DyLLM: Efficient Diffusion LLM Inference via Saliency-based Token Selection and Partial Attention

• URL: http://arxiv.org/abs/2603.08026v1
• Date: Mon, 09 Mar 2026 07:02:01 GMT
• Title: DyLLM: Efficient Diffusion LLM Inference via Saliency-based Token Selection and Partial Attention
• Abstract summary: We present DyLLM, a training-free inference framework that accelerates decoding by selectively computing only salient tokens.<n>Across diverse reasoning and code-generation benchmarks, DyLLM achieves up to 9.6x higher throughput.
• Score: 2.7422645382944935
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Masked Diffusion Language Models (MDLMs) enable parallel token decoding, providing a promising alternative to the sequential nature of autoregressive generation. However, their iterative denoising process remains computationally expensive because it repeatedly processes the entire sequence at every step. We observe that across these diffusion steps, most token representations remain stable; only a small subset, which we term salient tokens, contributes meaningfully to the next update. Leveraging this temporal sparsity, we present DyLLM, a training-free inference framework that accelerates decoding by selectively computing only these salient tokens. DyLLM identifies saliency by measuring the cosine similarity of attention contexts between adjacent denoising steps. It recomputes feed-forward and attention operations only for salient tokens while reusing cached activations for the remainder. Across diverse reasoning and code-generation benchmarks, DyLLM achieves up to 9.6x higher throughput while largely preserving the baseline accuracy of state-of-the-art models like LLaDA and Dream.

Related papers

• Masked Language Flow Models [10.204151640924454]
  Flow Language Models (FLMs) sidestep this limitation by learning a continuous flow that transports noise toward clean sequences.<n>We introduce Masked Language Flow Models (MLFMs) which incorporate masking into FLMs using a continuous interpolant to bridge partially masked and clean sequences.<n>We find, for the first time, that flow-based language models can be scaled to solve downstream reasoning and instruction-following tasks.
  arXiv  Detail & Related papers  (2026-06-26T00:16:40Z)
• SAID: Accelerating Diffusion-Based Language Models via Scaffold-Aware Iterative Decoding [11.233923872369202]
  Diffusion large language models (DLLMs) enable non-autoregressive computation generation by iteratively denoising corrupted tokens with bidirectional context.<n>We propose SAID, a Scaffold-Aware Iterative Decoding framework that acceleratesDLLMs by reallocating across tokens.<n>Experiments on LLaDA-8B and LLaDA 1.5 across math, coding, and knowledge benchmarks show that SAID significantly accelerates inference with a maximum speedup of 9.1x while maintaining competitive performance.
  arXiv  Detail & Related papers  (2026-06-03T14:56:42Z)
• SimSD: Simple Speculative Decoding in Diffusion Language Models [61.33773959352141]
  Diffusion large language models (dLLMs) offer faster inference through parallel or blockwise decoding.<n>Their masked language modeling formulation remains incompatible with standard token-level speculative decoding.<n>We propose a speculative decoding algorithm for dLLMs, named SimSD, that equips dLLMs with temporally valid token-level contexts.<n>Our method achieves up to 7.46x higher decoding throughput while maintaining and even improving average generation quality.
  arXiv  Detail & Related papers  (2026-06-01T17:46:46Z)
• Dependency-Aware Parallel Decoding via Attention for Diffusion LLMs [8.800270601225668]
  Parallel decoding for diffusion LLMs is difficult because each denoising step provides only token-wise marginal distributions.<n>We propose Dependency-Aware Parallel Decoding (DAPD), a simple, training-free decoding method that uses self-attention to induce a conditional dependency graph.
  arXiv  Detail & Related papers  (2026-03-13T13:52:02Z)
• Divide and Conquer: Accelerating Diffusion-Based Large Language Models via Adaptive Parallel Decoding [6.755667885643806]
  Diffusion-based large language models (dLLMs) have shown promising performance across various reasoning tasks.<n>We introduce an adaptive parallel decoding approach, namely DiCo, which features a three-phase divide-and-conquer paradigm.<n>Extensive experiments demonstrate that DiCo can achieve significant inference speedups while maintaining competitive generation quality.
  arXiv  Detail & Related papers  (2026-02-27T08:36:06Z)
• Prism: Efficient Test-Time Scaling via Hierarchical Search and Self-Verification for Discrete Diffusion Language Models [96.0074341403456]
  Inference-time compute has re-emerged as a practical way to improve LLM reasoning.<n>Most test-time scaling (TTS) algorithms rely on autoregressive decoding.<n>We propose Prism, an efficient TTS framework for dLLMs.
  arXiv  Detail & Related papers  (2026-02-02T09:14:51Z)
• Residual Context Diffusion Language Models [90.07635240595926]
  Residual Context Diffusion (RCD) is a module that converts discarded token representations into contextual residuals and injects them back for the next denoising step.<n>RCD consistently improves frontier dLLMs by 5-10 points in accuracy with minimal extra computation overhead.
  arXiv  Detail & Related papers  (2026-01-30T13:16:32Z)
• D$^{3}$ToM: Decider-Guided Dynamic Token Merging for Accelerating Diffusion MLLMs [22.78575203353886]
  Diffusion-based multimodal large language models (Diffusion MLLMs) exhibit substantially slower inference than autoregressive models.<n>We propose D$3$ToM, a Decider-guided dynamic token merging method to accelerate inference in Diffusion MLLMs.<n>Experiments show that D$3$ToM accelerates inference while preserving competitive performance.
  arXiv  Detail & Related papers  (2025-11-15T16:24:12Z)
• Continuous Autoregressive Language Models [56.49239051750678]
  We introduce Continuous Autoregressive Language Models (CALM)<n>CALM uses a high-fidelity autoencoder to compress a chunk of K tokens into a single continuous vector.<n>We develop a comprehensive likelihood-free framework that enables robust training, evaluation, and controllable sampling.
  arXiv  Detail & Related papers  (2025-10-31T17:58:11Z)
• Sparse-dLLM: Accelerating Diffusion LLMs with Dynamic Cache Eviction [72.27673320976933]
  Diffusion Large Language Models (dLLMs) enable breakthroughs in reasoning and parallel decoding.<n>Current caching techniques accelerate decoding by storing full-layer states, yet impose substantial memory usage.<n>We propose Sparse-dLLM, the first training-free framework integrating dynamic cache eviction with sparse attention.
  arXiv  Detail & Related papers  (2025-08-04T16:14:03Z)
• COrAL: Order-Agnostic Language Modeling for Efficient Iterative Refinement [80.18490952057125]
  Iterative refinement has emerged as an effective paradigm for enhancing the capabilities of large language models (LLMs) on complex tasks. We propose Context-Wise Order-Agnostic Language Modeling (COrAL) to overcome these challenges. Our approach models multiple token dependencies within manageable context windows, enabling the model to perform iterative refinement internally.
  arXiv  Detail & Related papers  (2024-10-12T23:56:19Z)
• Parallel Decoding via Hidden Transfer for Lossless Large Language Model Acceleration [54.897493351694195]
  We propose a novel parallel decoding approach, namely textithidden transfer, which decodes multiple successive tokens simultaneously in a single forward pass. In terms of acceleration metrics, we outperform all the single-model acceleration techniques, including Medusa and Self-Speculative decoding.
  arXiv  Detail & Related papers  (2024-04-18T09:17:06Z)
• SPEED: Speculative Pipelined Execution for Efficient Decoding [35.45955948053644]
  We propose SPEED, which improves inference efficiency by speculatively executing multiple future tokens in parallel with the current token. For Transformer decoders that employ parameter sharing, the memory operations for the tokens executing in parallel can be amortized. We demonstrate the efficiency of our method in terms of latency reduction relative to model accuracy and demonstrate how speculation allows for training deeper decoders with parameter sharing with minimal runtime overhead.
  arXiv  Detail & Related papers  (2023-10-18T16:07:01Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.