llama.cpp - faster and easier for everyone! - Xuan-Son Nguyen - Hugging Face - dotAI 2026

Why watch

Xuan-Son Nguyen, core maintainer of llama.cpp, presents the evolution of llama.cpp from a single-file experiment to a full local-inference platform. The talk covers speculative decoding and its various implementations (draft model, n-gram, MTP, DFlash, DSpark), explaining why it provides 'free' performance gains. The talk includes a live demo showing how to install and run llama.cpp in seconds. Ideal for developers interested in on-device AI and optimizing LLM inference.