---
id: 20260603-T0-11
title: "DLLM-JEPA：掩码扩散语言模型新架构"
title_en: "DLLM-JEPA: New Architecture for Masked Diffusion Language Models"
url: https://ai.daily.yangsir.net/daily/20260603-T0-11
issue_date: 2026-06-03
publish_date: 2026-06-02T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2606.00091
---

# DLLM-JEPA：掩码扩散语言模型新架构

arXiv论文提出DLLM-JEPA架构，将联合嵌入预测框架(JEPA)应用于掩码扩散语言模型。新方法解决了因果注意力机制的高计算成本问题，提升自监督学习效率。

## English Version

**DLLM-JEPA: New Architecture for Masked Diffusion Language Models**

The arXiv paper introduces DLLM-JEPA, applying Joint Embedding Predictive Architectures to masked diffusion language models. This method addresses the high computational costs of causal attention mechanisms, improving self-supervised learning efficiency.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2606.00091)

**详情页**：https://ai.daily.yangsir.net/daily/20260603-T0-11

---

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