3 GitHub Repos Worth Reading for the Code: nanoGPT, tinygrad, llama.cpp
cneuralnetwork · x · 2026-11-12
A curated list of 3 GitHub repos worth reading for the code itself, not just using:
- nanoGPT (karpathy): understand how GPT works from scratch — attention, transformer blocks, and token generation; start at model.py (330 lines).
- tinygrad: tensor.py teaches tensor ops, autograd, and how a deep learning framework is built.
- llama.cpp: src/llama-model.cpp covers LLM architectures, weight loading, and inference internals.
All three are open source and form a solid bottom-up learning path for LLM implementation.
More from coding & agent
- Meta open-sources Astryx, the design system behind 13,000+ internal apps — victor_explore · 2026-11-12
- 8 Parallel Agents Code an 83-Second Anime Short in Under 3 Hours — FinanceYF5 · 2026-11-12
- Fable 5.5 codes an 83-second animated short in under 3 hours, zero video models — FinanceYF5 · 2026-11-12
- Running bots as an expert team, author pivots to a scored "Checked Loop" workflow — JoeJustice · 2026-11-12
- Dev makes an AI agent CEO of a puzzle game company, shipping a new game daily — RichardsonDx · 2026-11-12
- Why AI agents forget: a primer on the memory layer missing from agent architecture — sharpeye_wnl · 2026-11-12