Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ma, Xiangyu, Xiao, Teng, Li, Zuchao, Zhang, Lefei
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2605.27387
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866916057766166528
author Ma, Xiangyu
Xiao, Teng
Li, Zuchao
Zhang, Lefei
author_facet Ma, Xiangyu
Xiao, Teng
Li, Zuchao
Zhang, Lefei
contents Diffusion models promise efficient parallel text generation but rely on bidirectional attention, creating a structural mismatch with pre-trained Autoregressive (AR) models. This incompatibility precludes reusing robust AR priors, necessitating prohibitive pre-training from scratch. To bridge this gap, we propose FLUID, a framework that efficiently adapts AR backbones to the diffusion paradigm. By enforcing Strictly Causal Alignment, FLUID enables seamless initialization from standard GPT-style checkpoints, circumventing the need for massive pre-training. Furthermore, we introduce Elastic Horizons, an entropy-driven mechanism that dynamically modulates denoising strides based on local information density rather than fixed schedules. Experiments demonstrate that FLUID achieves state-of-the-art performance while reducing training costs by orders of magnitude, effectively reconciling established AR foundations with efficient parallel generation. Our code is available at https://github.com/Oli-lab-nun/FLUID/tree/main.
format Preprint
id arxiv_https___arxiv_org_abs_2605_27387
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle From AR to Diffusion: Efficiently Adapting Large Language Models with Strictly Causal and Elastic Horizons
Ma, Xiangyu
Xiao, Teng
Li, Zuchao
Zhang, Lefei
Computation and Language
Artificial Intelligence
Diffusion models promise efficient parallel text generation but rely on bidirectional attention, creating a structural mismatch with pre-trained Autoregressive (AR) models. This incompatibility precludes reusing robust AR priors, necessitating prohibitive pre-training from scratch. To bridge this gap, we propose FLUID, a framework that efficiently adapts AR backbones to the diffusion paradigm. By enforcing Strictly Causal Alignment, FLUID enables seamless initialization from standard GPT-style checkpoints, circumventing the need for massive pre-training. Furthermore, we introduce Elastic Horizons, an entropy-driven mechanism that dynamically modulates denoising strides based on local information density rather than fixed schedules. Experiments demonstrate that FLUID achieves state-of-the-art performance while reducing training costs by orders of magnitude, effectively reconciling established AR foundations with efficient parallel generation. Our code is available at https://github.com/Oli-lab-nun/FLUID/tree/main.
title From AR to Diffusion: Efficiently Adapting Large Language Models with Strictly Causal and Elastic Horizons
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2605.27387