ReFusion: A Diffusion Large Language Model with Parallel Autoregressive Decoding

• URL: http://arxiv.org/abs/2512.13586v1
• Date: Mon, 15 Dec 2025 17:41:19 GMT
• Title: ReFusion: A Diffusion Large Language Model with Parallel Autoregressive Decoding
• Authors: Jia-Nan Li, Jian Guan, Wei Wu, Chongxuan Li,
• Abstract summary: Autoregressive models (ARMs) are hindered by slow sequential inference.<n>We introduce ReFusion, a novel masked diffusion model that achieves superior performance and efficiency.<n>ReFusion bridges the performance gap to strong ARMs while maintaining a 2.33$times$ average speedup.
• Score: 37.86179431483446
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Autoregressive models (ARMs) are hindered by slow sequential inference. While masked diffusion models (MDMs) offer a parallel alternative, they suffer from critical drawbacks: high computational overhead from precluding Key-Value (KV) caching, and incoherent generation arising from learning dependencies over an intractable space of token combinations. To address these limitations, we introduce ReFusion, a novel masked diffusion model that achieves superior performance and efficiency by elevating parallel decoding from the token level to a higher slot level, where each slot is a fixed-length, contiguous sub-sequence. This is achieved through an iterative ``plan-and-infill'' decoding process: a diffusion-based planning step first identifies a set of weakly dependent slots, and an autoregressive infilling step then decodes these selected slots in parallel. The slot-based design simultaneously unlocks full KV cache reuse with a unified causal framework and reduces the learning complexity from the token combination space to a manageable slot-level permutation space. Extensive experiments on seven diverse benchmarks show that ReFusion not only overwhelmingly surpasses prior MDMs with 34% performance gains and an over 18$\times$ speedup on average, but also bridges the performance gap to strong ARMs while maintaining a 2.33$\times$ average speedup.

Related papers

• Causal Autoregressive Diffusion Language Model [70.7353007255797]
  CARD reformulates the diffusion process within a strictly causal attention mask, enabling dense, per-token supervision in a single forward pass.<n>Our results demonstrate that CARD achieves ARM-level data efficiency while unlocking the latency benefits of parallel generation.
  arXiv  Detail & Related papers  (2026-01-29T17:38:29Z)
• WeDLM: Reconciling Diffusion Language Models with Standard Causal Attention for Fast Inference [44.87788417755154]
  We propose WeDLM, a diffusion decoding framework built entirely on standard causal attention.<n>We show that WeDLM preserves the quality of strong AR backbones while delivering substantial speedups.
  arXiv  Detail & Related papers  (2025-12-28T01:25:48Z)
• Memory- and Latency-Constrained Inference of Large Language Models via Adaptive Split Computing [8.705453442427585]
  Large language models (LLMs) have achieved near-human performance across diverse reasoning tasks.<n>Their deployment on resource-constrained Internet-of-Things (IoT) devices remains impractical due to massive parameter footprints and memory-intensive autoregressive decoding.<n>This work introduces the first autoregressive-aware split computing framework designed explicitly for LLM deployment on edge devices.
  arXiv  Detail & Related papers  (2025-11-06T02:55:07Z)
• Fast-dLLM v2: Efficient Block-Diffusion LLM [64.38006546510337]
  Fast-dLLM v2 is a block diffusion language model that adapts pretrained AR models into dLLMs for parallel text generation.<n>This represents a 500x reduction in training data compared to full-attention diffusion LLMs such as Dream (580B tokens)
  arXiv  Detail & Related papers  (2025-09-30T14:40:18Z)
• Fast and Fluent Diffusion Language Models via Convolutional Decoding and Rejective Fine-tuning [23.58934174168992]
  Autoregressive (AR) language models generate text one token at a time, which limits their inference speed.<n>We propose Convolutional decoding (Conv), a normalization-based method that narrows the decoding window without hard segmentation.<n>We also introduce Rejecting Rule-based Fine-Tuning (R2FT), a post-hoc training scheme that better aligns tokens at positions far from context.
  arXiv  Detail & Related papers  (2025-09-18T17:48:21Z)
• ToMA: Token Merge with Attention for Diffusion Models [8.079656935981193]
  Diffusion models excel in high-fidelity image generation but face scalability limits due to transformers' quadratic attention complexity.<n>We propose Token Merge with Attention (ToMA), an off-the-shelf method that negates token reduction for GPU-aligned efficiency.<n>ToMA reduces SDXL/Flux generation latency by 24%/23%, respectively (with DINO $Delta 0.07$), outperforming prior methods.
  arXiv  Detail & Related papers  (2025-09-13T17:35:00Z)
• Accelerating Diffusion LLMs via Adaptive Parallel Decoding [60.407727995313074]
  We introduce adaptive parallel decoding (APD), a novel method that dynamically adjusts the number of tokens sampled in parallel.<n>APD provides markedly higher throughput with minimal quality degradations on downstream benchmarks.
  arXiv  Detail & Related papers  (2025-05-31T06:10:10Z)
• Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding [51.711605076319216]
  Diffusion-based large language models (Diffusion LLMs) have shown promise for non-autoregressive text generation with parallel decoding capabilities.<n>We introduce a novel block-wise approximate KV Cache mechanism tailored for bidirectional diffusion models, enabling cache reuse with negligible performance drop.<n>We propose a confidence-aware parallel decoding strategy that selectively decodes tokens exceeding a confidence threshold, mitigating dependency violations and maintaining generation quality.
  arXiv  Detail & Related papers  (2025-05-28T17:39:15Z)
• HiRE: High Recall Approximate Top-$k$ Estimation for Efficient LLM Inference [68.59839755875252]
  HiRE comprises of two novel components: (i) a compression scheme to cheaply predict top-$k$ rows/columns with high recall, followed by full computation restricted to the predicted subset, and (ii) DA-TOP-$k$: an efficient multi-device approximate top-$k$ operator. We demonstrate that on a one billion parameter model, HiRE applied to both the softmax as well as feedforward layers, achieves almost matching pretraining and downstream accuracy, and speeds up inference latency by $1.47times$ on a single TPUv5e device.
  arXiv  Detail & Related papers  (2024-02-14T18:04:36Z)

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.