Custom training loop with Keras and MultiWorkerMirroredStrategy

View on TensorFlow.org Run in Google Colab View source on GitHub Download notebook

Overview

This tutorial demonstrates how to perform multi-worker distributed training with a Keras model and with custom training loops using the tf.distribute.Strategy API. The training loop is distributed via tf.distribute.MultiWorkerMirroredStrategy, such that a tf.keras model—designed to run on single-worker—can seamlessly work on multiple workers with minimal code changes. Custom training loops provide flexibility and a greater control on training, while also making it easier to debug the model. Learn more about writing a basic training loop, writing a training loop from scratch and custom training.

If you are looking for how to use MultiWorkerMirroredStrategy with tf.keras.Model.fit, refer to this tutorial instead.

Distributed Training in TensorFlow guide is available for an overview of the distribution strategies TensorFlow supports for those interested in a deeper understanding of tf.distribute.Strategy APIs.

Setup

First, some necessary imports.

import json
import os
import sys

Before importing TensorFlow, make a few changes to the environment:

  • Disable all GPUs. This prevents errors caused by all workers trying to use the same GPU. In a real-world application, each worker would be on a different machine.
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
  • Reset the 'TF_CONFIG' environment variable (you'll see more about this later).
os.environ.pop('TF_CONFIG', None)
  • Make sure that the current directory is on Python's path. This allows the notebook to import the files written by %%writefile later.
if '.' not in sys.path:
  sys.path.insert(0, '.')

Now import TensorFlow.

import tensorflow as tf

Dataset and model definition

Next, create an mnist.py file with a simple model and dataset setup. This Python file will be used by the worker-processes in this tutorial:

%%writefile mnist.py

import os
import tensorflow as tf
import numpy as np

def mnist_dataset(batch_size):
  (x_train, y_train), _ = tf.keras.datasets.mnist.load_data()
  # The `x` arrays are in uint8 and have values in the range [0, 255].
  # You need to convert them to float32 with values in the range [0, 1]
  x_train = x_train / np.float32(255)
  y_train = y_train.astype(np.int64)
  train_dataset = tf.data.Dataset.from_tensor_slices(
      (x_train, y_train)).shuffle(60000)
  return train_dataset

def dataset_fn(global_batch_size, input_context):
  batch_size = input_context.get_per_replica_batch_size(global_batch_size)
  dataset = mnist_dataset(batch_size)
  dataset = dataset.shard(input_context.num_input_pipelines,
                          input_context.input_pipeline_id)
  dataset = dataset.batch(batch_size)
  return dataset

def build_cnn_model():
  return tf.keras.Sequential([
      tf.keras.Input(shape=(28, 28)),
      tf.keras.layers.Reshape(target_shape=(28, 28, 1)),
      tf.keras.layers.Conv2D(32, 3, activation='relu'),
      tf.keras.layers.Flatten(),
      tf.keras.layers.Dense(128, activation='relu'),
      tf.keras.layers.Dense(10)
  ])
Writing mnist.py

Multi-worker configuration

Now let's enter the world of multi-worker training. In TensorFlow, the 'TF_CONFIG' environment variable is required for training on multiple machines. Each machine may have a different role. The 'TF_CONFIG' variable used below is a JSON string that specifies the cluster configuration on each worker that is part of the cluster. This is the default method for specifying a cluster, using cluster_resolver.TFConfigClusterResolver, but there are other options available in the distribute.cluster_resolver module. Learn more about setting up the 'TF_CONFIG' variable in the Distributed training guide.

Describe your cluster

Here is an example configuration:

tf_config = {
    'cluster': {
        'worker': ['localhost:12345', 'localhost:23456']
    },
    'task': {'type': 'worker', 'index': 0}
}

Note that tf_config is just a local variable in Python. To use it for training configuration, serialize it as a JSON and place it in a 'TF_CONFIG' environment variable. Here is the same 'TF_CONFIG' serialized as a JSON string:

json.dumps(tf_config)
'{"cluster": {"worker": ["localhost:12345", "localhost:23456"]}, "task": {"type": "worker", "index": 0} }'

There are two components of 'TF_CONFIG': 'cluster' and 'task'.

  • 'cluster' is the same for all workers and provides information about the training cluster, which is a dict consisting of different types of jobs such as 'worker'. In multi-worker training with MultiWorkerMirroredStrategy, there is usually one 'worker' that takes on a little more responsibility like saving checkpoints and writing summary files for TensorBoard in addition to what a regular 'worker' does. Such a worker is referred to as the 'chief' worker, and it is customary that the 'worker' with 'index' 0 is appointed as the chief worker.

  • 'task' provides information of the current task and is different on each worker. It specifies the 'type' and 'index' of that worker.

In this example, you set the task 'type' to 'worker' and the task 'index' to 0. This machine is the first worker and will be appointed as the chief worker and do more work than the others. Note that other machines will need to have the 'TF_CONFIG' environment variable set as well, and it should have the same 'cluster' dict, but different task 'type' or task 'index' depending on what the roles of those machines are.

For illustration purposes, this tutorial shows how one may set a 'TF_CONFIG' with two workers on 'localhost'. In practice, users would create multiple workers on external IP addresses/ports, and set 'TF_CONFIG' on each worker appropriately.

This example uses two workers. The first worker's 'TF_CONFIG' is shown above. For the second worker, set tf_config['task']['index']=1.

Environment variables and subprocesses in notebooks

Subprocesses inherit environment variables from their parent. So if you set an environment variable in this Jupyter Notebook process:

os.environ['GREETINGS'] = 'Hello TensorFlow!'

you can then access the environment variable from a subprocess:

echo ${GREETINGS}
Hello TensorFlow!

In the next section, you'll use this to pass the 'TF_CONFIG' to the worker subprocesses. You would never really launch your jobs this way, but it's sufficient for the purposes of this tutorial: To demonstrate a minimal multi-worker example.

MultiWorkerMirroredStrategy

Before training the model, first create an instance of tf.distribute.MultiWorkerMirroredStrategy:

strategy = tf.distribute.MultiWorkerMirroredStrategy()
2022-05-03 01:27:56.164414: E tensorflow/stream_executor/cuda/cuda_driver.cc:271] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
INFO:tensorflow:Single-worker MultiWorkerMirroredStrategy with local_devices = ('/device:CPU:0',), communication = CommunicationImplementation.AUTO

Use tf.distribute.Strategy.scope to specify that a strategy should be used when building your model. This allows the strategy to control things like variable placement—it will create copies of all variables in the model's layers on each device across all workers.

import mnist
with strategy.scope():
  # Model building needs to be within `strategy.scope()`.
  multi_worker_model = mnist.build_cnn_model()

Auto-shard your data across workers

In multi-worker training, dataset sharding is needed to ensure convergence and reproducibility. Sharding means handing each worker a subset of the entire dataset—it helps create the experience similar to training on a single worker. In the example below, you're relying on the default autosharding policy of tf.distribute. You can also customize it by setting the tf.data.experimental.AutoShardPolicy of the tf.data.experimental.DistributeOptions. To learn more, refer to the Sharding section of the Distributed input tutorial.

per_worker_batch_size = 64
num_workers = len(tf_config['cluster']['worker'])
global_batch_size = per_worker_batch_size * num_workers

with strategy.scope():
  multi_worker_dataset = strategy.distribute_datasets_from_function(
      lambda input_context: mnist.dataset_fn(global_batch_size, input_context))
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz
11490434/11490434 [==============================] - 0s 0us/step

Define a custom training loop and train the model

Specify an optimizer:

with strategy.scope():
  # The creation of optimizer and train_accuracy needs to be in
  # `strategy.scope()` as well, since they create variables.
  optimizer = tf.keras.optimizers.RMSprop(learning_rate=0.001)
  train_accuracy = tf.keras.metrics.SparseCategoricalAccuracy(
      name='train_accuracy')

Define a training step with tf.function:

@tf.function
def train_step(iterator):
  """Training step function."""

  def step_fn(inputs):
    """Per-Replica step function."""
    x, y = inputs
    with tf.GradientTape() as tape:
      predictions = multi_worker_model(x, training=True)
      per_batch_loss = tf.keras.losses.SparseCategoricalCrossentropy(
          from_logits=True,
          reduction=tf.keras.losses.Reduction.NONE)(y, predictions)
      loss = tf.nn.compute_average_loss(
          per_batch_loss, global_batch_size=global_batch_size)

    grads = tape.gradient(loss, multi_worker_model.trainable_variables)
    optimizer.apply_gradients(
        zip(grads, multi_worker_model.trainable_variables))
    train_accuracy.update_state(y, predictions)
    return loss

  per_replica_losses = strategy.run(step_fn, args=(next(iterator),))
  return strategy.reduce(
      tf.distribute.ReduceOp.SUM, per_replica_losses, axis=None)

Checkpoint saving and restoring

As you write a custom training loop, you need to handle checkpoint saving manually instead of relying on a Keras callback. Note that for MultiWorkerMirroredStrategy, saving a checkpoint or a complete model requires the participation of all workers, because attempting to save only on the chief worker could lead to a deadlock. Workers also need to write to different paths to avoid overwriting each other. Here's an example of how to configure the directories:

from multiprocessing import util
checkpoint_dir = os.path.join(util.get_temp_dir(), 'ckpt')

def _is_chief(task_type, task_id, cluster_spec):
  return (task_type is None
          or task_type == 'chief'
          or (task_type == 'worker'
              and task_id == 0
              and "chief" not in cluster_spec.as_dict()))

def _get_temp_dir(dirpath, task_id):
  base_dirpath = 'workertemp_' + str(task_id)
  temp_dir = os.path.join(dirpath, base_dirpath)
  tf.io.gfile.makedirs(temp_dir)
  return temp_dir

def write_filepath(filepath, task_type, task_id, cluster_spec):
  dirpath = os.path.dirname(filepath)
  base = os.path.basename(filepath)
  if not _is_chief(task_type, task_id, cluster_spec):
    dirpath = _get_temp_dir(dirpath, task_id)
  return os.path.join(dirpath, base)

Create one tf.train.Checkpoint that tracks the model, which is managed by a tf.train.CheckpointManager, so that only the latest checkpoints are preserved:

epoch = tf.Variable(
    initial_value=tf.constant(0, dtype=tf.dtypes.int64), name='epoch')
step_in_epoch = tf.Variable(
    initial_value=tf.constant(0, dtype=tf.dtypes.int64),
    name='step_in_epoch')
task_type, task_id = (strategy.cluster_resolver.task_type,
                      strategy.cluster_resolver.task_id)
# Normally, you don't need to manually instantiate a `ClusterSpec`, but in this 
# illustrative example you did not set `'TF_CONFIG'` before initializing the
# strategy. Check out the next section for "real-world" usage.
cluster_spec = tf.train.ClusterSpec(tf_config['cluster'])

checkpoint = tf.train.Checkpoint(
    model=multi_worker_model, epoch=epoch, step_in_epoch=step_in_epoch)

write_checkpoint_dir = write_filepath(checkpoint_dir, task_type, task_id,
                                      cluster_spec)
checkpoint_manager = tf.train.CheckpointManager(
    checkpoint, directory=write_checkpoint_dir, max_to_keep=1)

Now, when you need to restore a checkpoint, you can find the latest checkpoint saved using the convenient tf.train.latest_checkpoint function (or by calling tf.train.CheckpointManager.restore_or_initialize).

latest_checkpoint = tf.train.latest_checkpoint(checkpoint_dir)
if latest_checkpoint:
  checkpoint.restore(latest_checkpoint)

After restoring the checkpoint, you can continue with training your custom training loop.

num_epochs = 3
num_steps_per_epoch = 70

while epoch.numpy() < num_epochs:
  iterator = iter(multi_worker_dataset)
  total_loss = 0.0
  num_batches = 0

  while step_in_epoch.numpy() < num_steps_per_epoch:
    total_loss += train_step(iterator)
    num_batches += 1
    step_in_epoch.assign_add(1)

  train_loss = total_loss / num_batches
  print('Epoch: %d, accuracy: %f, train_loss: %f.'
                %(epoch.numpy(), train_accuracy.result(), train_loss))

  train_accuracy.reset_states()

  # Once the `CheckpointManager` is set up, you're now ready to save, and remove
  # the checkpoints non-chief workers saved.
  checkpoint_manager.save()
  if not _is_chief(task_type, task_id, cluster_spec):
    tf.io.gfile.rmtree(write_checkpoint_dir)

  epoch.assign_add(1)
  step_in_epoch.assign(0)
2022-05-03 01:27:57.910846: W tensorflow/core/framework/dataset.cc:768] Input of GeneratorDatasetOp::Dataset will not be optimized because the dataset does not implement the AsGraphDefInternal() method needed to apply optimizations.
Epoch: 0, accuracy: 0.839063, train_loss: 0.513761.
Epoch: 1, accuracy: 0.940290, train_loss: 0.196421.
Epoch: 2, accuracy: 0.959487, train_loss: 0.137908.

Complete code at a glance

To sum up all the procedures discussed so far:

  1. You create worker processes.
  2. Pass 'TF_CONFIG's to the worker processes.
  3. Let each work process run the script below that contains the training code.

File: main.py

Writing main.py

The current directory now contains both Python files:

ls *.py
main.py
mnist.py

So JSON-serialize the 'TF_CONFIG' and add it to the environment variables:

os.environ['TF_CONFIG'] = json.dumps(tf_config)

Now, you can launch a worker process that will run the main.py and use the 'TF_CONFIG':

# first kill any previous runs
%killbgscripts
All background processes were killed.
python main.py &> job_0.log

There are a few things to note about the above command:

  1. It uses the %%bash which is a notebook "magic" to run some bash commands.
  2. It uses the --bg flag to run the bash process in the background, because this worker will not terminate. It waits for all the workers before it starts.

The backgrounded worker process won't print the output to this notebook. The &> redirects its output to a file, so that you can inspect what happened.

Wait a few seconds for the process to start up:

import time
time.sleep(20)

Now, check the output to the worker's log file so far:

cat job_0.log
2022-05-03 01:28:08.018600: E tensorflow/stream_executor/cuda/cuda_driver.cc:271] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected

The last line of the log file should say: Started server with target: grpc://localhost:12345. The first worker is now ready, and is waiting for all the other worker(s) to be ready to proceed.

Update the tf_config for the second worker's process to pick up:

tf_config['task']['index'] = 1
os.environ['TF_CONFIG'] = json.dumps(tf_config)

Now launch the second worker. This will start the training since all the workers are active (so there's no need to background this process):

python main.py > /dev/null 2>&1

If you recheck the logs written by the first worker, notice that it participated in training that model:

cat job_0.log
2022-05-03 01:28:08.018600: E tensorflow/stream_executor/cuda/cuda_driver.cc:271] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
2022-05-03 01:28:29.041643: W tensorflow/core/framework/dataset.cc:768] Input of GeneratorDatasetOp::Dataset will not be optimized because the dataset does not implement the AsGraphDefInternal() method needed to apply optimizations.
Epoch: 0, accuracy: 0.847321, train_loss: 0.495879.
Epoch: 1, accuracy: 0.941853, train_loss: 0.188717.
Epoch: 2, accuracy: 0.961496, train_loss: 0.136167.
# Delete the `'TF_CONFIG'`, and kill any background tasks so they don't affect the next section.
os.environ.pop('TF_CONFIG', None)
%killbgscripts
All background processes were killed.

Multi-worker training in depth

This tutorial has demonstrated a custom training loop workflow of the multi-worker setup. Detailed descriptions of other topics is available in the Multi-worker training with Keras (tf.keras.Model.fit) tutorial applicable to custom training loops.

Learn more

  1. The Distributed training in TensorFlow guide provides an overview of the available distribution strategies.
  2. Official models, many of which can be configured to run multiple distribution strategies.
  3. The Performance section in the tf.function guide provides information about other strategies and tools you can use to optimize the performance of your TensorFlow models.