Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

fluid distribute doc #9288

Merged
merged 15 commits into from
Mar 21, 2018
143 changes: 143 additions & 0 deletions doc/fluid/howto/cluster/fluid_cluster_train_cn.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,143 @@
# Fluid 分布式版本使用指南
本篇文章将说明在PaddlePaddle Fluid版本下进行分布式训练的配置和执行

## 准备工作
* 可用的集群

包含一个或多个计算节点的集群,每一个节点都能够执行PaddlePaddle的训练任务且拥有唯一的IP地址,集群内的所有计算节点可以通过网络相互通信。
* 安装PaddlePaddle Fluid with Distributed版本
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Distributed
=>

Distribution

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done


所有的计算节点上均需要按照分布式版本的PaddlePaddle, 在用于GPU等设备的机器上还需要额外安装好相应的驱动程序和CUDA的库。
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里需要空一行,否则会和上一行显示在一起。

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

**注意:**当前对外提供的PaddlePaddle版本并不支持分布式,需要通过源码重新编译。编译和安装方法参见[编译和安装指南](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/build_and_install/index_en.html)。
cmake编译命令中需要将WITH_DISTRIBUTE设置为ON,下面是一个cmake编译指令示例:
```
cmake .. -DWITH_DOC=OFF -DWITH_GPU=OFF -DWITH_DISTRIBUTE=ON -DWITH_SWIG_PY=ON -DWITH_PYTHON=ON
```

## 更新训练脚本
这里,我们以[Deep Learing 101](http://www.paddlepaddle.org/docs/develop/book/01.fit_a_line/index.html)课程中的第一章 fit a line 为例。
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

xxx为例。。后面应该有具体做的事情,例如:
描述如何将单机训练脚本改造成支持集群训练的版本。

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

### 单机训练脚本示例
```python
import paddle.v2 as paddle
import paddle.fluid as fluid

x = fluid.layers.data(name='x', shape=[13], dtype='float32')
y_predict = fluid.layers.fc(input=x, size=1, act=None)
y = fluid.layers.data(name='y', shape=[1], dtype='float32')

cost = fluid.layers.square_error_cost(input=y_predict, label=y)
avg_cost = fluid.layers.mean(x=cost)

sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001)
sgd_optimizer.minimize(avg_cost)

BATCH_SIZE = 20

train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.uci_housing.train(), buf_size=500),
batch_size=BATCH_SIZE)

place = fluid.CPUPlace()
feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
exe = fluid.Executor(place)

exe.run(fluid.default_startup_program())

PASS_NUM = 100
for pass_id in range(PASS_NUM):
fluid.io.save_persistables(exe, "./fit_a_line.model/")
fluid.io.load_persistables(exe, "./fit_a_line.model/")
for data in train_reader():
avg_loss_value, = exe.run(fluid.default_main_program(),
feed=feeder.feed(data),
fetch_list=[avg_cost])

if avg_loss_value[0] < 10.0:
exit(0) # if avg cost less than 10.0, we think our code is good.
exit(1)
```

我们创建了一个简单的全连接神经网络程序,并且通过Fluid的Executor执行了100次迭代,现在我们需要将该单机版本的程序更新为分布式版本的程序。
### 介绍Parameter Server
在非分布式版本的训练脚本中,只存在Trainer一种角色,它不仅处理常规的计算任务,也处理参数相关的计算和保存任务。在分布式版本的训练过程中,由于存在多个Trainer节点进行同样的数据计算任务,因此需要有一个中心化的节点来统一处理参数相关的保存和分配。在PaddlePaddle中,我们称这样的节点为Parameter Server, [Parameter Server 设计文档](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/fluid/design/dist_train/parameter_server.md)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

因此需要有一个中心化的节点来统一处理参数相关的保存和分配

这里关于PServer的简短描述不是很准确,在Fluid中,PServer还负责参数的优化。

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done


**因此,在分布式的Fluid环境中,我们有两个角色需要创建,分别是Parameter Server和Trainer。**

### 分布式训练
Fliud专门提供了工具[Distributed Transpiler](https://github.com/PaddlePaddle/Paddle/blob/ba65d54d9d3b41cd3c5171b00f476d4e60133ddb/doc/fluid/design/dist_train/distributed_architecture.md#distributed-transpiler)用于将单机版的训练程序转换为分布式版本的训练程序。工具背后的理念是找出程序的优化算子和梯度参数,将他们分隔为两部分,通过send/recive 操作算子进行连接,优化算子和梯度参数可以在优化器的minimize函数的返回值中获取到。
```python
optimize_ops, params_grads = sgd_optimizer.minimize(avg_cost)
```
将Distributed Transpiler、优化算子和梯度函数放在一个代码中如下:
```python
... #define the program, cost, and create sgd optimizer

optimize_ops, params_grads = sgd_optimizer.minimize(avg_cost) #get optimize OPs and gradient parameters

t = fluid.DistributeTranspiler() # create the transpiler instance
# slice the program into 2 pieces with optimizer_ops and gradient parameters list, as well as pserver_endpoints, which is a comma separated list of [IP:PORT] and number of trainers
t.transpile(optimize_ops, params_grads, pservers=pserver_endpoints, trainers=2)

... #create executor

# in pserver, run this
#current_endpoint here means current pserver IP:PORT you wish to run on
pserver_prog = t.get_pserver_program(current_endpoint)
pserver_startup = t.get_startup_program(current_endpoint, pserver_prog)
exe.run(pserver_startup)
exe.run(pserver_prog)

# in trainer, run this
... # define data reader
exe.run(fluid.default_startup_program())
for pass_id in range(100):
for data in train_reader():
exe.run(t.get_trainer_program())
```
### 分布式训练脚本运行说明
分布式任务的运行需要将表格中说明的多个参数进行赋值:

| 参数名 | 值类型 | 说明 | 示例 |
|:-------------|:---|:---------------------------------------|:-------------|
| trainer_id | int | 当前训练节点的ID,训练节点ID编号为0 - n-1, n为trainers的值 | 0/1/2/3 |
| pservers | str | parameter server 列表 | 127.0.0.1:6710,127.0.0.1:6711 |
| trainers | int | 训练节点的总个数,>0的数字 | 4 |
| server_endpoint | str | 当前所起的服务节点的IP:PORT | 127.0.0.1:8789 |
| training_role | str | 节点角色, TRAINER/PSERVER | PSERVER |

**其中:training_role 是用来区分当前所起服务的角色的,用于训练程序中,用户可根据需要自行定义,其他参数为fluid.DistributeTranspiler的transpile函数所需要,需要在调用函数前进行定义,至于如何从外部环境传入,用户可自定义。**
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

注意training_role是用来区分当前所起服务的角色的,在训练程序中,用户可根据需要自行定义,其他参数为Distribute Transpiler所需要的参数,样例如下:

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done


参数赋值及使用的相关代码片段:
```python
t = fluid.DistributeTranspiler()
t.transpile(
optimize_ops,
params_grads,
trainer_id,
pservers=pserver,
trainers=trainers)
if training_role == "PSERVER":
pserver_prog = t.get_pserver_program(server_endpoint)
pserver_startup = t.get_startup_program(server_endpoint, pserver_prog)
```

### 启动顺序
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这行可以去掉了,Fluid中对启动顺序无要求。

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

先启动全部的PSERVER (Parameter Server)后,再启动TRAINER(Trainer)。

### Demo
完整的demo代码位于Fluid的test目录下的[book](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/fluid/tests/book/test_fit_a_line.py)中。
```
cd /paddle/python/paddle/fluid/tests/book
```
第一步:启动Parameter Server, 启动Parameter Server的命令:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

第一步,参考如下命令启动Parameter Server:

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

```
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

bash的命令可以用```bash ....```来格式化。

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

PADDLE_INIT_PORT=6174 PADDLE_INIT_PSERVERS=192.168.1.2 TRAINERS=2 POD_IP=192.168.1.2 PADDLE_INIT_TRAINER_ID=1 TRAINING_ROLE=PSERVER python test_fit_a_line.py
```
执行命令后请等待出现提示: ```Server listening on 192.168.1.2:6174 ```
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

等待提示......,表示Paramter Server已经正常启动。

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

第二步:启动Trainer, 启动Trainer的命令:
```
PADDLE_INIT_PORT=6174 PADDLE_INIT_PSERVERS=192.168.1.3 TRAINERS=2 POD_IP=192.168.1.3 PADDLE_INIT_TRAINER_ID=1 TRAINING_ROLE=TRAINER python test_fit_a_line.py
```
由于我们定义的Trainer的数量是2个,因此需要在另外一个计算节点上再启动一个Trainer。
现在我们就启动了一个包含一个Parameter Server和两个Trainer的分布式训练任务。