AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Vegas: Self-Speculative Decoding with Verification-Guided Sparse Attention

Yikang Yue, Yuqi Xue, Jian Huang

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Long-context KV and memory attention · 0 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
select KV by attention over all draft tokens instead of last accepted token (ablation)Vegas · LongBench-v2acceptance probability across draft chainImproved
include discarded draft tokens in KV selection (ablation)Vegas · LongBench-v2, gamma=7avg accepted tokens per iteration: 3%--14%Improved
use only first draft and bonus token for logits instead of all draft tokens (ablation)Vegas · LongBench-v2, Qwen3-8B gamma=7avg accepted tokens: 6.13 -> 6.11No clear effect
use only first draft and bonus token for logits instead of all draft tokens (ablation)Vegas · LongBench-v2, Qwen3-8B gamma=11avg accepted tokens: 8.91 -> 8.72No clear effect
use only first draft and bonus token for logits instead of all draft tokens (ablation)Vegas · FA3 kernel, Qwen3-8Blogit collection overhead: 53% -> 5%Improved
use only first draft and bonus token for logits instead of all draft tokens (ablation)Vegas · FA3 kernel, gpt-oss-20blogit collection overhead: 189% -> 37%Improved
increase sparsity ratio (ablation)Vegas · LongBench-v2accepted tokens per iterationMixed
use page size 1 instead of larger pages up to 16 (ablation)Vegas sparse attention · FA3 kernel latencylatency: 197.6 vs 198.3 us (Qwen3-8B)No clear effect
use verification-guided KV selection instead of MagicDec-StreamVegas · AIME25 and CodeElodecoding throughput: 1.15x--1.23xImproved
use Vegas instead of vanilla vLLM decodingVegas · AIME25 and CodeElodecoding throughput: 1.25x--2.70xImproved

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