Saved in:
Bibliographic Details
Main Authors: Lin, Ke, Luo, Yiyang, Su, Zhaolong, Song, Yunya, Rao, Anyi
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.25969
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917531809218560
author Lin, Ke
Luo, Yiyang
Su, Zhaolong
Song, Yunya
Rao, Anyi
author_facet Lin, Ke
Luo, Yiyang
Su, Zhaolong
Song, Yunya
Rao, Anyi
contents Causal Transformer language models suffer from strictly sequential decoding and a quadratic per-step attention cost. While linear-time causal models and discrete diffusion models each address these weaknesses, their integration remains inherently inconsistent: diffusion requires bidirectional attention, while causal models are unidirectional. To unify these architectures, we propose $B^3D-RWKV$, a diffusion RWKV variant that integrates the model's $O(L)$ inference efficiency with parallel, bidirectional discrete-diffusion through a \emph{triplet-block layout} method. $B^3D-RWKV-7.2B$ reaches comparable accuracy on an 8-task suite versus existing models while significantly outperforming baselines in decoding throughput with an average of $\mathbf{1.6\times}$ speedup.
format Preprint
id arxiv_https___arxiv_org_abs_2605_25969
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Triplet-Block Diffusion RWKV
Lin, Ke
Luo, Yiyang
Su, Zhaolong
Song, Yunya
Rao, Anyi
Computation and Language
Causal Transformer language models suffer from strictly sequential decoding and a quadratic per-step attention cost. While linear-time causal models and discrete diffusion models each address these weaknesses, their integration remains inherently inconsistent: diffusion requires bidirectional attention, while causal models are unidirectional. To unify these architectures, we propose $B^3D-RWKV$, a diffusion RWKV variant that integrates the model's $O(L)$ inference efficiency with parallel, bidirectional discrete-diffusion through a \emph{triplet-block layout} method. $B^3D-RWKV-7.2B$ reaches comparable accuracy on an 8-task suite versus existing models while significantly outperforming baselines in decoding throughput with an average of $\mathbf{1.6\times}$ speedup.
title Triplet-Block Diffusion RWKV
topic Computation and Language
url https://arxiv.org/abs/2605.25969