ReduceSum#

Domain: ai.onnx
Since version: 13

Computes the sum of the input tensor’s elements along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned. The axes are provided as an optional second input tensor. Negative axes are supported. If axes is not provided or is empty and noop_with_empty_axes is set to 1, the input tensor is returned unchanged. If axes is not provided or is empty and noop_with_empty_axes is not set, all dimensions are reduced.

Inputs

data (T): An input tensor.
axes (tensor(int64)): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if noop_with_empty_axes is false, else act as an Identity op when noop_with_empty_axes is true. Accepted range is [-r, r-1] where r = rank(data).

Outputs

reduced (T): Reduced output tensor.

Attributes

keepdims (int): Keep the reduced dimension or not, default 1 means keep reduced dimension.
noop_with_empty_axes (int): Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced, and the output tensor would be equivalent to input tensor.

Type Constraints

T: Constrain input and output types to high-precision numeric tensors. Allowed types: tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64).

Examples#

test_cc_reducesum_default_axes_keepdims

Inputs:
  data: shape=(3, 2, 2), dtype=float32
    [[[ 1.,  2.],
      [ 3.,  4.]],

     [[ 5.,  6.],
      [ 7.,  8.]],

     [[ 9., 10.],
      [11., 12.]]]

Outputs:
  reduced: shape=(1, 1, 1), dtype=float32
    [[[78.]]]

test_cc_reducesum_do_not_keepdims

Attributes:
  keepdims = 0

Inputs:
  data: shape=(3, 2, 2), dtype=float32
    [[[ 1.,  2.],
      [ 3.,  4.]],

     [[ 5.,  6.],
      [ 7.,  8.]],

     [[ 9., 10.],
      [11., 12.]]]
  axes: shape=(1,), dtype=int64
    [1]

Outputs:
  reduced: shape=(3, 2), dtype=float32
    [[ 4.,  6.],
     [12., 14.],
     [20., 22.]]

test_cc_reducesum_negative_axes_keepdims

Inputs:
  data: shape=(3, 2, 2), dtype=float32
    [[[ 1.,  2.],
      [ 3.,  4.]],

     [[ 5.,  6.],
      [ 7.,  8.]],

     [[ 9., 10.],
      [11., 12.]]]
  axes: shape=(1,), dtype=int64
    [-2]

Outputs:
  reduced: shape=(3, 1, 2), dtype=float32
    [[[ 4.,  6.]],

     [[12., 14.]],

     [[20., 22.]]]

Differences with previous version (11)#

SchemaDiff: ReduceSum (domain 'ai.onnx')

old version: 11
new version: 13
breaking: yes

Breaking reasons:

input ‘axes’ (added): at position 1; option=Single; type_str=’tensor(int64)’
attribute ‘axes’ (removed): type=INTS; required=False

Inputs:

[BREAKING] added ‘axes’: at position 1; option=Single; type_str=’tensor(int64)’

Attributes:

[BREAKING] removed ‘axes’: type=INTS; required=False
added ‘noop_with_empty_axes’: type=INT; required=False; default=0

Type constraints:

changed ‘T’: added types: [‘tensor(bfloat16)’]

Documentation:

line similarity: 0.50 (+3/-3 lines)

--- ReduceSum v11
+++ ReduceSum v13
@@ -1,6 +1,6 @@
 Computes the sum of the input tensor's elements along the provided axes.
 The resulting tensor has the same rank as the input if keepdims equals 1.
 If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
-The axes attribute specifies which dimensions to reduce. Negative axes are supported.
-If axes is not provided, all dimensions are reduced.
-If noop_with_empty_axes is set and axes is empty, the input tensor is returned unchanged.
+The axes are provided as an optional second input tensor. Negative axes are supported.
+If axes is not provided or is empty and noop_with_empty_axes is set to 1, the input tensor is returned unchanged.
+If axes is not provided or is empty and noop_with_empty_axes is not set, all dimensions are reduced.

ReduceSum#

Examples#

Differences with previous version (11)#

Version History#