LLM Digest

Story

arxiv_cs_cl · Nov 10, 2026 · paper

Source brief

REAL Loss: Optimizing Real Expected Acceptance Length for Speculative Decoding

arxiv.orgNov 10, 2026
original source linked

In brief

Speculative decoding accelerates LLM inference by having a draft model propose multiple tokens for the target model to verify in one pass. The speedup is governed by the expected acceptance length (EAL), yet standard...

Feed lens

evaluation

Read the original at arxiv.org →Open in live feed Read that day’s brief

REAL Loss: Optimizing Real Expected Acceptance Length for Speculative Decoding

Earlier in this thread 4 items

Training Parallel Speculative Draft Models by Directly Minimizing Expected Decoding Rounds

Phrase-Level Speculative Decoding on a Frozen Target

DForkTree: Accelerating Speculative Decoding with Fork-Aware Block Diffusion Draft Trees

D-REAL: Acceptance-Length Optimization with Unbiased Gradients for Speculative Decoding