AI Paper Atlas

Research map / Learning theory and algorithms / Optimal transport

SpecTr: Fast Speculative Decoding via Optimal Transport

Ziteng Sun, Ananda Theertha Suresh, Jae Hun Ro, Ahmad Beirami, Himanshu Jain, Felix Yu

NeurIPS 2023 · Learning theory and algorithms › Optimal transport · approach: Neural optimal transport · 8 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use SpecTr K=8 instead of speculative K=1 (L=4)speculative decoding · LM1B PaLM-2block efficiency: 2.4 -> 3.1Improved
use SpecTr K=8 instead of speculative K=1 (L=4)speculative decoding · LM1B PaLM-2wall clock speedup: 1.67 -> 2.08Improved
use SpecTr K=8 instead of speculative K=1 (L=8)speculative decoding · LM1B PaLM-2block efficiency: 2.9 -> 4.0Improved
use SpecTr K=8 instead of speculative K=1 (L=8)speculative decoding · LM1B PaLM-2wall clock speedup: 1.56 -> 2.13Improved
increase draft length L from 4 to 8 (ablation)speculative decoding K=1 · LM1B PaLM-2wall clock speedup: 1.67 -> 1.56Hurt
increase draft length L from 4 to 8 (ablation)SpecTr K=8 · LM1B PaLM-2wall clock speedup: 2.08 -> 2.13Improved
increase number of drafts K from 1 to 8 (L=4) (ablation)SpecTr, 2M draft · LM1B 97M largeblock efficiency: 1.86 -> 2.56Improved
increase number of drafts K from 1 to 8 (L=8) (ablation)SpecTr, 2M draft · LM1B 97M largeblock efficiency: 1.91 -> 2.68Improved
increase batch from 1 to 8 for large model scoring (length 4) (ablation)97M transformer · GPU latencyrelative latency: 1.00 -> 1.10No clear effect
increase batch from 1 to 8 for large model scoring (length 8) (ablation)97M transformer · GPU latencyrelative latency: 1.01 -> 1.42Mixed
increase batch from 1 to 8 for 6M draft (length 4) (ablation)6M transformer · GPU latencyrelative latency: 0.18 -> 0.20No clear effect
wall clock vs theoretical block efficiencySpecTr K=8 L=8 · LM1B PaLM-2speedup: 4.0 -> 2.13Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy