Loopholing Discrete Diffusion: Deterministic Bypass of the Sampling Wall
This addresses a key bottleneck for non-autoregressive text generation, offering a scalable solution with concrete performance gains.
The paper tackles the sampling wall problem in discrete diffusion models where categorical sampling collapses distributional information, by introducing Loopholing, a deterministic latent pathway that preserves this information. The resulting Loopholing Discrete Diffusion Models (LDDMs) reduce generative perplexity by up to 61% over prior baselines, close the gap with autoregressive models, and improve performance on arithmetic reasoning benchmarks.
Discrete diffusion models offer a promising alternative to autoregressive generation through parallel decoding, but they suffer from a sampling wall: once categorical sampling occurs, rich distributional information collapses into one-hot vectors and cannot be propagated across steps, forcing subsequent steps to operate with limited information. To mitigate this problem, we introduce Loopholing, a novel and simple mechanism that preserves this information via a deterministic latent pathway, leading to Loopholing Discrete Diffusion Models (LDDMs). Trained efficiently with a self-conditioning strategy, LDDMs achieve substantial gains-reducing generative perplexity by up to 61% over prior baselines, closing (and in some cases surpassing) the gap with autoregressive models, and producing more coherent text. Applied to reasoning tasks, LDDMs also improve performance on arithmetic benchmarks such as Countdown and Game of 24. These results also indicate that loopholing mitigates idle steps and oscillations, providing a scalable path toward high-quality non-autoregressive text generation.