[NOT A CRUCIAL QUESTION] Are built-in torch functions really "faster"?

[sachinlodhi]

So I was watching the matrix multiplication section and was trying to compare the manual (for-loop) method and the PyTorch built-in method i.e. torch.matmul().
Interestingly, I found that, even for 1D vector, for loop was faster than torch.matmul().
The screenshot is attached.

[LuluW8071]

Interesting experiment. Yes they could be faster for 1D ones. But for 2D and higher dimensions, torch is faster. This is because matmul and other operations are written in C++ which is faster than regular python

Matmul refrence docs