dMoE: dLLMs with Learnable Block Experts

• URL: http://arxiv.org/abs/2605.30876v1
• Date: Fri, 29 May 2026 06:03:50 GMT
• Title: dMoE: dLLMs with Learnable Block Experts
• Abstract summary: We propose dMoE, a simple yet effective block-level MoE framework.<n>We show that dMoE substantially reduces the number of uniquely activated experts during inference.<n>It also reduces memory usage by 76.64% to 79.84% and achieves 1.14$times$ to 1.66$times$ end-to-end latency speedup.
• Score: 71.572316901001
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Diffusion Large Language Models (dLLMs) have recently emerged as a promising alternative to autoregressive models, offering competitive performance while naturally supporting parallel decoding. However, as dLLMs are increasingly integrated with Mixture-of-Experts (MoE) architectures to scale model capacity, a fundamental mismatch arises between block parallel decoding and token-level expert selection. Specifically, each dLLM forward pass processes multiple tokens with bidirectional dependencies, whereas conventional MoE layers route each token independently. This mismatch substantially increases the number of uniquely activated experts, making inference increasingly memory-bound. To address this, we propose dMoE, a simple yet effective block-level MoE framework. The central idea of dMoE is to aggregate token-level expert distributions within each block into a unified block-level expert distribution, which is then used to guide expert routing in a more coherent manner. In this way, dMoE substantially reduces the number of uniquely activated experts during inference without sacrificing performance, thereby mitigating the memory-bound bottleneck. Extensive experiments across a variety of benchmarks demonstrate the effectiveness of dMoE. On average, dMoE reduces the number of uniquely activated experts from 69.5 to 14.6 while retaining 99.11% of the original performance. Meanwhile, it reduces memory usage by 76.64% to 79.84% and achieves 1.14$\times$ to 1.66$\times$ end-to-end latency speedup. Code is available at: https://github.com/fscdc/dMoE

Related papers

• SERE: Similarity-based Expert Re-routing for Efficient Batch Decoding in MoE Models [19.56443760368644]
  We present SERE, a Similarity-based Expert Re-routing method for Efficient batch decoding in MoE models.<n>SERE dynamically reduces the number of active experts by re-routing tokens from secondary experts to their most similar primary counterparts.<n>SERE achieves up to 2.0x speedup with minimal quality loss, providing a practical solution for cost-efficient and latency-sensitive large-scale MoE deployment.
  arXiv  Detail & Related papers  (2026-02-07T16:51:16Z)
• Dynamic Expert Sharing: Decoupling Memory from Parallelism in Mixture-of-Experts Diffusion LLMs [22.399470395813577]
  Dynamic Expert Sharing (DES) is a novel technique that shifts MoE optimization from token-centric pruning to sequence-level coreset selection.<n>DES reduces unique expert activations by over 55% and latency by up to 38%, while retaining 99% of vanilla accuracy.
  arXiv  Detail & Related papers  (2026-01-31T20:01:47Z)
• SonicMoE: Accelerating MoE with IO and Tile-aware Optimizations [54.303301888915406]
  Mixture of Experts (MoE) models have emerged as the de facto architecture for scaling up language models without significantly increasing the computational cost.<n>We propose a memory-efficient algorithm to compute the forward and backward passes of MoEs with minimal activation caching.<n>We also propose a novel "token rounding" method that minimizes the wasted compute due to padding in Grouped GEMM kernels.
  arXiv  Detail & Related papers  (2025-12-16T04:39:10Z)
• MoDES: Accelerating Mixture-of-Experts Multimodal Large Language Models via Dynamic Expert Skipping [52.02659589971978]
  We propose MoDES, the first training-free framework that adaptively skips experts to enable efficient and accurate MoE MLLM inference.<n>MoDES significantly enhances inference speed, improving the prefilling time by 2.16$times$ and the decoding time by 1.26$times$.
  arXiv  Detail & Related papers  (2025-11-19T18:48:27Z)
• PuzzleMoE: Efficient Compression of Large Mixture-of-Experts Models via Sparse Expert Merging and Bit-packed inference [17.441141633991197]
  We introduce PuzzleMoE, a training-free MoE compression method that achieves high accuracy and efficient inference through two key innovations.<n>First, PuzzleMoE performs sparse expert merging by identifying element-wise weight redundancy and specialization.<n>Second, to avoid the overhead of storing binary masks and signs, PuzzleMoE introduces a bit-packed encoding scheme that reuses underutilized exponent bits.
  arXiv  Detail & Related papers  (2025-11-06T20:53:02Z)
• MoBiLE: Efficient Mixture-of-Experts Inference on Consumer GPU with Mixture of Big Little Experts [17.518573710849513]
  MoBiLE is a plug-and-play offloading-based MoE inference framework with textitmixture of big-little experts.<n>MoBiLE achieves a speedup of 1.60x to 1.72x compared to the baseline on a consumer GPU system, with negligible degradation in accuracy.
  arXiv  Detail & Related papers  (2025-10-14T10:22:44Z)
• ResMoE: Space-efficient Compression of Mixture of Experts LLMs via Residual Restoration [61.579842548990754]
  Mixture-of-Experts (MoE) Transformer, the backbone of multiple phenomenal language models, leverages sparsity by activating only a fraction of model parameters for each input token.<n>We introduce ResMoE, an innovative MoE approximation framework that utilizes Wasserstein barycenter to extract a common expert (barycenter expert) and approximate the residuals between this barycenter expert and the original ones.
  arXiv  Detail & Related papers  (2025-03-10T03:15:54Z)
• MoE++: Accelerating Mixture-of-Experts Methods with Zero-Computation Experts [63.67734699877724]
  MoE++ is a general and heterogeneous MoE framework that integrates both Feed-Forward Network(FFN) and zero-computation experts. MoE++ achieves better performance while delivering 1.1-2.1x expert forward throughput compared to a vanilla MoE model of the same size.
  arXiv  Detail & Related papers  (2024-10-09T18:01:27Z)
• Unchosen Experts Can Contribute Too: Unleashing MoE Models' Power by Self-Contrast [58.98411447739218]
  Mixture-of-Experts (MoE) has emerged as a prominent architecture for scaling model size while maintaining computational efficiency. We propose Self-Contrast Mixture-of-Experts (SCMoE), a training-free strategy that utilizes unchosen experts in a self-contrast manner during inference. Our method is conceptually simple and computationally lightweight, as it incurs minimal latency compared to greedy decoding.
  arXiv  Detail & Related papers  (2024-05-23T12:45:29Z)

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.