mindspore.Tensor.addbmm

Tensor.addbmm()

Applies batch matrix multiplication to batch1 and batch2, with a reduced add step and add self to the result.

The optional values alpha and beta are the matrix-matrix product between batch1 and batch2 and the scale factor for the added tensor self respectively. If beta is 0, then self will be ignored.

\[output = \beta self + \alpha (\sum_{i=0}^{b-1} {batch1_i @ batch2_i})\]

Warning

This is an experimental API that is subject to change or deletion.

Parameters

  • batch1 (Tensor) – The first batch of tensor to be multiplied.

  • batch2 (Tensor) – The second batch of tensor to be multiplied.

Keyword Arguments

  • beta (Union[int, float], optional) – Multiplier for input. Default: 1 .

  • alpha (Union[int, float], optional) – Multiplier for batch1 @ batch2. Default: 1 .

Returns

Tensor, has the same dtype as self.

Raises

  • TypeError – If alpha or beta is not an int or float.

  • ValueError – If batch1, batch2 cannot apply batch matrix multiplication.

  • ValueError – If batch1 and batch2 are not 3-D tensors.

Supported Platforms:

Ascend

Examples

>>> import numpy as np
>>> from mindspore import Tensor, ops
>>> m = np.ones((3, 3)).astype(np.float32)
>>> arr1 = np.arange(24).astype(np.float32).reshape((2, 3, 4))
>>> arr2 = np.arange(24).astype(np.float32).reshape((2, 4, 3))
>>> a = Tensor(arr1)
>>> b = Tensor(arr2)
>>> c = Tensor(m)
>>> output = c.addbmm(a, b)
>>> print(output)
[[ 949. 1009. 1069.]
 [1285. 1377. 1469.]
 [1621. 1745. 1869.]]