A (poorly) made LLM. It is made with just Python and no external libraries/dependencies (except for numPy). You can check out a demo here.
I need numPy for efficient matrix multiplication, as trying to do that in just python would be too slow (each training pass would take hours as opposed to like seconds). numPy does this in C, which is faster, but that's a bit out of the scope for this project so I'm just using numPy.
Download the files, use pip to install numPy, and run each file. It's not organized very well (and named quite poorly), but in general I made a new file for each new video in his series, and didn't remove old files.
I plan on creating website where you can easily run each project I did (with each different technique), so it's easy to access.
It is heavily inspired by Andrej Karpathy's Micrograd/makeMore series.