FilesLast commitLast update
12 days ago
2 months ago
6 months ago
19 days ago
2 months ago
19 days ago
6 months ago
1 month ago
12 days ago
README

SplitD

English | 简体中文

Product Support Status

Product Supported
Ascend 950PR&950DT products √
Atlas A3 products √
Atlas A2 products √
Atlas 200I/500 A2 inference products √
Atlas inference products √
Atlas training products √

Function Description

  • Operator function: evenly splits a tensor into num_split smaller tensors along the specified dimension split_dim. Unlike the Split operator, split_dim is provided as an attribute instead of an input.

Parameter Description

Parameter Name Input/Output/Attribute Description Data Type Data Format
x Input Tensor to be split. FLOAT16, FLOAT, DOUBLE, INT8, INT16, INT32, INT64, UINT8, UINT16, UINT32, UINT64, BOOL ND
split_dim Attribute Dimension along which the tensor is split. INT -
num_split Attribute Number of tensors to be split. INT -
y Output Output result, which is a dynamic output. The number of outputs is determined by num_split. FLOAT16, FLOAT, DOUBLE, INT8, INT16, INT32, INT64, UINT8, UINT16, UINT32, UINT64, BOOL ND

Constraint Description

  • split_dim must be in the range [0, number of dimensions of value). If it is negative, counting starts from the end
  • The size of value on the split_dim dimension must be integer-divisible by num_split
  • num_split must be greater than or equal to 1

Invocation Description

Invocation Method Sample Code Description
Graph-mode API test_geir_split_d Invokes the SplitD operator through the operator IR API.