar
fengzch-das committed
6
7
8
9
10
   | PyTorch版本 | fastpt版本  |PyTorch-Encoding版本      | DTK版本                  | Python版本       | 推荐编译方式 |
   | ----------- | ----------- | ----------- | ------------------------ | -----------------| ------------ |
   | 2.5.1       | 2.1.0       |master       | >= 25.04                 | 3.8、3.10、3.11  | fastpt不转码 |
   | 2.4.1       | 2.0.1       |master       | >= 25.04                 | 3.8、3.10、3.11  | fastpt不转码 |
   | 其他        | 其他         | 其他        | 其他                     | 3.8、3.10、3.11  | hip转码      |
Hang Zhang's avatar
v0.1.0  
Hang Zhang committed
11

fengzch-das's avatar
fengzch-das committed
12
+ pytorch版本大于2.4.1 && dtk版本大于25.04 推荐使用fastpt不转码编译。
Hang Zhang's avatar
v0.2.0  
Hang Zhang committed
13

fengzch-das's avatar
fengzch-das committed
14
15
16
17
18
19
20
21
22
### 1、使用pip方式安装
PyTorch-Encoding whl包下载目录:[光和开发者社区](https://download.sourcefind.cn:65024/4/main),选择对应的pytorch版本和python版本下载对应PyTorch-Encoding的whl包
```shell
pip install torch* (下载torch的whl包)
pip install fastpt* --no-deps (下载fastpt的whl包)
source  /usr/local/bin/fastpt -E
pip install torch_encoding* (下载的PyTorch-Encoding的whl包)
```
### 2、使用源码编译方式安装
Hang Zhang's avatar
v0.1.0  
Hang Zhang committed
23

fengzch-das's avatar
fengzch-das committed
24
25
#### 编译环境准备
提供基于fastpt不转码编译:
Hang Zhang's avatar
Hang Zhang committed
26

fengzch-das's avatar
fengzch-das committed
27
1. 基于光源pytorch基础镜像环境:镜像下载地址:[光合开发者社区](https://sourcefind.cn/#/image/dcu/pytorch),根据pytorch、python、dtk及系统下载对应的镜像版本。
Hang Zhang's avatar
Hang Zhang committed
28

fengzch-das's avatar
fengzch-das committed
29
30
31
32
33
34
35
2. 基于现有python环境:安装pytorch,fastpt whl包下载目录:[光合开发者社区](https://sourcefind.cn/#/image/dcu/pytorch),根据python、dtk版本,下载对应pytorch的whl包。安装命令如下:
```shell
pip install torch* (下载torch的whl包)
pip install fastpt* --no-deps (下载fastpt的whl包, 安装顺序,先安装torch,后安装fastpt)
pip install pytest
pip install wheel
```
Hang Zhang's avatar
v0.1.0  
Hang Zhang committed
36

fengzch-das's avatar
fengzch-das committed
37
38
39
40
#### 源码编译安装
- 代码下载
```shell
git clone http://developer.sourcefind.cn/codes/OpenDAS/PyTorch-Encoding.git # 根据编译需要切换分支
Hang Zhang's avatar
Hang Zhang committed
41
```
fengzch-das's avatar
fengzch-das committed
42
- 提供2种源码编译方式(进入PyTorch-Encoding目录):
Hang Zhang's avatar
Hang Zhang committed
43
```
fengzch-das's avatar
fengzch-das committed
44
45
1. 设置不转码编译环境变量
source /usr/local/bin/fastpt -C
Hang Zhang's avatar
Hang Zhang committed
46

fengzch-das's avatar
fengzch-das committed
47
48
49
50
51
52
2. 编译whl包并安装
python3 setup.py bdist_wheel
pip install dist/torch_encoding* --no-deps

3. 源码编译安装
python3 setup.py install --no-deps
Hang Zhang's avatar
v0.2.0  
Hang Zhang committed
53
```
fengzch-das's avatar
fengzch-das committed
54
55
56
57
#### 注意事项
+ 若使用pip install下载安装过慢,可添加pypi清华源:-i https://pypi.tuna.tsinghua.edu.cn/simple/
+ ROCM_PATH为dtk的路径,默认为/opt/dtk
+ 在pytorch2.5.1环境下编译需要支持c++17语法,打开setup.py文件,把文件中的 -std=c++14 修改为 -std=c++17
Hang Zhang's avatar
init  
Hang Zhang committed
58

fengzch-das's avatar
fengzch-das committed
59
## 验证
Hang Zhang's avatar
init  
Hang Zhang committed
60
```
fengzch-das's avatar
fengzch-das committed
61
62
63
64
65
66
67
python3
Python 3.10.12 (main, May 27 2025, 17:12:29) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import encoding
>>> encoding.__version__
'1.2.2b20250928'
>>> 
Hang Zhang's avatar
init  
Hang Zhang committed
68
```
fengzch-das's avatar
fengzch-das committed
69
70
71
72
73
74
75
76
版本号与官方版本同步,查询该软件的版本号,例如0.3;

## Known Issue
-

## 参考资料
- [README_ORIGIN](README_ORIGIN.md)
- [https://github.com/princeton-vl/PyTorch-Encoding](https://github.com/zhanghang1989/PyTorch-Encoding)