Verification-Aware Training for Speculative Decoding

类型：arxiv
标识：2608.30135
链接：https://arxiv.org/abs/2608.30135
主分类：engineering
形态：position
TLDR：Speculative decoding accelerates large language model inference by using a draft model to generate candidate tokens, which are verified by the target model in a single forward pass. Verification proceeds sequentially and discards every position from the first rejection onward, yet existing draft training relies on token-level imitation of the target with a fixed per-position weighting that reflects neither property. We introduce Verification-Aware Training (VAT), a plug-in framework that simulates verification at every training step and turns the resulting accept and reject patterns into super
待LLM分类：否
来源文件：
/inbox/tom/_candidates/2026-09-01-agent-rag-longcontext-candidates.json

2026-07-04

笔记 KVAE Tokenizers arXiv:2608.05798 · multimodal · 精读与批判 v2

2026-08-09

仓库 griver/Q-RAG

2026-08-13

笔记精读笔记：Characterizing State Space Model and Hybrid Language Model Performance with Long Context

2026-08-30

仓库 leon2k2k2k/llm-feedback-alignment

2026-08-11

仓库 LearnPrompt/humanize-ppt

2026-07-31

arXiv

2608.30135

TLDR

Speculative decoding accelerates large language model inference by using a draft model to generate candidate tokens, which are verified by the target model in a single forward pass. Verification proceeds sequentially and discards every position from the first rejection onward, yet existing draft training relies on token-level imitation of the target with a fixed per-position weighting that reflects neither property. We introduce Verification-Aware Training (VAT), a plug-in framework that simulates verification at every training step and turns the resulting accept and reject patterns into super

Verification-Aware Training for Speculative Decoding

相关 · 按语义匹配