FastDiSS: Few-step Match Many-step Diffusion Language Model on Sequence-to-Sequence Generation--Full Version
This work addresses a key bottleneck for deploying diffusion models in fast inference scenarios, offering incremental improvements for sequence-to-sequence generation tasks.
The paper tackles the degradation of self-conditioning in continuous diffusion language models during few-step sampling, which leads to significant approximation gaps and poor sample quality; the proposed training framework with perturbed self-conditioning and token-level noise-awareness improves robustness and optimization, achieving up to 400x faster inference while surpassing standard models and competing with other one-step frameworks.
Self-conditioning has been central to the success of continuous diffusion language models, as it allows models to correct previous errors. Yet its ability degrades precisely in the regime where diffusion is most attractive for deployment: few-step sampling for fast inference. In this study, we show that when models only have a few denoising steps, inaccurate self-conditioning induces a substantial approximation gap; this mistake compounds across denoising steps and ultimately dominate the sample quality. To address this, we propose a novel training framework that handles these errors during learning by perturbing the self-conditioning signal to match inference noise, improving robustness to prior estimation errors. In addition, we introduce a token-level noise-awareness mechanism that prevents training from saturation, hence improving optimization. Extensive experiments across conditional generation benchmarks demonstrate that our framework surpasses standard continuous diffusion models while providing up to 400x faster inference speed, and remains competitive against other one-step diffusion frameworks.