Minimal GPT2 Inference in C++
Here’s a little project which runs GPT2 inference with every operation rolled out by hand to be inspectable and debuggable, in C++, with no dependences. Ideal for learning about the Transformer mechanism.
Written on June 11, 2025