An Op to exchange data across TPU replicas.
On each replica, the input is split into `split_count` blocks along `split_dimension` and send to the other replicas given group_assignment. After receiving `split_count` - 1 blocks from other replicas, we concatenate the blocks along `concat_dimension` as the output.
For example, suppose there are 2 TPU replicas: replica 0 receives input: `[[A, B]]` replica 1 receives input: `[[C, D]]`
group_assignment=`[[0, 1]]` concat_dimension=0 split_dimension=1 split_count=2
replica 0's output: `[[A], [C]]` replica 1's output: `[[B], [D]]`
Constants
String | OP_NAME | The name of this op, as known by TensorFlow core engine |
Public Methods
Output<T> |
asOutput()
Returns the symbolic handle of the tensor.
|
static <T extends TType> AllToAll<T> | |
Output<T> |
output()
The exchanged result.
|
Inherited Methods
Constants
public static final String OP_NAME
The name of this op, as known by TensorFlow core engine
Public Methods
public Output<T> asOutput ()
Returns the symbolic handle of the tensor.
Inputs to TensorFlow operations are outputs of another TensorFlow operation. This method is used to obtain a symbolic handle that represents the computation of the input.
public static AllToAll<T> create (Scope scope, Operand<T> input, Operand<TInt32> groupAssignment, Long concatDimension, Long splitDimension, Long splitCount)
Factory method to create a class wrapping a new AllToAll operation.
Parameters
scope | current scope |
---|---|
input | The local input to the sum. |
groupAssignment | An int32 tensor with shape [num_groups, num_replicas_per_group]. `group_assignment[i]` represents the replica ids in the ith subgroup. |
concatDimension | The dimension number to concatenate. |
splitDimension | The dimension number to split. |
splitCount | The number of splits, this number must equal to the sub-group size(group_assignment.get_shape()[1]) |
Returns
- a new instance of AllToAll