Code Explorer

This list is read from data/generated/code-index.json, so it holds every Python file the extractor found and nothing else. Files in tests/ are labelled as tests.

File	Kind	Lines	Functions	Purpose
checkpoint.py	Source	46	2	Save and load weights, model shape and tokenizer
config.py	Source	44	0	Hyperparameters as module-level constants, plus two import-time assertions
data.py	Source	43	5	Corpus loading, encoding, split, windows, batches
generate.py	Source	35	2	Sampling and the generation loop
gradcheck.py	Source	33	3	Finite-difference gradient checker, used only by tests
ops.py	Source	60	12	Differentiable operations with hand-written backward functions
optim.py	Source	50	5	Adam with decoupled weight decay, learning-rate schedule, gradient clipping
talk.py	Source	30	1	Interactive inference entry point
tokenizer.py	Source	119	10	Byte-pair-encoding (BPE) tokenizer, working on characters
train.py	Source	85	3	Training entry point
transformer.py	Source	190	15	Parameters, forward pass, backward pass
`tests/`
test_checkpoint.py	Test	76	8	Checkpoint tests
test_data.py	Test	83	11	Data pipeline tests
test_generate.py	Test	82	13	Sampling tests
test_gradcheck.py	Test	39	4	Gradient-check tests
test_ops.py	Test	88	10	Operation tests
test_optim.py	Test	49	6	Optimizer, schedule and clipping tests
test_tokenizer.py	Test	50	9	Tokenizer tests
test_transformer.py	Test	110	11	Model tests

Each file's page shows where the file fits in the pipeline, what it imports, what uses it according to the call graph, the concepts that link to it, and its source with real line numbers.

A link to a function scrolls to it and highlights it, for example ops.softmax_forward. A link to one line works the same way, for example the line that saves the weights.

The call graph is static. It follows calls by name and by known types, so a call on a value of unknown type, such as a callback parameter, creates no "used by" entry.