How Efficient Are Diffusion Language Models? A Critical Examination of Efficiency Evaluation Practices
This work addresses the practical efficiency gap for researchers and practitioners considering DLMs as alternatives to autoregressive models, though it is incremental in highlighting evaluation issues rather than proposing new methods.
This paper systematically examines the efficiency of diffusion language models (DLMs) compared to autoregressive models, finding that DLMs consistently underperform in throughput despite their parallelizable decoding potential, with acceleration strategies offering only limited gains at scale.
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.