Show HN: Every number in a tiny GPT, from one forward pass to one step of RLHF
TL;DR
com/manogyasingh/transformer-visualisation Made this with Opus 5.5 to understand the exact working of an LLM. It's a step by step walkthrough of a toy GPT-2 like model. Every matrix fits on screen, every number is computed live in the browser, and hovering over any cell shows the exact math of that step. It covers inference -- following the generation of one token from tokensiation to sampling, one training step of pretraining, and one PPO iteration of RLHF for post-training.
Nauti's Take
This is worth a short hands-on test as a learning and debugging aid: trace one token through the browser, then check which parts of the explanation carry over to your own model work. The source base is thin, with almost no Hacker News discussion so far.
Treat the RLHF walkthrough as a teaching simplification and verify the details against the repository before using it to guide production decisions.
Sources