|View source on GitHub|
A state & compute distribution policy on a list of devices.
tf.compat.v2.distribute.Strategy( extended )
See the guide for overview and examples.
- To use it with Keras
fit, please read.
- You may pass descendant of
tf.estimator.RunConfigto specify how a
tf.estimator.Estimatorshould distribute its computation. See guide.
- Otherwise, use
tf.distribute.Strategy.scopeto specify that a strategy should be used when building an executing your model. (This puts you in the "cross-replica context" for this strategy, which means the strategy is put in control of things like variable placement.)
If you are writing a custom training loop, you will need to call a few more methods, see the guide:
Start by either creating a
tf.data.Datasetnormally or using
tf.distribute.experimental_make_numpy_datasetto make a dataset out of a
tf.distribute.Strategy.experimental_distribute_datasetto convert a
tf.data.Datasetto something that produces "per-replica" values. If you want to manually specify how the dataset should be partitioned across replicas, use
tf.distribute.Strategy.experimental_run_v2to run a function once per replica, taking values that may be "per-replica" (e.g. from a distributed dataset) and returning "per-replica" values. This function is executed in "replica context", which means each operation is performed separately on each replica.
Finally use a method (such as
tf.distribute.Strategy.reduce) to convert the resulting "per-replica" values into ordinary
A custom training loop can be as simple as:
with my_strategy.scope(): @tf.function def distribute_train_epoch(dataset): def replica_fn(input): # process input and return result return result total_result = 0 for x in dataset: per_replica_result = my_strategy.experimental_run_v2(replica_fn, args=(x,)) total_result += my_strategy.reduce(tf.distribute.ReduceOp.SUM, per_replica_result, axis=None) return total_result dist_dataset = my_strategy.experimental_distribute_dataset(dataset) for _ in range(EPOCHS): train_result = distribute_train_epoch(dist_dataset)
This takes an ordinary
replica_fn and runs it
distributed using a particular
my_strategy above. Any variables created in
replica_fn are created
my_strategy's policy, and library functions called by
replica_fn can use the
get_replica_context() API to implement
You can use the
reduce API to aggregate results across replicas and use
this as a return value from one iteration over the distributed dataset. Or
you can use
tf.keras.metrics (such as loss, accuracy, etc.) to
accumulate metrics across steps in a given epoch.
See the custom training loop tutorial for a more detailed example.
||Returns number of replicas over which gradients are aggregated.|
experimental_distribute_dataset( dataset )
Distributes a tf.data.Dataset instance provided via
The returned distributed dataset can be iterated over similar to how regular datasets can. NOTE: Currently, the user cannot add any more transformations to a distributed dataset.
The following is an example:
strategy = tf.distribute.MirroredStrategy() # Create a dataset dataset = dataset_ops.Dataset.TFRecordDataset([ "/a/1.tfr", "/a/2.tfr", "/a/3.tfr", "/a/4.tfr"]) # Distribute that dataset dist_dataset = strategy.experimental_distribute_dataset(dataset) # Iterate over the distributed dataset for x in dist_dataset: # process dataset elements strategy.experimental_run_v2(train_step, args=(x,))
We will assume that the input dataset is batched by the global batch size. With this assumption, we will make a best effort to divide each batch across all the replicas (one or more workers).
In a multi-worker setting, we will first attempt to distribute the dataset by attempting to detect whether the dataset is being created out of ReaderDatasets (e.g. TFRecordDataset, TextLineDataset, etc.) and if so, attempting to shard the input files. Note that there has to be at least one input file per worker. If you have less than one input file per worker, we suggest that you should disable distributing your dataset using the method below.
If that attempt is unsuccessful (e.g. the dataset is created from a
Dataset.range), we will shard the dataset evenly at the end by appending a
.shard operation to the end of the processing pipeline. This will cause
the entire preprocessing pipeline for all the data to be run on every
worker, and each worker will do redundant work. We will print a warning
if this method of sharding is selected. In this case, consider using
You can disable dataset sharding across workers using the
Within each worker, we will also split the data among all the worker devices (if more than one a present), and this will happen even if multi-worker sharding is disabled using the method above.
If the above batch splitting and dataset sharding logic is undesirable,
experimental_distribute_datasets_from_function instead, which
does not do any automatic splitting or sharding.
experimental_distribute_datasets_from_function( dataset_fn )
tf.data.Dataset instances created by calls to
dataset_fn will be called once for each worker in the strategy. Each
replica on that worker will dequeue one batch of inputs from the local
Dataset (i.e. if a worker has two replicas, two batches will be dequeued
Dataset every step).
This method can be used for several purposes. For example, where
experimental_distribute_dataset is unable to shard the input files, this
method might be used to manually shard the dataset (avoiding the slow
fallback behavior in
experimental_distribute_dataset). In cases where the
dataset is infinite, this sharding can be done by creating dataset replicas
that differ only in their random seed.
experimental_distribute_dataset may also sometimes fail to split the
batch across replicas on a worker. In that case, this method can be used
where that limitation does not exist.
dataset_fn should take an
tf.distribute.InputContext instance where
information about batching and input replication can be accessed:
def dataset_fn(input_context): batch_size = input_context.get_per_replica_batch_size(global_batch_size) d = tf.data.Dataset.from_tensors([[1.]]).repeat().batch(batch_size) return d.shard( input_context.num_input_pipelines, input_context.input_pipeline_id) inputs = strategy.experimental_distribute_datasets_from_function(dataset_fn) for batch in inputs: replica_results = strategy.experimental_run_v2(replica_fn, args=(batch,))
A function taking a
experimental_local_results( value )
Returns the list of all local per-replica values contained in
A value returned by
A tuple of values contained in
experimental_make_numpy_dataset( numpy_input )
tf.data.Dataset for input provided via a numpy array.
This avoids adding
numpy_input as a large constant in the graph,
and copies the data to the machine or machines that will be processing
Note that you will likely need to use
with the returned dataset to further distribute it with the strategy.
numpy_input = np.ones(, dtype=np.float32) dataset = strategy.experimental_make_numpy_dataset(numpy_input) dist_dataset = strategy.experimental_distribute_dataset(dataset)
A nest of NumPy input arrays that will be converted into a
dataset. Note that lists of Numpy arrays are stacked, as that is normal
experimental_run_v2( fn, args=(), kwargs=None )
fn on each replica, with the given arguments.
Executes ops specified by
fn on each replica. If
"per-replica" values, such as those produced by a "distributed
fn is executed on a particular replica, it will be executed with the
component of those "per-replica" values that correspond to that replica.
fn may call
tf.distribute.get_replica_context() to access members such
All arguments in
kwargs should either be nest of tensors or
per-replica objects containing tensors or composite tensors.
The function to run. The output must be a
(Optional) Positional arguments to |