Last Updated on 2025-07-01 by Clay Recently, I’ve still been diving into inference acceleration techniques, but work has kept me too busy to publish any updates. Today, I’m introducing a classic multi-head decoding architecture called Medusa. Medusa, inspired by the mythological Greek figure also known as the “snake-haired woman,” has each decoding head metaphorically representing … Continue reading [Paper Reading] Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads