Getting Started with Parallel Distributed Training

Overview

In deep learning, the increasing number of datasets and parameters prolongs the training time and requires more hardware resources, becoming a training bottleneck. Parallel distributed training is an important optimization method for training, which can reduce requirements on hardware, such as memory and computing performance. Based on different parallel principles and modes, parallelism is generally classified into the following types:

  • Data parallelism: splits data into many batches and then allocates the batches to each worker for model computation.

  • Model parallelism: splits a model. MindSpore supports the intra-layer model parallelism. Parameters are split and then allocated to each worker for training.

  • Hybrid parallelism: contains data parallelism and model parallelism.

MindSpore also provides the parallel distributed training function. It supports the following modes:

  • DATA_PARALLEL: data parallelism.

  • AUTO_PARALLEL: automatic parallelism, which integrates data parallelism, model parallelism, and hybrid parallelism. A cost model can be automatically created to select one parallel mode for users. Creating a cost model refers to modeling the training time based on the memory-based computation and communication overheads of the Ascend 910 chip, and designing efficient algorithms to develop a parallel strategy with a relatively short training time.

  • HYBRID_PARALLEL: On MindSpore, users manually split parameters to implement intra-layer model parallelism.

This tutorial describes how to train the ResNet-50 network in data parallel and automatic parallel modes on MindSpore.

The example in this tutorial applies to hardware platforms based on the Ascend 910 AI processor, whereas does not support CPU and GPU scenarios. Download address of the complete sample code: https://gitee.com/mindspore/docs/blob/r0.3/tutorials/tutorial_code/distributed_training/resnet50_distributed_training.py

Preparations

Downloading the Dataset

This sample uses the CIFAR-10 dataset, which consists of color images of 32 x 32 pixels in 10 classes, with 6000 images per class. There are 50,000 images in the training set and 10,000 images in the test set.

Download the dataset and decompress it to a local path. The folder generated after the decompression is cifar-10-batches-bin.

Configuring Distributed Environment Variables

When distributed training is performed in the bare-metal environment (compared with the cloud environment where the Ascend 910 AI processor is deployed on the local host), you need to configure the networking information file for the current multi-device environment. If the HUAWEI CLOUD environment is used, skip this section because the cloud service has been configured.

The following uses the Ascend 910 AI processor as an example. The JSON configuration file for an environment with eight devices is as follows. In this example, the configuration file is named rank_table_8pcs.json. For details about how to configure the 2-device environment, see the rank_table_2pcs.json file in the sample code.

{
    "board_id": "0x0000",
    "chip_info": "910",
    "deploy_mode": "lab",
    "group_count": "1",
    "group_list": [
        {
            "device_num": "8",
            "server_num": "1",
            "group_name": "",
            "instance_count": "8",
            "instance_list": [
                {"devices": [{"device_id": "0","device_ip": "192.1.27.6"}],"rank_id": "0","server_id": "10.155.111.140"},
                {"devices": [{"device_id": "1","device_ip": "192.2.27.6"}],"rank_id": "1","server_id": "10.155.111.140"},
                {"devices": [{"device_id": "2","device_ip": "192.3.27.6"}],"rank_id": "2","server_id": "10.155.111.140"},
                {"devices": [{"device_id": "3","device_ip": "192.4.27.6"}],"rank_id": "3","server_id": "10.155.111.140"},
                {"devices": [{"device_id": "4","device_ip": "192.1.27.7"}],"rank_id": "4","server_id": "10.155.111.140"},
                {"devices": [{"device_id": "5","device_ip": "192.2.27.7"}],"rank_id": "5","server_id": "10.155.111.140"},
                {"devices": [{"device_id": "6","device_ip": "192.3.27.7"}],"rank_id": "6","server_id": "10.155.111.140"},
                {"devices": [{"device_id": "7","device_ip": "192.4.27.7"}],"rank_id": "7","server_id": "10.155.111.140"},
                ]
        }
    ],
    "para_plane_nic_location": "device",
    "para_plane_nic_name": ["eth0","eth1","eth2","eth3","eth4","eth5","eth6","eth7"],
    "para_plane_nic_num": "8",
    "status": "completed"
}

The following parameters need to be modified based on the actual training environment:

  • board_id: current running environment. Set this parameter to 0x0000 for x86, and to 0x0020 for ARM.

  • server_num: number of hosts.

  • server_id: IP address of the local host.

  • device_num, para_plane_nic_num, and instance_count: number of devices.

  • rank_id: logical sequence number of a device, which starts from 0.

  • device_id: physical sequence number of a device, that is, the actual sequence number of the device on the corresponding host.

  • device_ip: IP address of the integrated NIC. You can run the cat /etc/hccn.conf command on the current host. The key value of address_x is the IP address of the NIC.

  • para_plane_nic_name: name of the corresponding NIC.

Calling the Collective Communication Library

The Huawei Collective Communication Library (HCCL) is used for the communication of MindSpore parallel distributed training and can be found in the Ascend 310 AI processor software package. In addition, mindspore.communication.management encapsulates the collective communication API provided by the HCCL to help users configure distributed information.

HCCL implements multi-device multi-node communication based on the Ascend AI processor. The common restrictions on using the distributed service are as follows. For details, see the HCCL documentation.

  • In a single-node system, a cluster of 1, 2, 4, or 8 devices is supported. In a multi-node system, a cluster of 8 x N devices is supported.

  • Each host has four devices numbered 0 to 3 and four devices numbered 4 to 7 deployed on two different networks. During training of 2 or 4 devices, the devices must be connected and clusters cannot be created across networks.

  • The server hardware architecture and operating system require the symmetrical multi-processing (SMP) mode.

The sample code for calling the HCCL as follows:

import os
from mindspore import context
from mindspore.communication.management import init

if __name__ == "__main__":
    context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", device_id=int(os.environ["DEVICE_ID"]))
    init()
    ...   

In the preceding code:

  • mode=context.GRAPH_MODE: sets the running mode to graph mode for distributed training. (The PyNative mode does not support parallel running.)

  • device_id: physical sequence number of a device, that is, the actual sequence number of the device on the corresponding host.

  • init(): enables HCCL communication and completes the distributed training initialization.

Loading the Dataset in Data Parallel Mode

During distributed training, data is imported in data parallel mode. The following takes the CIFAR-10 dataset as an example to describe how to import the CIFAR-10 dataset in data parallel mode. data_path indicates the dataset path, which is also the path of the cifar-10-batches-bin folder.

import mindspore.common.dtype as mstype
import mindspore.dataset as ds
import mindspore.dataset.transforms.c_transforms as C
import mindspore.dataset.transforms.vision.c_transforms as vision
from mindspore.communication.management import get_rank, get_group_size

def create_dataset(data_path, repeat_num=1, batch_size=32, rank_id=0, rank_size=1):
    resize_height = 224
    resize_width = 224
    rescale = 1.0 / 255.0
    shift = 0.0
    
    # get rank_id and rank_size
    rank_id = get_rank()
    rank_size = get_group_size()
    data_set = ds.Cifar10Dataset(data_path, num_shards=rank_size, shard_id=rank_id)
    
    # define map operations
    random_crop_op = vision.RandomCrop((32, 32), (4, 4, 4, 4))
    random_horizontal_op = vision.RandomHorizontalFlip()
    resize_op = vision.Resize((resize_height, resize_width))
    rescale_op = vision.Rescale(rescale, shift)
    normalize_op = vision.Normalize((0.4465, 0.4822, 0.4914), (0.2010, 0.1994, 0.2023))
    changeswap_op = vision.HWC2CHW()
    type_cast_op = C.TypeCast(mstype.int32)

    c_trans = [random_crop_op, random_horizontal_op]
    c_trans += [resize_op, rescale_op, normalize_op, changeswap_op]

    # apply map operations on images
    data_set = data_set.map(input_columns="label", operations=type_cast_op)
    data_set = data_set.map(input_columns="image", operations=c_trans)

    # apply shuffle operations
    data_set = data_set.shuffle(buffer_size=10)

    # apply batch operations
    data_set = data_set.batch(batch_size=batch_size, drop_remainder=True)

    # apply repeat operations
    data_set = data_set.repeat(repeat_num)

    return data_set

Different from the single-node system, the multi-node system needs to transfer the num_shards and shard_id parameters to the dataset API. The two parameters correspond to the number of devices and logical sequence numbers of devices, respectively. You are advised to obtain the parameters through the HCCL API.

  • get_rank: obtains the ID of the current device in the cluster.

  • get_group_size: obtains the number of devices.

Defining the Network

In data parallel and automatic parallel modes, the network definition method is the same as that in a single-node system. The reference code is as follows: https://gitee.com/mindspore/docs/blob/r0.3/tutorials/tutorial_code/resnet/resnet.py

Defining the Loss Function and Optimizer

Defining the Loss Function

Automatic parallelism splits models using the operator granularity and obtains the optimal parallel strategy through algorithm search. Therefore, to achieve a better parallel training effect, you are advised to use small operators to implement the loss function.

In the Loss function, the SoftmaxCrossEntropyWithLogits is expanded into multiple small operators for implementation according to a mathematical formula. The sample code is as follows:

from mindspore.ops import operations as P
from mindspore import Tensor
import mindspore.ops.functional as F
import mindspore.common.dtype as mstype
import mindspore.nn as nn

class SoftmaxCrossEntropyExpand(nn.Cell):
    def __init__(self, sparse=False):
        super(SoftmaxCrossEntropyExpand, self).__init__()
        self.exp = P.Exp()
        self.sum = P.ReduceSum(keep_dims=True)
        self.onehot = P.OneHot()
        self.on_value = Tensor(1.0, mstype.float32)
        self.off_value = Tensor(0.0, mstype.float32)
        self.div = P.Div()
        self.log = P.Log()
        self.sum_cross_entropy = P.ReduceSum(keep_dims=False)
        self.mul = P.Mul()
        self.mul2 = P.Mul()
        self.mean = P.ReduceMean(keep_dims=False)
        self.sparse = sparse
        self.max = P.ReduceMax(keep_dims=True)
        self.sub = P.Sub()
        
    def construct(self, logit, label):
        logit_max = self.max(logit, -1)
        exp = self.exp(self.sub(logit, logit_max))
        exp_sum = self.sum(exp, -1)
        softmax_result = self.div(exp, exp_sum)
        if self.sparse:
            label = self.onehot(label, F.shape(logit)[1], self.on_value, self.off_value)
        softmax_result_log = self.log(softmax_result)
        loss = self.sum_cross_entropy((self.mul(softmax_result_log, label)), -1)
        loss = self.mul2(F.scalar_to_array(-1.0), loss)
        loss = self.mean(loss, -1)

        return loss

Defining the Optimizer

The Momentum optimizer is used as the parameter update tool. The definition is the same as that in the single-node system. For details, see the implementation in the sample code.

Training the Network

context.set_auto_parallel_context() is an API for users to set parallel training parameters and must be called before the initialization of Model. If no parameters are specified, MindSpore will automatically set parameters to the empirical values based on the parallel mode. For example, in data parallel mode, parameter_broadcast is enabled by default. The related parameters are as follows:

  • parallel_mode: parallel distributed mode. The default value is ParallelMode.STAND_ALONE. The options are ParallelMode.DATA_PARALLEL and ParallelMode.AUTO_PARALLEL.

  • parameter_broadcast: whether to broadcast initialized parameters. The default value is True in DATA_PARALLEL and HYBRID_PARALLEL mode.

  • mirror_mean: During backward computation, the framework collects gradients of parameters in data parallel mode across multiple hosts, obtains the global gradient value, and transfers the global gradient value to the optimizer for update. The default value is False, which indicates that the allreduce_sum operation is applied. The value True indicates that the allreduce_mean operation is applied.

You are advised to set device_num and global_rank to their default values. The framework calls the HCCL API to obtain the values.

If multiple network cases exist in the script, call context.reset_auto_parallel_context() to restore all parameters to default values before executing the next case.

In the following sample code, the automatic parallel mode is specified. To switch to the data parallel mode, you only need to change parallel_mode to DATA_PARALLEL.

from mindspore import context
from mindspore.nn.optim.momentum import Momentum
from mindspore.train.callback import LossMonitor
from mindspore.train.model import Model, ParallelMode
from resnet import resnet50

device_id = int(os.getenv('DEVICE_ID'))
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
context.set_context(device_id=device_id) # set device_id

def test_train_cifar(num_classes=10, epoch_size=10):
    context.set_auto_parallel_context(parallel_mode=ParallelMode.AUTO_PARALLEL, mirror_mean=True)
    loss_cb = LossMonitor()
    dataset = create_dataset(epoch_size)
    net = resnet50(32, num_classes)
    loss = SoftmaxCrossEntropyExpand(sparse=True)
    opt = Momentum(filter(lambda x: x.requires_grad, net.get_parameters()), 0.01, 0.9)
    model = Model(net, loss_fn=loss, optimizer=opt)
    model.train(epoch_size, dataset, callbacks=[loss_cb], dataset_sink_mode=True)

In the preceding code:

  • dataset_sink_mode=True: uses the dataset sink mode. That is, the training computing is sunk to the hardware platform for execution.

  • LossMonitor: returns the loss value through the callback function to monitor the loss function.

Running the Script

After the script required for training is edited, run the corresponding command to call the script.

Currently, MindSpore distributed execution uses the single-device single-process running mode. That is, one process runs on each device, and the number of total processes is the same as the number of devices that are being used. For device 0, the corresponding process is executed in the foreground. For other devices, the corresponding processes are executed in the background. You need to create a directory for each process to store log information and operator compilation information. The following takes the distributed training script for eight devices as an example to describe how to run the script:

#!/bin/bash

DATA_PATH=$1
export DATA_PATH=${DATA_PATH}
RANK_SIZE=$2

EXEC_PATH=$(pwd)

test_dist_8pcs()
{
    export MINDSPORE_HCCL_CONFIG_PATH=${EXEC_PATH}/rank_table_8pcs.json
    export RANK_SIZE=8
}

test_dist_2pcs()
{
    export MINDSPORE_HCCL_CONFIG_PATH=${EXEC_PATH}/rank_table_2pcs.json
    export RANK_SIZE=2
}

test_dist_${RANK_SIZE}pcs

for((i=1;i<${RANK_SIZE};i++))
do
    rm -rf device$i
    mkdir device$i
    cp ./resnet50_distributed_training.py ./resnet.py ./device$i
    cd ./device$i
    export DEVICE_ID=$i
    export RANK_ID=$i
    echo "start training for device $i"
    env > env$i.log
    pytest -s -v ./resnet50_distributed_training.py > train.log$i 2>&1 &
    cd ../
done
rm -rf device0
mkdir device0
cp ./resnet50_distributed_training.py ./resnet.py ./device0
cd ./device0
export DEVICE_ID=0
export RANK_ID=0
echo "start training for device 0"
env > env0.log
pytest -s -v ./resnet50_distributed_training.py > train.log0 2>&1
if [ $? -eq 0 ];then
    echo "training success"
else
    echo "training failed"
    exit 2
fi
cd ../

The variables DATA_PATH and RANK_SIZE need to be transferred to the script, which indicate the path of the dataset and the number of devices, respectively.

The necessary environment variables are as follows:

  • MINDSPORE_HCCL_CONFIG_PATH: path for storing the networking information file.

  • DEVICE_ID: actual sequence number of the current device on the corresponding host.

  • RANK_ID: logical sequence number of the current device. For details about other environment variables, see configuration items in the installation guide.

The running time is about 5 minutes, which is mainly occupied by operator compilation. The actual training time is within 20 seconds. You can use ps -ef | grep pytest to monitor task processes.

Log files are saved in the device directory. The env.log file records environment variable information. The train.log file records the loss function information. The following is an example:

resnet50_distributed_training.py::test_train_feed ===============ds_num 195
global_step: 194, loss: 1.997
global_step: 389, loss: 1.655
global_step: 584, loss: 1.723
global_step: 779, loss: 1.807
global_step: 974, loss: 1.417
global_step: 1169, loss: 1.195
global_step: 1364, loss: 1.238
global_step: 1559, loss: 1.456
global_step: 1754, loss: 0.987
global_step: 1949, loss: 1.035
end training
PASSED