Qwen3 Speculative Decoding on the DGX Spark: Two Models, Four Methods, One Surprising Lesson

A beginner's guide to one of the most practical speed techniques for AI inference, tested on real hardware with real numbers.