---
id: 20260718-T0-06
title: "Polestar：漂移感知缓存校准与Token提交机制，加速扩散LLM推理"
title_en: "Polestar: Drift-Aware Cache Calibration Speeds Up Diffusion LLM Inference"
url: https://ai.daily.yangsir.net/daily/20260718-T0-06
issue_date: 2026-07-18
publish_date: 2026-07-17T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2607.14107
---

# Polestar：漂移感知缓存校准与Token提交机制，加速扩散LLM推理

研究团队提出Polestar，一种针对扩散大语言模型（dLLM）的推理效率优化方案。该方法解决了两大挑战：双向注意力机制导致KV缓存无法有效复用，以及静态置信度阈值解码并行性不佳。Polestar通过漂移感知缓存校准和Token提交机制，在保证生成质量的前提下显著提升dLLM的推理速度。

## English Version

**Polestar: Drift-Aware Cache Calibration Speeds Up Diffusion LLM Inference**

Researchers propose Polestar, a method to accelerate inference for diffusion large language models (dLLMs). It tackles two key bottlenecks: bidirectional attention preventing efficient KV-cache reuse, and static confidence thresholds limiting decoding parallelism. Polestar employs drift-aware cache calibration and token commitment mechanisms to boost inference speed while maintaining generation quality.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2607.14107)

**详情页**：https://ai.daily.yangsir.net/daily/20260718-T0-06

---

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