A Survey on Diffusion Language Models

TL;DR

Diffusion Language Models are an emerging class of models that generate text in parallel, offering advantages in speed and bidirectional context understanding, with recent advancements making them competitive with traditional autoregressive models.

Contribution

This survey provides a comprehensive overview, taxonomy, and analysis of DLMs, including their principles, techniques, inference strategies, and multimodal extensions, highlighting current challenges and future directions.

Findings

DLMs achieve several-fold speed-up over autoregressive models.

Recent DLMs show performance comparable to autoregressive counterparts.

Advancements include improved decoding, caching, and multimodal capabilities.

Abstract

Diffusion Language Models (DLMs) are rapidly emerging as a powerful and promising alternative to the dominant autoregressive (AR) paradigm. By generating tokens in parallel through an iterative denoising process, DLMs possess inherent advantages in reducing inference latency and capturing bidirectional context, thereby enabling fine-grained control over the generation process. While achieving a several-fold speed-up, recent advancements have allowed DLMs to show performance comparable to their autoregressive counterparts, making them a compelling choice for various natural language processing tasks. In this survey, we provide a holistic overview of the current DLM landscape. We trace its evolution and relationship with other paradigms, such as autoregressive and masked language models, and cover both foundational principles and state-of-the-art models. Our work offers an up-to-date, comprehensive taxonomy and an in-depth analysis of current techniques, from pre-training strategies to advanced post-training methods. Another contribution of this survey is a thorough review of DLM inference strategies and optimizations, including improvements in decoding parallelism, caching mechanisms, and generation quality. We also highlight the latest approaches to multimodal extensions of DLMs and delineate their applications across various practical scenarios. Furthermore, our discussion addresses the limitations and challenges of DLMs, including efficiency, long-sequence handling, and infrastructure requirements, while outlining future research directions to sustain progress in this rapidly evolving field. Project GitHub is available at https://github.com/VILA-Lab/Awesome-DLMs.