DiffuSeq-v2: Bridging Discrete and Continuous Text Spaces for Accelerated Seq2Seq Diffusion Models

Shansan Gong, Mukai Li, Jiangtao Feng, Zhiyong Wu, Lingpeng Kong

Submission Type: Regular Short Paper

Submission Track: Natural Language Generation

Submission Track 2: NLP Applications

Keywords: diffusion model, sequence to sequence, text generation

TL;DR: Accelerate continuous text diffusion model by connecting the discrete nature.

Abstract: Diffusion models have gained prominence in generating high-quality sequences of text. Nevertheless, current approaches predominantly represent discrete text within a continuous diffusion space, which incurs substantial computational overhead during training and results in slower sampling speeds. In this paper, we introduce a soft absorbing state that facilitates the diffusion model in learning to reconstruct discrete mutations based on the underlying Gaussian space, thereby enhancing its capacity to recover conditional signals. During the sampling phase, we employ state-of-the-art ODE solvers within the continuous space to expedite the sampling process. Comprehensive experimental evaluations reveal that our proposed method effectively accelerates the training convergence by 4x and generates samples of similar quality 800x faster, rendering it significantly closer to practical application. \footnote{The code is released at \url{https://github.com/Shark-NLP/DiffuSeq/tree/diffuseq-v2}.}

Submission Number: 5563