Salvato in:
Dettagli Bibliografici
Autori principali: Wen, Zhuofan, Feng, Yang
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:https://arxiv.org/abs/2604.12247
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917406149967872
author Wen, Zhuofan
Feng, Yang
author_facet Wen, Zhuofan
Feng, Yang
contents Speculative decoding has emerged as a promising approach to accelerate autoregressive inference in large language models (LLMs). Self-draft methods, which leverage the base LLM itself for speculation, avoid the overhead of auxiliary draft models but face limitations: shallow layers often produce overconfident yet incorrect token predictions, and the presence of difficult tokens in a draft sequence forces redundant computation through deeper layers, undermining both draft acceptance and overall speedup. To address these issues, we propose a novel self-draft framework that suppresses spurious confidence via layer-wise temperature annealing in early-exit decision and adaptively bounds speculation length based on token-wise decoding difficulty. By reprocessing the hidden states of draft tokens in a unified parallel pass through deep layers, our method maintains exact output equivalence with the original model while maximizing computational efficiency. It requires no modifications to the base LLM parameters and achieves up to 2.33x wall-time speedup over standard autoregressive decoding across diverse long-form generation tasks and multiple model architectures.
format Preprint
id arxiv_https___arxiv_org_abs_2604_12247
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle SpecBound: Adaptive Bounded Self-Speculation with Layer-wise Confidence Calibration
Wen, Zhuofan
Feng, Yang
Computation and Language
Artificial Intelligence
Machine Learning
Speculative decoding has emerged as a promising approach to accelerate autoregressive inference in large language models (LLMs). Self-draft methods, which leverage the base LLM itself for speculation, avoid the overhead of auxiliary draft models but face limitations: shallow layers often produce overconfident yet incorrect token predictions, and the presence of difficult tokens in a draft sequence forces redundant computation through deeper layers, undermining both draft acceptance and overall speedup. To address these issues, we propose a novel self-draft framework that suppresses spurious confidence via layer-wise temperature annealing in early-exit decision and adaptively bounds speculation length based on token-wise decoding difficulty. By reprocessing the hidden states of draft tokens in a unified parallel pass through deep layers, our method maintains exact output equivalence with the original model while maximizing computational efficiency. It requires no modifications to the base LLM parameters and achieves up to 2.33x wall-time speedup over standard autoregressive decoding across diverse long-form generation tasks and multiple model architectures.
title SpecBound: Adaptive Bounded Self-Speculation with Layer-wise Confidence Calibration
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2604.12247