20. Glossary and resources

These are the terms to use when you search or ask for help; the right word usually finds the right answer.

Term	Meaning	Part
AST	the parsed tree structure of source code	1
Autograd	PyTorch’s system that records operations and computes gradients	15
Backoff	waiting progressively longer between retries	8, 16
Broadcasting	combining arrays of different shapes by virtually stretching size-1 axes	13
Bytecode	the instructions CPython compiles source code into	1
Closure	a function that keeps variables from the scope where it was defined	5
Collate	combining individual examples into one batch	15
Coroutine	a function defined with async def that can pause at await	12
CPython	the reference Python implementation, written in C	1
Decorator	a function that wraps or replaces another function	5
Descriptor	an object with __get__ that controls attribute access	7
Dunder method	a special method such as __len__ that syntax calls	7
Eval	a test set that scores probabilistic model behavior	17
Event loop	the scheduler that runs coroutines in asyncio	12
Generator	a function with yield that produces values lazily	6
GIL	the lock that lets one thread run Python bytecode at a time	12
Hashable	usable as a dict key: has a stable hash consistent with equality	3
Idempotent	safe to repeat with the same effect	16
Iterable / iterator	something you can loop over / the one-shot object doing the looping	6
JSON Schema	a JSON description of data shape, used for tools and structured output	10, 16
Leaf tensor	a tensor created directly (like a parameter) whose .grad is filled	15
MRO	method resolution order: the class lookup chain	7
Mutable	its value can change in place	2
Reference count	the number of references to an object; at zero it is freed	2
SSE	server-sent events, the HTTP streaming format most LLM APIs use	6, 16
state_dict	a model’s parameters and buffers as a name-to-tensor dict	15
Stride	bytes to step in memory to move one position along an axis	13
ufunc	a NumPy function applied elementwise in compiled code	13
View	an array sharing memory with another array	13
Virtual environment	an isolated interpreter plus its own installed packages	9
Wheel	a prebuilt package file, often containing compiled code	9

Resources, in the order this guide uses them

• Python documentation: the official tutorial, then the “Data model” chapter of the Language Reference. That one chapter underlies Parts 2 to 7.
• Ned Batchelder, “Facts and Myths about Python names and values” (PyCon talk): Part 2 in about half an hour.
• Luciano Ramalho, Fluent Python: the best book-length treatment of Parts 2 to 8 and 12.
• Brett Slatkin, Effective Python: short, practical items on idiomatic code.
• David Beazley’s talks on generators and concurrency: the clearest explanations of Parts 6 and 12.
• Jake VanderPlas, Python Data Science Handbook, and Nicolas Rougier, From Python to NumPy: Parts 13 and 14.
• Andrej Karpathy, “Neural Networks: Zero to Hero”: micrograd, makemore, nanoGPT and a tokenizer build; the backbone of Part 19, phase 5.
• PyTorch official tutorials and the “Autograd mechanics” notes: Part 15.
• FastAPI, Pydantic and your LLM provider’s official docs: Parts 10 and 16. Prefer them over blog posts, because these APIs change quickly.
• Hugging Face course: transformers, tokenizers and datasets in practice.