Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Liu, Liming, Huang, Binxuan, Zhang, Zixuan, Liu, Xin, Yin, Bing, Zhao, Tuo
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2601.06428
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910158078083072
author Liu, Liming
Huang, Binxuan
Zhang, Zixuan
Liu, Xin
Yin, Bing
Zhao, Tuo
author_facet Liu, Liming
Huang, Binxuan
Zhang, Zixuan
Liu, Xin
Yin, Bing
Zhao, Tuo
contents Diffusion Language Models (DLMs) decode multiple tokens in parallel, but aggressive multi-token decoding amplifies cross-token dependency errors and can sharply degrade generation quality. We propose BackPlay, a frozen-backbone self-correction framework that trains only a lightweight correction head on a finetuned DLM without updating any backbone or adapter parameters. Because the head is trained on errors produced by the same frozen generator used at inference time, its training distribution aligns with the error patterns of the deployed model. We further introduce Look-back Correction, a training mechanism that injects predictions from earlier, more corrupted denoising states into later, richer contexts, enabling the head to leverage later context to detect mistakes made in earlier generation steps. During inference, BackPlay periodically revisits previously generated tokens through selective remasking and regeneration to limit error accumulation. Across mathematical reasoning and code generation benchmarks, BackPlay improves the speed--quality trade-off of the underlying DLM under multi-token decoding.
format Preprint
id arxiv_https___arxiv_org_abs_2601_06428
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle BackPlay: Head-Only Look-Back Self-Correction for Diffusion Language Models
Liu, Liming
Huang, Binxuan
Zhang, Zixuan
Liu, Xin
Yin, Bing
Zhao, Tuo
Machine Learning
Diffusion Language Models (DLMs) decode multiple tokens in parallel, but aggressive multi-token decoding amplifies cross-token dependency errors and can sharply degrade generation quality. We propose BackPlay, a frozen-backbone self-correction framework that trains only a lightweight correction head on a finetuned DLM without updating any backbone or adapter parameters. Because the head is trained on errors produced by the same frozen generator used at inference time, its training distribution aligns with the error patterns of the deployed model. We further introduce Look-back Correction, a training mechanism that injects predictions from earlier, more corrupted denoising states into later, richer contexts, enabling the head to leverage later context to detect mistakes made in earlier generation steps. During inference, BackPlay periodically revisits previously generated tokens through selective remasking and regeneration to limit error accumulation. Across mathematical reasoning and code generation benchmarks, BackPlay improves the speed--quality trade-off of the underlying DLM under multi-token decoding.
title BackPlay: Head-Only Look-Back Self-Correction for Diffusion Language Models
topic Machine Learning
url https://arxiv.org/abs/2601.06428