Python tf.distribute.experimental.coordinator.ClusterCoordinator.fetch用法及代碼示例

用法

fetch(
    val
)

阻止調用以從遠程值中獲取結果。

這是RemoteValue 結構的tf.distribute.experimental.coordinator.RemoteValue.fetch 的包裝器；它返回RemoteValue s的執行結果。如果尚未準備好，請在阻止調用者的同時等待他們。

例子：

strategy = ...
coordinator = tf.distribute.experimental.coordinator.ClusterCoordinator(
    strategy)

def dataset_fn():
  return tf.data.Dataset.from_tensor_slices([1, 1, 1])

with strategy.scope():
  v = tf.Variable(initial_value=0)

@tf.function
def worker_fn(iterator):
  def replica_fn(x):
    v.assign_add(x)
    return v.read_value()
  return strategy.run(replica_fn, args=(next(iterator),))

distributed_dataset = coordinator.create_per_worker_dataset(dataset_fn)
distributed_iterator = iter(distributed_dataset)
result = coordinator.schedule(worker_fn, args=(distributed_iterator,))
assert coordinator.fetch(result) == 1