Dynamic-dLLM: Dynamic Cache-Budget and Adaptive Parallel Decoding for Training-Free Acceleration of Diffusion LLM
Quick Answer
Dynamic-dLLM introduces a training-free framework that enhances inference efficiency of diffusion LLMs like LLaDA-8B-Instruct by over 3 times.
Quick Take
It employs Dynamic Cache Updating and Adaptive Parallel Decoding to optimize performance on benchmarks such as and GSM8K, outperforming existing acceleration methods. This solution allows for efficient deployment without sacrificing model performance.
Key Points
- Dynamic Cache Updating allocates cache budgets based on token dynamics across layers.
- Adaptive Parallel Decoding calibrates thresholds for balancing quality and efficiency.
- Achieves over 3x speedup in inference while maintaining performance on various benchmarks.
- Outperforms state-of-the-art acceleration methods in efficiency and effectiveness.
- Code available for implementation at GitHub: https://github.com/TianyiWu233/DYNAMIC-DLLM.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Diffusion Large Language Models (dLLMs) offer a promising alternative to autoregressive models, excelling in text generation tasks due to their bidirectional attention mechanisms. However, their computational complexity scales on the order of L cubed with the sequence length L. This poses significant challenges for long-sequence and real-time applications, primarily due to the lack of compatibility with key-value caching and the non-autoregressive nature of denoising steps. Existing acceleration methods rely on static caching or parallel decoding strategies, which fail to account for the dynamic behavior of token properties across layers and decoding steps. We propose Dynamic-dLLM, a training-free framework that enhances dLLM inference efficiency through two components: Dynamic Cache Updating (DCU), which adaptively allocates cache-update budgets based on layer-wise token dynamics, and Adaptive Parallel Decoding (APD), which dynamically calibrates decoding thresholds to balance generation quality and efficiency. Extensive experiments on models like LLaDA-8B-Instruct, LLaDA-1.5, and Dream-v0-7B-Instruct across benchmarks such as MMLU, GSM8K, and HumanEval demonstrate that Dynamic-dLLM significantly improves inference speed. It attains an average speedup exceeding 3 times while maintaining performance. Dynamic-dLLM outperforms state-of-the-art acceleration methods and provides a plug-and-play solution for efficient dLLM deployment without compromising performance. The code is available at this https URL.
| Subjects: | Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2606.26120 [cs.CL] |
| (or arXiv:2606.26120v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2606.26120 arXiv-issued DOI via DataCite |
Submission history
From: Yulin Li [view email]
[v1]
Wed, 27 May 2026 02:47:50 UTC (5,328 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →TriAgent: Divergence-Aware Committees for Cost-Efficient Financial Sentiment Analysis
TriAgent introduces a cost-efficient multi-agent system for financial sentiment analysis, combining VADER, FinBERT, and Qwen2.5. It achieves an F1 score of ~0.87 with significant savings of $9.3M/year at a 10M-user scale compared to GPT-4o-mini, while also detecting hallucinations with an AUC of 0.90.