sum(inp, axis=None, keepdims=False)[源代码]

Calculates the sum of tensor elements over a given axis (or axes).

  • inp (Tensor) – input tensor. Should have a numeric data type.

  • axis (Union[int, Sequence[int], None]) – axis or axes along which sums must be computed. By default, the sum must be computed over the entire tensor. If a sequence of integers, sums must be computed over multiple axes.

  • keepdims (bool) – if True, the reduced axes (dimensions) must be included in the result as singleton dimensions, and, accordingly, the result must be compatible with the input tensor (see 广播机制与规则). Otherwise, if False, the reduced axes (dimensions) must not be included in the result.




if the sum was computed over the entire tensor, a zero-dimensional tensor containing the sum; otherwise, a tensor containing the sums. The returned tensor must have a data type determined by 类型提升规则.

Special Cases

Let N equal the number of elements over which to compute the sum.

  • If N is 0, the sum is 0 (i.e., the empty sum).

  • If \(x_i\) is NaN, the sum is NaN (i.e., NaN values propagate).


If the accumulator is too small, overflow occurs:

>>> x = F.ones(128, dtype="int8")
>>> F.sum(x)
Tensor(-128, dtype=int8, device=xpux:0)


The sum of an empty tensor is the neutral element 0:

>>> F.sum(Tensor([]))
Tensor(0.0, device=xpux:0)

Normal case:

>>> F.sum(Tensor([1, 2, 3]))
Tensor(6, dtype=int32, device=xpux:0)
>>> F.sum(Tensor([0.5, 1.5]))
Tensor(2.0, device=xpux:0)

Along an axis:

>>> F.sum(Tensor([[1, 2, 3], [4, 5, 6]]), axis=0)
Tensor([5 7 9], dtype=int32, device=xpux:0)
>>> F.sum(Tensor([[1, 2, 3], [4, 5, 6]]), axis=1)
Tensor([ 6 15], dtype=int32, device=xpux:0)