PyTorch Quickstart

In this guide we will describe how to scale out PyTorch (v1.5 or above) programs using Orca in 4 simple steps.

Run in Google Colab  View source on GitHub

Step 0: Prepare Environment

Conda is needed to prepare the Python environment for running this example. Please refer to the install guide for more details.

conda create -n zoo python=3.7 # zoo is conda environment name, you can use any name you like.
conda activate zoo
pip install analytics-zoo # install either version 0.9 or latest nightly build
pip install torch==1.7.1 torchvision==0.8.2
pip install six cloudpickle
pip install jep==3.9.0

Step 1: Init Orca Context

from zoo.orca import init_orca_context, stop_orca_context

if args.cluster_mode == "local":
    init_orca_context(cores=1, memory="2g")   # run in local mode
elif args.cluster_mode == "k8s":
    init_orca_context(cluster_mode="k8s", num_nodes=2, cores=4) # run on K8s cluster
elif args.cluster_mode == "yarn":
    cluster_mode="yarn-client", cores=4, num_nodes=2, memory="2g",
    driver_memory="10g", driver_cores=1,
    conf={"spark.rpc.message.maxSize": "1024",
        "spark.task.maxFailures": "1",
        "spark.driver.extraJavaOptions": "-Dbigdl.failure.retryTimes=1"})   # run on Hadoop YARN cluster

This is the only place where you need to specify local or distributed mode. View Orca Context for more details.

Note: You should export HADOOP_CONF_DIR=/path/to/hadoop/conf/dir when you run on Hadoop YARN cluster.

Step 2: Define the Model

You may define your model, loss and optimizer in the same way as in any standard (single node) PyTorch program.

import torch
import torch.nn as nn
import torch.nn.functional as F

class LeNet(nn.Module):
    def __init__(self):
        super(LeNet, self).__init__()
        self.conv1 = nn.Conv2d(1, 20, 5, 1)
        self.conv2 = nn.Conv2d(20, 50, 5, 1)
        self.fc1 = nn.Linear(4*4*50, 500)
        self.fc2 = nn.Linear(500, 10)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        x = F.max_pool2d(x, 2, 2)
        x = F.relu(self.conv2(x))
        x = F.max_pool2d(x, 2, 2)
        x = x.view(-1, 4*4*50)
        x = F.relu(self.fc1(x))
        x = self.fc2(x)
        return F.log_softmax(x, dim=1)

model = LeNet()
criterion = nn.NLLLoss()
adam = torch.optim.Adam(model.parameters(), 0.001)

Step 3: Define Train Dataset

You can define the dataset using standard Pytorch DataLoader. Orca also supports a data creator function or Orca SparkXShards.

import torch
from torchvision import datasets, transforms

train_loader =
    datasets.MNIST(dir, train=True, download=True,
                       transforms.Normalize((0.1307,), (0.3081,))
    batch_size=batch_size, shuffle=True)
test_loader =
    datasets.MNIST(dir, train=False,
                       transforms.Normalize((0.1307,), (0.3081,))
    batch_size=test_batch_size, shuffle=False) 

Step 4: Fit with Orca Estimator

First, Create an Estimator

from zoo.orca.learn.pytorch import Estimator
from zoo.orca.learn.metrics import Accuracy 

est = Estimator.from_torch(model=model, optimizer=adam, loss=criterion, metrics=[Accuracy()])

Next, fit and evaluate using the Estimator

from zoo.orca.learn.trigger import EveryEpoch, epochs=10, validation_data=test_loader,

result = est.evaluate(data=test_loader)
for r in result:

Note: You should call stop_orca_context() when your application finishes.