I have done a matrix multiplication program that runs ~50x faster than plain naive implementation and ~3x faster than IKJ-order with modular tricks in just 55 lines of code.

Typically, to get this kind of speed, you would have to spend 300+ lines of code for a Strassen implementation.

Here is the link: https://judge.yosupo.jp/submission/310249. A lot of the code was based on https://codeforces.me/blog/entry/101655, aside from the tmp part.