◆ Machine Learning // 3 posts
2026
Training GPT from Scratch
Training a GPT from scratch on my own RTX 4090, written down as I go. Part 1: preparing the dataset.
The Normal Equation: Linear Regression in One Line
Why you would fit a line at all, how the normal equation does it in O(nd² + d³), and a live chart to feel it: turn the noise, slope and intercept knobs and watch the fit, the residuals and the MSE respond.
Matrix Multiplication, From Definition to Cache Lines
One multiply, three implementations, three orders of magnitude — measured on a Skylake desktop, then explained: cache lines, arithmetic intensity, and the tiling idea every BLAS and every GPU matmul kernel is built on.