Add distributed training examples of PyTorch

docs/manual/cluster-user/README.md

@@ -11,6 +11,7 @@ This manual is for cluster users to learn how to submit job, debug job, manage d
 3. [How to Manage Data](./how-to-manage-data.md)
 4. [How to Debug Jobs](./how-to-debug-jobs.md)
 5. [How to Use Advanced Job Settings](./how-to-use-advanced-job-settings.md)
-6. [Use Marketplace](./use-marketplace.md)
-7. [Use VSCode Extension](./use-vscode-extension.md)
-8. [Use Jupyter Notebook Extension](./use-jupyter-notebook-extension.md)
+6. [How to Run Distributed Job](./how-to-run-distributed-job.md)
+7. [Use Marketplace](./use-marketplace.md)
+8. [Use VSCode Extension](./use-vscode-extension.md)
+9. [Use Jupyter Notebook Extension](./use-jupyter-notebook-extension.md)

docs/manual/cluster-user/how-to-run-distributed-job.md

@@ -0,0 +1,34 @@
+## How OpenPAI Deploy Distributed Jobs
+### Taskrole and Instance
+When we execute distributed programs on PAI, we can add different task roles for our job. For single server jobs, there is only one task role. For distributed jobs, there may be multiple task roles. For example, when TensorFlow is used to running distributed jobs, it has two roles, including the parameter server and the worker. In distributed jobs, each role may have one or more instances. For example, if it's 8 instances in a worker role of TensorFlow. It means there should be 8 Docker containers for the worker role. Please visit [here](how-to-use-advanced-job-settings.html#multiple-task-roles) for specific operations.
+
+### Environmental variables
+In a distributed job, one task might communicate with others (When we say task, we mean a single instance of a task role). So a task need to be aware of other tasks' runtime information such as IP, port, etc. The system exposes such runtime information as environment variables to each task's Docker container. For mutual communication, users can write code in the container to access those runtime environment variables. Please visit [here](how-to-use-advanced-job-settings.html#environmental-variables-and-port-reservation) for specific operations.
+
+### Retry policy and Completion policy
+If unknown error happens, PAI will retry the job according to user settings. To set a retry policy and completion policy for user's job，PAI asks user to switch to Advanced mode.  Please visit [here](how-to-use-advanced-job-settings.html#job-exit-spec-retry-policy-and-completion-policy) for specific operations.
+### Run PyTorch Distributed Jobs in OpenPAI
+Example Name | Multi-GPU | Multi-Node | Backend |Apex| Job protocol |
+---|---|---|---|---|---| 
+Single-Node DataParallel CIFAR-10 | ✓| x | -|-| [cifar10-single-node-gpus-cpu-DP.yaml](../../../examples/Distributed-example/cifar10-single-node-gpus-cpu-DP.yaml)|
+cifar10-single-mul-DDP-gloo.yaml | ✓|  ✓ | gloo|-| [cifar10-single-mul-DDP-gloo.yaml](../../../examples/Distributed-example/cifar10-single-mul-DDP-gloo.yaml)|
+cifar10-single-mul-DDP-nccl | ✓| ✓ |nccl|-| [cifar10-single-mul-DDP-nccl.yaml](../../../examples/Distributed-example/cifar10-single-mul-DDP-nccl.yaml)|
+cifar10-single-mul-DDP-gloo-Apex-mixed | ✓|  ✓ | gloo|✓ | [cifar10-single-mul-DDP-gloo-Apex-mixed.yaml](../../../examples/Distributed-example/cifar10-single-mul-DDP-gloo-Apex-mixed.yaml)|
+cifar10-single-mul-DDP-nccl-Apex-mixed | ✓|  ✓ | nccl|  ✓ | [cifar10-single-mul-DDP-gloo-Apex-mixed.yaml](../../../examples/Distributed-example/cifar10-single-mul-DDP-gloo-Apex-mixed.yaml)|
+imagenet-single-mul-DDP-gloo | ✓|  ✓| gloo|-| [imagenet-single-mul-DDP-gloo.yaml](../../../examples/Distributed-example/Lite-imagenet-single-mul-DDP-gloo.yaml)|
+## DataParallel
+The single node program is simple. The program executed in PAI is exactly the same as the program in our machine. It should be noted that an Worker can be applied in PAI and a Instance can be applied in Worker. In a worker, we can apply for GPUs that we need. We provide an [example](../../../examples/Distributed-example/cifar10-single-node-gpus-cpu-DP.py) of DP.
+
+## DistributedDataParallel
+DDP requires users set a master node ip and port for synchronization in PyTorch. For the port, you can simply set one certain port, such as `5000` as your master port. However, this port may conflict with others. To prevent port conflict, you can reserve a port in OpenPAI, as we mentioned [here](how-to-use-advanced-job-settings.html#environmental-variables-and-port-reservation). The port you reserved is available in environmental variables like `PAI_PORT_LIST_$taskRole_$taskIndex_$portLabel`, where `$taskIndex` means the instance index of that task role. For example, if your task role name is `work` and port label is `SyncPort`, you can add the following code in your PyTorch DDP program:
+
+```
+os.environ['MASTER_ADDR'] = os.environ['PAI_HOST_IP_worker_0']
+os.environ['MASTER_PORT'] = os.environ['PAI_worker_0_SynPort_PORT']
+```
+
+If you are using `gloo` as your DDP communication backend, please set correct network interface such as `export GLOO_SOCKET_IFNAME=eth0`.
+
+
+We provide examples with [gloo](../../../examples/Distributed-example/cifar10-single-mul-DDP-gloo.yaml) and [nccl](../../../examples/Distributed-example/cifar10-single-mul-DDP-nccl.yaml) as backend.
+

examples/Distributed-example/Lite-imagenet-single-mul-DDP-gloo.yaml

@@ -0,0 +1,50 @@
+protocolVersion: 2
+name: imagenet-gloo_8ba8ed42_7606233c
+type: job
+jobRetryCount: 0
+prerequisites:
+  - type: dockerimage
+    uri: 'openpai/standard:python_3.6-pytorch_1.2.0-gpu'
+    name: docker_image_0
+taskRoles:
+  worker:
+    instances: 2
+    completion:
+      minFailedInstances: 1
+    taskRetryCount: 0
+    dockerImage: docker_image_0
+    resourcePerInstance:
+      gpu: 4
+      cpu: 16
+      memoryMB: 32768
+      ports:
+        SynPort: 1
+    commands:
+      - export GLOO_SOCKET_IFNAME=eth0
+      - 'git clone https://github.com/NVIDIA/apex'
+      - cd apex
+      - python setup.py install
+      - cd ..
+      - apt update
+      - apt install -y nfs-common
+      - mkdir -p /mnt/data
+      - 'mount 10.151.40.32:/mnt/zhiyuhe /mnt/data'
+      - >-
+        wget
+        https://raw.githubusercontent.com/vvfreesoul/pai/master/examples/Distributed-example/Lite-imagenet-single-mul-DDP-nccl-gloo.py
+      - >-
+        python Lite-imagenet-single-mul-DDP-nccl-gloo.py  -n 2 -g 4
+        --dist-backend gloo --epochs 2 /mnt/data/imagenet/unzipped
+defaults:
+  virtualCluster: default
+extras:
+  com.microsoft.pai.runtimeplugin:
+    - plugin: ssh
+      parameters:
+        jobssh: true
+        userssh: {}
+  hivedScheduler:
+    taskRoles:
+      worker:
+        skuNum: 1
+        skuType: null

examples/Distributed-example/Lite-imagenet-single-mul-DDP-nccl-gloo.py

@@ -0,0 +1,117 @@
+import os
+from datetime import datetime
+import argparse
+import torch.multiprocessing as mp
+import torch.backends.cudnn as cudnn
+import torchvision
+import torchvision.transforms as transforms
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.distributed as dist
+from apex.parallel import DistributedDataParallel as DDP
+from apex import amp
+
+import torchvision.datasets as datasets
+import torchvision.models as models
+model_names = sorted(name for name in models.__dict__
+    if name.islower() and not name.startswith("__")
+    and callable(models.__dict__[name]))
+def main():
+    print('run main')
+    parser = argparse.ArgumentParser()
+    parser.add_argument('data', metavar='DIR',
+                        help='path to dataset')
+    parser.add_argument('-a', '--arch', metavar='ARCH', default='resnet18',
+                        choices=model_names,
+                        help='model architecture: ' +
+                             ' | '.join(model_names) +
+                             ' (default: resnet18)')
+    parser.add_argument('-n', '--nodes', default=1, type=int, metavar='N',
+                        help='number of data loading workers (default: 4)')
+    parser.add_argument('-g', '--gpus', default=1, type=int,
+                        help='number of gpus per node')
+    parser.add_argument('-nr', '--nr', default=0, type=int,
+                        help='ranking within the nodes')
+    parser.add_argument('-b', '--batch-size', default=256, type=int,
+                        metavar='N',
+                        help='mini-batch size (default: 256), this is the total '
+                             'batch size of all GPUs on the current node when '
+                             'using Data Parallel or Distributed Data Parallel')
+    parser.add_argument('--epochs', default=2, type=int, metavar='N',
+                        help='number of total epochs to run')
+    parser.add_argument('--dist-backend',  default='nccl', type=str,
+                        help='distributed backend')
+    args = parser.parse_args()
+    args.world_size = args.gpus * args.nodes
+    print('world_size:',args.world_size)
+    os.environ['MASTER_ADDR'] = os.environ['PAI_HOST_IP_worker_0']
+    os.environ['MASTER_PORT'] = os.environ['PAI_worker_0_SynPort_PORT']
+    print('master:', os.environ['MASTER_ADDR'], 'port:', os.environ['MASTER_PORT'])
+    mp.spawn(train, nprocs=args.gpus, args=(args,))
+
+def train(gpu, args):
+    print("start train")
+    rank = int(os.environ['PAI_TASK_INDEX']) * args.gpus + gpu
+    dist.init_process_group(backend=args.dist_backend, init_method='env://', world_size=args.world_size, rank=rank)
+    torch.manual_seed(0)
+    model=model = models.__dict__[args.arch]()
+    torch.cuda.set_device(gpu)
+    model.cuda(gpu)
+    batch_size = args.batch_size
+    # define loss function (criterion) and optimizer
+    criterion = nn.CrossEntropyLoss().cuda(gpu)
+    optimizer = torch.optim.SGD(model.parameters(), 1e-4)
+    # Wrap the model
+    model = nn.parallel.DistributedDataParallel(model, device_ids=[gpu])
+    # Data loading code
+    traindir = os.path.join(args.data, 'train')
+    valdir = os.path.join(args.data, 'val')
+    normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
+                                     std=[0.229, 0.224, 0.225])
+    train_dataset = datasets.ImageFolder(
+        traindir,
+        transforms.Compose([
+            transforms.RandomResizedCrop(224),
+            transforms.RandomHorizontalFlip(),
+            transforms.ToTensor(),
+            normalize,
+        ]))
+    train_sampler = torch.utils.data.distributed.DistributedSampler(train_dataset)
+
+    train_loader = torch.utils.data.DataLoader(
+        train_dataset, batch_size=args.batch_size, shuffle=(train_sampler is None),
+        num_workers=args.nodes, pin_memory=True, sampler=train_sampler)
+
+    val_loader = torch.utils.data.DataLoader(
+        datasets.ImageFolder(valdir, transforms.Compose([
+            transforms.Resize(256),
+            transforms.CenterCrop(224),
+            transforms.ToTensor(),
+            normalize,
+        ])),
+        batch_size=args.batch_size, shuffle=False,
+        num_workers=args.nodes, pin_memory=True)
+    start = datetime.now()
+    total_step = len(train_loader)
+    for epoch in range(args.epochs):
+        for i, (images, labels) in enumerate(train_loader):
+            images = images.cuda(non_blocking=True)
+            labels = labels.cuda(non_blocking=True)
+            # Forward pass
+            outputs = model(images)
+            loss = criterion(outputs, labels)
+
+            # Backward and optimize
+            optimizer.zero_grad()
+            loss.backward()
+            optimizer.step()
+            #if (i + 1) % 100 == 0 and gpu == 0:
+            print('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}'.format(epoch + 1, args.epochs, i + 1, total_step,
+                                                                         loss.item()))
+    if gpu == 0:
+        print("Training complete in: " + str(datetime.now() - start))
+
+
+if __name__ == '__main__':
+    main()

examples/Distributed-example/LiteApex-imagenet-single-mul-DDP-nccl-gloo.py

@@ -0,0 +1,122 @@
+import os
+from datetime import datetime
+import argparse
+import torch.multiprocessing as mp
+import torch.backends.cudnn as cudnn
+import torchvision
+import torchvision.transforms as transforms
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.distributed as dist
+from apex.parallel import DistributedDataParallel as DDP
+from apex import amp
+
+import torchvision.datasets as datasets
+import torchvision.models as models
+model_names = sorted(name for name in models.__dict__
+    if name.islower() and not name.startswith("__")
+    and callable(models.__dict__[name]))
+def main():
+    print('run main')
+    parser = argparse.ArgumentParser()
+    parser.add_argument('data', metavar='DIR',
+                        help='path to dataset')
+    parser.add_argument('-a', '--arch', metavar='ARCH', default='resnet18',
+                        choices=model_names,
+                        help='model architecture: ' +
+                             ' | '.join(model_names) +
+                             ' (default: resnet18)')
+    parser.add_argument('-n', '--nodes', default=1, type=int, metavar='N',
+                        help='number of data loading workers (default: 4)')
+    parser.add_argument('-g', '--gpus', default=1, type=int,
+                        help='number of gpus per node')
+    parser.add_argument('-nr', '--nr', default=0, type=int,
+                        help='ranking within the nodes')
+    parser.add_argument('-b', '--batch-size', default=256, type=int,
+                        metavar='N',
+                        help='mini-batch size (default: 256), this is the total '
+                             'batch size of all GPUs on the current node when '
+                             'using Data Parallel or Distributed Data Parallel')
+    parser.add_argument('--epochs', default=2, type=int, metavar='N',
+                        help='number of total epochs to run')
+    parser.add_argument('--dist-backend',  default='nccl', type=str,
+                        help='distributed backend')
+    args = parser.parse_args()
+    args.world_size = args.gpus * args.nodes
+    print('world_size:',args.world_size)
+    os.environ['MASTER_ADDR'] = os.environ['PAI_HOST_IP_worker_0']
+    os.environ['MASTER_PORT'] = os.environ['PAI_worker_0_SynPort_PORT']
+    print('master:', os.environ['MASTER_ADDR'], 'port:', os.environ['MASTER_PORT'])
+    mp.spawn(train, nprocs=args.gpus, args=(args,))
+
+def train(gpu, args):
+    print("start train")
+    rank = int(os.environ['PAI_TASK_INDEX']) * args.gpus + gpu
+    dist.init_process_group(backend=args.dist_backend, init_method='env://', world_size=args.world_size, rank=rank)
+    torch.manual_seed(0)
+    model=model = models.__dict__[args.arch]()
+    torch.cuda.set_device(gpu)
+    model.cuda(gpu)
+    batch_size = args.batch_size
+    # define loss function (criterion) and optimizer
+    criterion = nn.CrossEntropyLoss().cuda(gpu)
+    optimizer = torch.optim.SGD(model.parameters(), 1e-4)
+    # Wrap the model
+    model = nn.parallel.DistributedDataParallel(model, device_ids=[gpu])
+    # Wrap the model
+    model, optimizer = amp.initialize(model, optimizer, opt_level='O2')
+    model = DDP(model)
+    # Data loading code
+    traindir = os.path.join(args.data, 'train')
+    valdir = os.path.join(args.data, 'val')
+    normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
+                                     std=[0.229, 0.224, 0.225])
+
+    train_dataset = datasets.ImageFolder(
+        traindir,
+        transforms.Compose([
+            transforms.RandomResizedCrop(224),
+            transforms.RandomHorizontalFlip(),
+            transforms.ToTensor(),
+            normalize,
+        ]))
+    train_sampler = torch.utils.data.distributed.DistributedSampler(train_dataset)
+
+    train_loader = torch.utils.data.DataLoader(
+        train_dataset, batch_size=args.batch_size, shuffle=(train_sampler is None),
+        num_workers=args.nodes, pin_memory=True, sampler=train_sampler)
+
+    val_loader = torch.utils.data.DataLoader(
+        datasets.ImageFolder(valdir, transforms.Compose([
+            transforms.Resize(256),
+            transforms.CenterCrop(224),
+            transforms.ToTensor(),
+            normalize,
+        ])),
+        batch_size=args.batch_size, shuffle=False,
+        num_workers=args.nodes, pin_memory=True)
+    start = datetime.now()
+    total_step = len(train_loader)
+    for epoch in range(args.epochs):
+        for i, (images, labels) in enumerate(train_loader):
+            images = images.cuda(non_blocking=True)
+            labels = labels.cuda(non_blocking=True)
+            # Forward pass
+            outputs = model(images)
+            loss = criterion(outputs, labels)
+
+            # Backward and optimize
+            optimizer.zero_grad()
+            with amp.scale_loss(loss, optimizer) as scaled_loss:
+                scaled_loss.backward()
+            loss.backward()
+            optimizer.step()
+            #if (i + 1) % 100 == 0 and gpu == 0:
+            print('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}'.format(epoch + 1, args.epochs, i + 1, total_step,
+                                                                         loss.item()))
+    if gpu == 0:
+        print("Training complete in: " + str(datetime.now() - start))
+
+if __name__ == '__main__':
+    main()