The Local AI Handbook
Build a custom GPU rig at home, from electrician wiring and sockets through cooling and GPU tradeoffs for training vs local inference. Covers NVIDIA, Apple, AMD, and Intel so you know what to buy.
$30
Books
Porting Video Games Into Fast Simulators With Agents
Reimplement games in rust, C, and CUDA until they run fast enough to train neural nets in them. Covers Minecraft 1.11.2, New Super Mario Bros Wii, Mario Kart Wii, and Terraria, plus the goals and decisions at each step.
$30
The Physics of LLM Inference
Build your own LLM serving engine from scratch. Covers hardware-level optimization, memory management, custom CUDA and Triton kernel development, and throughput maximization. 113 pages + full code repository.
$5+
Grokking Megakernels
Your GPU is sitting 50% idle during inference. This book shows you exactly why and how to fix it with a single fused CUDA kernel. Covers kernel fusion, memory bandwidth bottlenecks, and building an end-to-end megakernel.
$5+
The RL Post-Training Handbook
Implement reinforcement learning techniques for building reasoning capabilities into language models on a single GPU. Covers policy gradients, GRPO, think tokens, and memory-efficient training. 60% code, 40% prose.
$5
LLM Pre-Training for Dummies
Learn GPT-2 pre-training on a single GPU. Covers tokenization, embeddings, attention mechanisms, transformer architecture, training loops, and optimization. Based on nanoGPT - minimal code, maximum understanding.
$5+
Raw JAX
Stop translating NumPy tutorials into JAX by trial and error. This book gives you every core concept side-by-side with runnable code and real GPU output. Covers JAX fundamentals, transformations, and GPU acceleration.
$5+