使用MirroredStrategy时,tensorflow Estimator是否为工作人员分配了不同的批次?

时间:2019-01-23 12:45:58

标签: tensorflow tensorflow-datasets tensorflow-estimator

我正在将GANEstimator与MirroredStrategy一起用于单个实例的多个GPU。在我的情况下,input_fntf.data.Dataset,具有以下设置:

dataset = dataset.repeat()
dataset = dataset.shuffle(buffer_size=100)
dataset = dataset.batch(self.batch_size, drop_remainder=True)
dataset = dataset.prefetch(100)

之所以这样问,是因为我需要手动指定类似dataset.shard()之类的内容,才能将不同的数据传递给工作人员吗?我正在研究EstimatorMirroredStrategy的代码,但是我不清楚发生了什么。从description of distributed strategies会造成其他混乱:

MirroredStrategy: This does in-graph replication with synchronous 
training on many GPUs on one machine. Essentially, we create copies of all
variables in the model's layers on each device. We then use all-reduce 
to combine gradients across the devices before applying them 
to the variables to keep them in sync.

CollectiveAllReduceStrategy: This is a version of MirroredStrategy 
for multi-worker training. 

那么MirroredStratedy是否只使用一名工人?我不明白我需要指定等于一个塔的容量的批处理大小,否则得到OOM。有人可以将我指向代码,并说明这种简单的设置如何与批处理一起工作:

def create_dataset():
    ...
    dataset = dataset.repeat()
    dataset = dataset.shuffle(buffer_size=100)
    dataset = dataset.batch(self.batch_size, drop_remainder=True)
    dataset = dataset.prefetch(100)
    return dataset



NUM_GPUS = 4
strategy = tf.contrib.distribute.MirroredStrategy(num_gpus=NUM_GPUS)

optimizer = tf.train.RMSPropOptimizer(learning_rate=0.01, use_locking=True)
optimizer_d = tf.train.RMSPropOptimizer(learning_rate=0.01, use_locking=True)

config = tf.estimator.RunConfig(save_checkpoints_steps=100, 
          save_summary_steps=1, keep_checkpoint_max=50, 
          train_distribute=strategy)

# I have more hooks here, just simplified to show 
def get_hooks_fn(GANTrainOps):

    disjoint_train_hook_func = tfgan.get_sequential_train_hooks(
                 train_steps=tfgan.GANTrainSteps(10, 1)
                 ) # g steps, d steps
    disjoint_train_hooks = disjoint_train_hook_func(GANTrainOps)
    return [update_hook, summary_hook] + disjoint_train_hooks


# Create GAN estimator.
gan_estimator = tfgan.estimator.GANEstimator(
    model_dir = '/data/checkpoints/estimator_model', 
    generator_fn = generator_fn,
    discriminator_fn = discriminator_fn,
    generator_loss_fn = generator_loss_fn, 
    discriminator_loss_fn = discriminator_loss_fn, 
    generator_optimizer = optimizer,
    discriminator_optimizer = optimizer_d, 
    use_loss_summaries=True,
    config=config,
    get_hooks_fn=get_hooks_fn)


gan_estimator.train(input_fn=create_dataset, steps=10000)

谢谢!

MirroredStrategy的代码包含:

1)奇怪的措辞:

  

此类的多工作人员版本将一个副本映射到服务器上的一个设备。     工人。它在所有副本上镜像所有模型变量。例如,如果您     有两个worker,每个worker有4个GPU,它将创建8个副本     这8个GPU上的模型变量。然后像在MirroredStrategy(???)中一样,每个     副本使用自己的变量副本执行计算,除非在     发生变量或张量减少的交叉复制模型。

2)

  

auto_shard_dataset:是否在存在以下情况时自动分片数据集       多名工人。

此参数默认为False。

编辑:

到目前为止,我发现tf.estimator.train()在一段时间后指向似乎是strategy.make_input_fn_iterator()的地方:

def _get_iterator_from_input_fn(self, input_fn, mode, distribution=None):
    if distribution is not None:
      iterator = distribution.make_input_fn_iterator(
          lambda _: self._call_input_fn(input_fn, mode))
      input_hooks = [
          estimator_util.DistributedIteratorInitializerHook(iterator)]
    else:
      result = self._call_input_fn(input_fn, mode)
      iterator = result.make_initializable_iterator()
      input_hooks = [estimator_util._DatasetInitializerHook(iterator)]  
return iterator, input_hooks

make_input_fn_iterator()

但是它已从MirroredStrategy的代码中删除,并且不再存在!我不知道它是如何工作的以及数据集实际上在哪里拆分。

EDIT2:我在使用grep的tensorflow 1.12.0发行版中找不到行make_input_fn_iterator。似乎代码中完全不存在。

2 个答案:

答案 0 :(得分:1)

好吧,花了一些时间研究github之后,我发现它已经和我的tf 1.12.0不同了。因此,进入1.12.0的本地文件可以得到:

GANEstimator继承了tf.python.estimator.Estimator

Estimator.init():

# The distribute field contains an instance of DistributionStrategy.
    self._train_distribution = self._config.train_distribute

然后向下的路径是:

tf.contrib.gan.GANEstimator -> tf.python.estimator.Estimator.train() --> 
tf.python.estimator.Estimator._train_model(input_fn, hooks, saving_listeners) --> 
._train_model_distributed(input_fn, hooks, saving_listeners) --> 
._get_iterator_from_input_fn(input_fn, model_fn_lib.ModeKeys.TRAIN, self._train_distribution) --> 
distribution.distribute_dataset(lambda: self._call_input_fn(input_fn, mode))

在我的情况下要求MirrorredStrategy.distribute_dataset():

def distribute_dataset(self, dataset_fn):
    if self._cluster_spec:
      return values.MultiWorkerDataset(
          partial(self._call_dataset_fn, dataset_fn), self._worker_device_map,
          self._prefetch_on_device, self._auto_shard_dataset)
    else:
      return values.PerDeviceDataset(
          self._call_dataset_fn(dataset_fn), self._devices,
          self._prefetch_on_device)

tensorflow/python/training/distribute.py

  def _call_dataset_fn(self, dataset_fn):
    result = dataset_fn()
    if not isinstance(result, dataset_ops.Dataset):
      raise ValueError(
          "dataset_fn() must return a tf.data.Dataset when using a "
          "DistributionStrategy.")
    return result

我假设使用了PerDeviceDataset,所以最后我在values.py中找到了这两个类:

class PerDeviceDataset(object):
  """Like `tf.data.Dataset` split devices, producing `PerDevice` data."""

  def __init__(self, dataset, devices, prefetch_on_device=None):
    self._devices = devices

    # Default to using prefetching in graph mode, unless specified.
    # TODO(priyag): Enable prefetching in eager mode.
    self._prefetch_on_device = prefetch_on_device
    if self._prefetch_on_device is None:
      self._prefetch_on_device = not context.executing_eagerly()
    assert not (self._prefetch_on_device and context.executing_eagerly()), (
        "Prefetching is only supported in graph mode currently")

    if self._prefetch_on_device:
      self._dataset = dataset.apply(
          prefetching_ops_v2.prefetch_to_devices(self._devices))
    else:
      # TODO(priyag): If dropping remainder is not appropriate, find another
      # approach to distributing the dataset when not possible to divide evenly.
      # Possibly not an issue when we start using PartitionedDataset.
      self._dataset = dataset.batch(len(devices), drop_remainder=True)

  def make_one_shot_iterator(self):
    """Get a one time use iterator for the distributed PerDeviceDataset."""
    dataset_iterator = self._dataset.make_one_shot_iterator()
    return PerDeviceDataIterator(dataset_iterator, self._devices,
                                 self._prefetch_on_device)

  def make_initializable_iterator(self):
    """Get an initializable iterator for the distributed PerDeviceDataset."""
    dataset_iterator = self._dataset.make_initializable_iterator()
    return PerDeviceDataIterator(dataset_iterator, self._devices,
                                 self._prefetch_on_device)


class PerDeviceDataIterator(object):
  """An iterator (like `tf.data.Iterator`) into a `PerDeviceDataset`."""

  def __init__(self, iterator, devices, prefetch_on_device=None):
    self._iterator = iterator
    self._devices = devices
    self._prefetch_on_device = prefetch_on_device

  @property
  def initializer(self):
    return self._iterator.initializer

  def get_next(self, name=None):
    """Scatter the input across devices."""
    if self._prefetch_on_device:
      data_list = self._iterator.get_next(name=name)
      index = dict(zip(self._devices, data_list))
    else:
      batch = self._iterator.get_next(name=name)
      index = {}
      def get_ith(i):
        return lambda x: x[i]

      for i, d in enumerate(self._devices):
        index[d] = nest.map_structure(get_ith(i), batch)
        if context.executing_eagerly():
          with ops.device(d):
            index[d] = nest.map_structure(array_ops.identity, index[d])

    return regroup(index)

据我所知,首先,我的dataset_fn()函数被调用来获取数据集对象,然后在其之上应用一批具有GPU数量的大小。该批次的元素必须是我在dataset_fn()内的数据集初始化中定义的实际批次,已分配给不同的设备。

答案 1 :(得分:0)

如果有帮助,我会做一些澄清,但真的不确定这是否是您的意思。

MirroredStrategy是否仅使用一名工作人员?

是的。 MirroredStrategy旨在仅在一台Worker上工作(也就是一台节点,一台计算机等)

我需要指定等于一塔容量的批次大小

不。您需要将批次大小乘以塔数之和。

注意:仅供参考,Tower是模型的副本,等于GPU的数量,也称为副本

从此Keras tutorial开始,这是简单计算批次大小的方法:

BATCH_SIZE_PER_REPLICA = 64
BATCH_SIZE = BATCH_SIZE_PER_REPLICA * strategy.num_replicas_in_sync

train_dataset = mnist_train.map(scale).cache().shuffle(BUFFER_SIZE).batch(BATCH_SIZE)
eval_dataset = mnist_test.map(scale).batch(BATCH_SIZE)

在这种情况下,每个GPU的批处理大小为64。然后乘以GPU的数量。 为什么要乘以GPU数量? 计算梯度和损耗。它将除以批处理大小的总和(而不是GPU批处理大小)

  1. 措辞怪异:

这是将MirroredStrategy与Multi-WorkerStrategy进行比较。对于集群,您的塔将被复制到每个工作者(例如,在此示例中为2个节点)。每个工作人员将负责将模型分发到他们的GPU(例如,在这种情况下为4个GPU)。在该示例中,您将有8个模型副本。

[...]然后,就像在MirroredStrategy(???)中一样,每个副本都使用自己的变量副本[...]

无论您使用多工作人员还是单个工作人员,每个GPU(或副本)都将独立计算其模型并随后进行同步。 我猜他们会提到“变量的副本”,因为存在另一种带有参数服务器(ps)的分布式计算拓扑,其中ps将收集所有副本的权重,求和,然后将其重新分配给所有副本,以进行下一轮。