SplitDedupData

public final class SplitDedupData

An op splits input deduplication data XLA tuple into integer and floating point tensors.

Deduplication data is an XLA tuple, which consists of integer and floating point values. This op is to split these values into two groups for two types, and construct each group as one tensor to return.

Nested Classes

class SplitDedupData.Options Optional attributes for SplitDedupData  

Public Methods

static SplitDedupData.Options
config(String config)
static <T extends Number, U extends Number> SplitDedupData<T, U>
create(Scope scope, Operand<?> input, Class<T> integerType, Class<U> floatType, String tupleMask, Options... options)
Factory method to create a class wrapping a new SplitDedupData operation.
Output<U>
floatTensor()
A 1-D float tensor, includes float elements of deduplication data tuple.
Output<T>
integerTensor()
A 1-D integer tensor, includes integer elements of deduplication data tuple.

Inherited Methods

Public Methods

public static SplitDedupData.Options config (String config)

public static SplitDedupData<T, U> create (Scope scope, Operand<?> input, Class<T> integerType, Class<U> floatType, String tupleMask, Options... options)

Factory method to create a class wrapping a new SplitDedupData operation.

Parameters
scope current scope
input An XLA tuple including integer and float elements as deduplication data tuple.
integerType integer_tensor type. Allowed types: int32, int64, uint32, uint64.
floatType float_tensor type. Allowed types: half, bfloat16, float.
tupleMask A serialized TensorProto string of output tuple mask. This mask is a 2-D tensor, with first column as tuple element type, and second column as span of this type. For example, an output tuple of (1, 2, 0.1, 3), its mask is [[0, 2], [1, 1], [0, 1]]. We expect only two types of elements: integer(0) and float(1).
options carries optional attributes values
Returns
  • a new instance of SplitDedupData

public Output<U> floatTensor ()

A 1-D float tensor, includes float elements of deduplication data tuple.

public Output<T> integerTensor ()

A 1-D integer tensor, includes integer elements of deduplication data tuple.