addmv

→Tensor

addmv(input: Tensor, mat: Tensor, vec: Tensor, beta: float = ..., alpha: float = ...)

source

Matrix-vector multiply with a scaled accumulator (BLAS-2 gemv).

Computes $\beta \cdot \text{input} + \alpha \cdot (\text{mat} \cdot \text{vec})$ , the canonical BLAS-2 gemv operation.

Parameters

inputTensor

Accumulator vector of shape (M,).

matTensor

Matrix of shape (M, N).

vecTensor

Vector of shape (N,).

betafloat

Scalar multiplier on input. Defaults to 1.0.

alphafloat

Scalar multiplier on mat @ vec. Defaults to 1.0.

Returns

Tensor

Vector of shape (M,).

Notes

Mathematical definition:

\text{out}_i = \beta \cdot \text{input}_i + \alpha \cdot \sum_{j = 0}^{N - 1} \text{mat}_{ij} \cdot \text{vec}_j.

Implemented by promoting vec to a column (N, 1), calling lucid.matmul, and squeezing the trailing axis.

Examples

>>> import lucid
>>> M = lucid.zeros(2)
>>> A = lucid.tensor([[1., 2.], [3., 4.]])
>>> v = lucid.tensor([5., 6.])
>>> lucid.addmv(M, A, v)
Tensor([17., 39.])