Open-source LLM Optimization Handbook: 48 topics with hands-on quantization, KV cache and kernel problems

udmrzn · x · 2026-11-18

TensorTonic open-sourced the LLM Optimization Handbook, a roadmap for making LLMs faster, cheaper and smaller. It covers 48 topics across 7 stages — inference fundamentals, quantization, KV cache, speculative decoding, GPU kernels and serving — each with a short explanation and a hands-on problem where you implement the idea in Python, CUDA or Triton and check against tests. The repo also curates the 49 most common inference/LLM engineer interview questions with explanations and solvable problems. MIT licensed.

Original post →

More from Infra

Infra channel →