torch mm

Guri Bee

2 min read

·

23-10-2024

1

0

Share

Demystifying PyTorch's torch.mm: Matrix Multiplication Made Easy

In the world of deep learning and scientific computing, matrix multiplication is a fundamental operation. PyTorch, a powerful deep learning framework, provides a convenient and efficient way to perform this operation through its torch.mm function.

This article aims to guide you through the intricacies of torch.mm, equipping you with the knowledge to leverage its power for your own projects.

Understanding torch.mm

At its core, torch.mm is a function designed to perform matrix multiplication between two tensors. It stands for "matrix multiplication" and accepts two inputs:

• Matrix 1 (tensor1): A 2-dimensional tensor representing the first matrix in the multiplication.
• Matrix 2 (tensor2): Another 2-dimensional tensor representing the second matrix.

The output of torch.mm is a new 2-dimensional tensor representing the product of the two input matrices.

Usage Examples

Let's explore some practical examples to solidify our understanding:

Example 1: Basic Matrix Multiplication

import torch

matrix1 = torch.tensor([[1, 2], [3, 4]])
matrix2 = torch.tensor([[5, 6], [7, 8]])

result = torch.mm(matrix1, matrix2)

print(result)

Output:

tensor([[19, 22],
        [43, 50]])

In this example, we create two matrices, matrix1 and matrix2, and perform matrix multiplication using torch.mm. The resulting matrix result holds the product of the two input matrices.

Example 2: Transpose and Multiplication

import torch

matrix1 = torch.tensor([[1, 2, 3], [4, 5, 6]])
matrix2 = torch.tensor([[7, 8], [9, 10], [11, 12]])

result = torch.mm(matrix1.T, matrix2)

print(result)

Output:

tensor([[58, 64],
        [139, 154]])

This example demonstrates the use of transposition with torch.mm. We first transpose matrix1 using .T and then perform matrix multiplication with matrix2. This results in a matrix with dimensions determined by the transposed matrix1 and matrix2.

Example 3: Broadcasting

import torch

matrix1 = torch.tensor([[1, 2], [3, 4]])
vector = torch.tensor([5, 6])

result = torch.mm(matrix1, vector.view(1, -1).T)

print(result)

Output:

tensor([[17],
        [39]])

Here, we perform matrix multiplication between a matrix (matrix1) and a vector (vector). We need to reshape the vector to a 1x2 matrix using .view(1, -1).T to ensure the dimensions are compatible for matrix multiplication.

Beyond torch.mm: torch.matmul and @ Operator

While torch.mm is well-suited for matrix multiplication, PyTorch offers more versatile alternatives:

• torch.matmul: This function handles multiplication between tensors of various dimensions, including matrices, vectors, and higher-order tensors. It is more flexible than torch.mm and allows for broadcasting.
• @ operator: This operator provides a more concise and Pythonic way to perform matrix multiplication. It works similarly to torch.matmul.

For most use cases, torch.matmul and the @ operator are recommended due to their greater flexibility and compatibility. However, torch.mm still holds its place for performing efficient matrix multiplication specifically between two 2-dimensional tensors.

Key Takeaways

• torch.mm is a PyTorch function specifically designed for matrix multiplication between two 2-dimensional tensors.
• It provides a convenient and efficient way to perform this fundamental operation.
• Remember the dimension requirements and explore the more versatile torch.matmul and @ operator for broader tensor multiplication scenarios.

By understanding the nuances of torch.mm and its alternatives, you can confidently implement efficient matrix multiplication operations within your PyTorch projects.

Acknowledgement: This article was inspired by and borrows from the official PyTorch documentation and discussions on GitHub.