torch.mm

torch.mm(input, mat2, *, out=None) → Tensor

Performs a matrix multiplication of the matrices input and mat2.

If input is a $(n \times m)$ tensor, mat2 is a $(m \times p)$ tensor, out will be a $(n \times p)$ tensor.

Note

This function does not broadcast. For broadcasting matrix products, see torch.matmul().

Supports strided and sparse 2-D tensors as inputs, autograd with respect to strided inputs.

This operation has support for arguments with sparse layouts. If out is provided its layout will be used. Otherwise, the result layout will be deduced from that of input.

Warning

Sparse support is a beta feature and some layout(s)/dtype/device combinations may not be supported, or may not have autograd support. If you notice missing functionality please open a feature request.

This operator supports TensorFloat32.

On certain ROCm devices, when using float16 inputs this module will use different precision for backward.

Parameters:

• input (Tensor) – the first matrix to be matrix multiplied

• mat2 (Tensor) – the second matrix to be matrix multiplied

Keyword Arguments:

out (Tensor, optional) – the output tensor.

Example:

>>> mat1 = torch.randn(2, 3)
>>> mat2 = torch.randn(3, 3)
>>> torch.mm(mat1, mat2)
tensor([[ 0.4851,  0.5037, -0.3633],
        [-0.0760, -3.6705,  2.4784]])

torch.mm(input, mat2, out_dtype, *, out=None) → Tensor

Parameters:

• input (Tensor) – the first matrix to be matrix multiplied

• mat2 (Tensor) – the second matrix to be matrix multiplied

• out_dtype (dtype) – the dtype of the output tensor. On CUDA and XPU, only torch.float32 is supported given torch.float16/torch.bfloat16 input dtypes. Other backends (including out-of-tree accelerators) may support additional input/output dtype combinations.

Keyword Arguments:

out (Tensor, optional) – the output tensor.