01_tokenizer
tokenization
naive, byte-level, and byte-pair tokenization in one inspectable notebook.
6 saved output blocks
read notebook →a route through the work
start with tokenization, then the bigram baseline, then attention. autograd branches into the differentiation mechanism underneath. each record exposes the original code and retained outputs; incomplete work is labeled as such.
start with tokenization →02_bigram
the bigram baseline
token streams, training and validation splits, and a count-based starting point.
5 saved output blocks
read notebook →03_attention
attention
authored implementation and comparison notes. this revision does not contain saved cell outputs.
source only · no saved cell outputs
read notebook →04_autograd
automatic differentiation
an implementation notebook with retained cell outputs. source and output are shown together.
7 saved output blocks
read notebook →publication boundary
these are static, commit-pinned exports of selected public notebooks. this site never executes training or notebook cells. source-only records are implementation workspaces, not completed experimental results.
inspect the source revision ↗