A maintained reference
Foundations
The math, implementation, and intuition behind modern ML systems. Start with Transformer Internals: an ordered path from attention to a working decoder.
Unlike blog posts, these are maintained over time as understanding deepens. Check the 'last updated' date on each.
Read alongside the RLVR implementation and test suite . Published parts appear below; later parts remain planned.
Transformer Internals
A ground-up walkthrough of the transformer architecture — from the math of attention through positional encoding to a complete forward pass, with tested PyTorch implementations at every step.