---
id: 20260530-T0-11
title: "AR到扩散模型的高效适配方法"
title_en: "Efficient AR to diffusion model adaptation"
url: https://ai.daily.yangsir.net/daily/20260530-T0-11
issue_date: 2026-05-30
publish_date: 2026-05-29T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2605.27387
---

# AR到扩散模型的高效适配方法

研究提出严格因果与弹性跨度方法，解决预训练自回归模型与扩散模型的结构不兼容问题。该方法能够在保持模型性能的同时，实现更高效的并行文本生成。

## English Version

**Efficient AR to diffusion model adaptation**

A new method adapts Autoregressive models to diffusion models using strictly causal and elastic horizons, solving structural incompatibility while enabling efficient parallel text generation.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2605.27387)

**详情页**：https://ai.daily.yangsir.net/daily/20260530-T0-11

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*