Fast Inference from Transformers via Speculative Decoding
Source
arxiv.org
Author
Yaniv Leviathan et al.
Date
Why it matters
It makes generation faster without changing what the model outputs — a small model drafts, the large one verifies in parallel, rejected tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → fall back. Every fast inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → provider runs some version of it, and the identical-distribution guarantee is why it costs no quality.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
