AIbase
Product LibraryTool Navigation

SpS-SpecDec

Public

SpS-SpecDec: a fast Python lib that boosts autoregressive LM inference with speculative decoding. Inspired by DeepMind, it guesses multiple tokens using a small draft model, verifies with a big one. Get 2-2.5x speedups, no quality drop!

Creat2025-03-06T15:44:04
Update2025-03-09T14:34:04
2
Stars
0
Stars Increase