tf.distribute.StrategyExtended

TensorFlow 1 version View source on GitHub

Additional APIs for algorithms that need to be distribution-aware.

Lower-level concepts:

  • Wrapped values: In order to represent values parallel across devices (either replicas or the devices associated with a particular value), we wrap them in a "PerReplica" or "Mirrored" object that contains a map from replica id to values. "PerReplica" is used when the value may be different across replicas, and "Mirrored" when the value are the same.
  • Unwrapping and merging: Consider calling a function fn on multiple replicas, like experimental_run_v2(fn, args=[w]) with an argument w that is a wrapped value. This means w will have a map taking replica id 0 to w0, replica id 11 to w1, etc. experimental_run_v2() unwraps w before calling fn, so it calls fn(w0) on d0, fn(w1) on d1, etc. It then merges the return values from fn(), which can possibly result in wrapped values. For example, let's say fn() returns a tuple with three components: (x, a, v0) from replica 0, (x, b, v1) on replica 1, etc. If the first component is the same object x from every replica, then the first component of the merged result will also be x. If the second component is different (a, b, ...) from each replica, then the merged value will have a wrapped map from replica device to the different values. If the third component is the members of a mirrored variable (v maps d0 to v0, d1 to v1, etc.), then the merged result will be that mirrored variable (v).
  • Worker devices vs. parameter devices: Most replica computations will happen on worker devices. Since we don't yet support model parallelism, there will be one worker device per replica. When using parameter servers or central storage, the set of devices holding variables may be different, otherwise the parameter devices might match the worker devices.

Replica context vs. Cross-replica context

replica context is when we are in some function that is being called once for each replica. Otherwise we are in cross-replica context, which is useful for calling tf.distribute.Strategy methods which operate across the replicas (like reduce_to()). By default you start in a replica context (the "default single replica context") and then some methods can switch you back and forth. There is a third mode you can be in called update context used when updating variables.

In a replica context, you may freely read the values of variables, but you may only update their value if they specify a way to aggregate the update using the aggregation parameter in the variable's constructor. In a cross-replica context, you may read or write variables (writes may need to be broadcast to all copies of the variable if it is mirrored).

Sync on read variables

In some cases, such as a metric, we want to accumulate a bunch of updates on each replica independently and only aggregate when reading. This can be a big performance win when the value is read only rarely (maybe the value is only read at the end of an epoch or when checkpointing). These are variables created by passing synchronization=ON_READ to the variable's constructor (and some value for aggregation).

The strategy may choose to put the variable on multiple devices, like mirrored variables, but unlike mirrored variables we don't synchronize the updates to them to make sure they have the same value. Instead, the synchronization is performed when reading in cross-replica context. In a replica context, reads and writes are performed on the local copy (we allow reads so you can write code like v = 0.9*v + 0.1*update). We don't allow operations like v.assign_add in a cross-replica context for sync on read variables; right now we don't have a use case for such updates and depending on the aggregation mode such updates may not be sensible.

Locality

Depending on how a value is produced, it will have a type that will determine how it may be used.

"Per-replica" values exist on the worker devices, with a different value for each replica. They are produced by iterating through a "distributed Dataset" returned by tf.distribute.Strategy.experimental_distribute_dataset and tf.distribute.Strategy.experimental_distribute_datasets_from_function. They are also the typical result returned by tf.distribute.Strategy.experimental_run_v2. You typically can't use a per-replica value directly in a cross-replica context, without first resolving how to aggregate the values across replicas, for instance by using tf.distribute.Strategy.reduce.

"Mirrored" values are like per-replica values, except we know that the value on all replicas are the same. We can safely read a mirrored value in a cross-replica context by using the value on any replica. You can convert a per-replica value into a mirrored value by using tf.distribute.ReplicaContext.all_reduce.

Values can also have the same locality as a variable, which is a mirrored value but residing on the same devices as the variable (as opposed to the compute devices). Such values may be passed to a call to tf.distribute.StrategyExtended.update to update the value of a variable. You may use tf.distribute.StrategyExtended.colocate_vars_with to give a variable the same locality as another variable. This is useful, for example, for "slot" variables used by an optimizer for keeping track of statistics used to update a primary/model variable. You may convert a per-replica value to a variable's locality by using tf.distribute.StrategyExtended.reduce_to or tf.distribute.StrategyExtended.batch_reduce_to.

In addition to slot variables which should be colocated with their primary variables, optimizers also define non-slot variables. These can be things like "number of step updates performed" or "beta1^t" and "beta2^t". Each strategy has some policy for which devices those variables should be copied too, called the "non-slot devices" (some subset of the parameter devices). We require that all non-slot variables are allocated on the same device, or mirrored across the same set of devices. You can use tf.distribute.StrategyExtended.non_slot_devices to pick a consistent set of devices to pass to both tf.distribute.StrategyExtended.colocate_vars_with and tf.distribute.StrategyExtended.update_non_slot.

How to update a variable

The standard pattern for updating variables is to:

  1. In your function passed to tf.distribute.Strategy.experimental_run_v2, compute a list of (update, variable) pairs. For example, the update might be a the gradient of the loss with respect to the variable.
  2. Switch to cross-replica mode by calling tf.distribute.get_replica_context().merge_call() with the updates and variables as arguments.
  3. Call tf.distribute.StrategyExtended.reduce_to(VariableAggregation.SUM, t, v) (for one variable) or tf.distribute.StrategyExtended.batch_reduce_to (for a list of variables) to sum the updates. and broadcast the result to the variable's devices.
  4. Call tf.distribute.StrategyExtended.update(v) for each variable to update its value.

Steps 2 through 4 are done automatically by class tf.keras.optimizers.Optimizer if you call its tf.keras.optimizers.Optimizer.apply_gradients method in a replica context. They are also done automatically if you call an assign* method on a (non sync-on-read) variable that was constructed with an aggregation method (which is used to determine the reduction used in step 3).

Distribute-aware layers

Layers are generally called in a replica context, except when defining a functional model. tf.distribute.in_cross_replica_context will let you determine which case you are in. If in a replica context, the tf.distribute.get_replica_context function will return a tf.distribute.ReplicaContext object. The ReplicaContext object has an all_reduce method for aggregating across all replicas. Alternatively, you can update variables following steps 2-4 above.

experimental_require_static_shapes Returns True if static shape is required; False otherwise.
parameter_devices Returns the tuple of all devices used to place variables.
worker_devices Returns the tuple of all devices used to for compute replica execution.

Methods

batch_reduce_to

View source

Combine multiple reduce_to calls into one for faster execution.

Args
reduce_op Reduction type, an instance of tf.distribute.ReduceOp enum.
value_destination_pairs A sequence of (value, destinations) pairs. See reduce_to() for a description.

Returns
A list of mirrored values, one per pair in value_destination_pairs.

colocate_vars_with

View source

Scope that controls which devices variables will be created on.

No operations should be added to the graph inside this scope, it should only be used when creating variables (some implementations work by changing variable creation, others work by using a tf.compat.v1.colocate_with() scope).

This may only be used inside self.scope().

Example usage:

with strategy.scope():
  var1 = tf.Variable(...)
  with strategy.extended.colocate_vars_with(var1):
    # var2 and var3 will be created on the same device(s) as var1
    var2 = tf.Variable(...)
    var3 = tf.Variable(...)

  def fn(v1, v2, v3):
    # operates on v1 from var1, v2 from var2, and v3 from var3

  # `fn` runs on every device `var1` is on, `var2` and `var3` will be there
  # too.
  strategy.extended.update(var1, fn, args=(var2, var3))

Args
colocate_with_variable A variable created in this strategy's scope(). Variables created while in the returned context manager will be on the same set of devices as colocate_with_variable.

Returns
A context manager.

non_slot_devices

View source

Device(s) for non-slot variables.

Create variables on these devices in a with colocate_vars_with(non_slot_devices(...)): block. Update those using update_non_slot().

Args
var_list The list of variables being optimized, needed with the default tf.distribute.Strategy.

Returns
A sequence of devices for non-slot variables.

reduce_to

View source

Combine (via e.g. sum or mean) values across replicas.

Args
reduce_op Reduction type, an instance of tf.distribute.ReduceOp enum.
value A per-replica value with one value per replica.
destinations A mirrored variable, a per-replica tensor, or a device string. The return value will be copied to all destination devices (or all the devices where the destinations value resides). To perform an all-reduction, pass value to destinations.

Returns
A tensor or value mirrored to destinations.

update

View source

Run fn to update var using inputs mirrored to the same devices.

If var is mirrored across multiple devices, then this implements logic like:

results = {}
for device, v in var:
  with tf.device(device):
    # args and kwargs will be unwrapped if they are mirrored.
    results[device] = fn(v, *args, **kwargs)
return merged(results)

Otherwise this returns fn(var, *args, **kwargs) colocated with var.

Neither args nor kwargs may contain per-replica values. If they contain mirrored values, they will be unwrapped before calling fn.

Args
var Variable, possibly mirrored to multiple devices, to operate on.
fn Function to call. Should take the variable as the first argument.
args Tuple or list. Additional positional arguments to pass to fn().
kwargs Dict with keyword arguments to pass to fn().
group Boolean. Defaults to True. If False, the return value will be unwrapped.

Returns
By default, the merged return value of fn across all replicas. The merged result has dependencies to make sure that if it is evaluated at all, the side effects (updates) will happen on every replica. If instead "group=False" is specified, this function will return a nest of lists where each list has an element per replica, and the caller is responsible for ensuring all elements are executed.

update_non_slot

View source

Runs fn(*args, **kwargs) on colocate_with devices.

Args
colocate_with The return value of non_slot_devices().
fn Function to execute.
args Tuple or list. Positional arguments to pass to fn().
kwargs Dict with keyword arguments to pass to fn().
group Boolean. Defaults to True. If False, the return value will be unwrapped.

Returns
Return value of fn, possibly merged across devices.

value_container

View source

Returns the container that this per-replica value belongs to.

Args
value A value returned by experimental_run_v2() or a variable created in scope().

Returns
A container that value belongs to. If value does not belong to any container (including the case of container having been destroyed), returns the value itself. value in experimental_local_results(value_container(value)) will always be true.

variable_created_in_scope

View source

Tests whether v was created while this strategy scope was active.

Variables created inside the strategy scope are "owned" by it:

with strategy.scope():
  v = tf.Variable(1.)
strategy.variable_created_in_scope(v)
True

Variables created outside the strategy are not owned by it:

v = tf.Variable(1.)
strategy.variable_created_in_scope(v)
False

Args
v A tf.Variable instance.

Returns
True if v was created inside the scope, False if not.