Paper 2512.13586
ReFusion: A Diffusion Large Language Model with Parallel Autoregressive Decoding
- Published
- Dec 2025
- Research lab
- Independent
- Citations
- 14
- GitHub
- 63 stars
01 In brief
Summary
ReFusion is a masked diffusion language model that integrates sequence reorganization into a causal attention framework to overcome the limitations of standard masked diffusion models (MDMs), which suffer from high computational overhead due to the lack of Key-Value (KV) caching and incoherent generation from intractable token combination learning.
ReFusion partitions sequences into slots, using diffusion-based selection for inter-slot parallel decoding and autoregressive infilling for intra-slot generation, reordering generated slots ahead of remaining masks to enable full KV cache reuse and reduce learning complexity to a slot-level permutation space.
Experiments on seven benchmarks show ReFusion surpasses prior MDMs with a 34% performance gain and over 18x speedup, and bridges the performance gap to autoregressive models (ARMs) while maintaining a 2.33x average speedup, outperforming Qwen3-8B on GSM8K and MBPP.
Controlled comparisons confirm the gains stem from architectural innovations, not data or initialization advantages.
Hyperparameter analysis reveals a robust operating region where ReFusion exceeds ARM baselines in both performance and throughput.
02 From the paper
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 \textsc{ReFusion}, a novel masked diffusion model that integrates sequence reorganization into the causal attention framework. By elevating parallel decoding from the token level to a higher slot level, \textsc{ReFusion} interleaves inter-slot diffusion-based selection with intra-slot autoregressive infilling, while reordering newly generated slots ahead of the remaining masks after each iteration. Consequently, this design simultaneously unlocks full KV cache reuse and reduces learning complexity from an intractable token combination space to a manageable slot-level permutation space. Extensive experiments on seven diverse benchmarks show that \textsc{ReFusion} not only overwhelmingly surpasses prior MDMs with a 34\% performance gain 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.