How Efficient Are Diffusion Language Models? A Critical Examination of Efficiency Evaluation Practices

TL;DR

This paper critically examines the efficiency of diffusion language models (DLMs), revealing that they underperform autoregressive models in speed and highlighting the need for better evaluation and acceleration techniques.

Contribution

It provides a systematic analysis of DLM efficiency, identifying flaws in prior evaluation practices and assessing the effectiveness of acceleration strategies.

Findings

AR models have higher throughput than DLMs

Acceleration techniques are effective only at small batch sizes

Robust evaluation methods are essential for fair comparison

Abstract

Diffusion language models (DLMs) have emerged as a promising alternative to the long-dominant autoregressive (AR) paradigm, offering a parallelable decoding process that could yield greater efficiency. Yet, in practice, current open-source DLMs often underperform their AR counterparts in speed, limiting their real-world utility. This work presents a systematic study of DLM efficiency, identifying key issues in prior evaluation methods. Through empirical benchmarking and a theoretical analysis, we demonstrate that AR models generally achieve higher throughput, while DLMs consistently lag. We also investigate acceleration strategies, finding that techniques like dual cache and parallel decoding mainly offer gains at small batch sizes, with their benefits diminishing upon scaling. Our findings underscore the necessity of robust evaluation methods and improved acceleration strategies to advance research on DLMs.