The paper introduces adaptive second‑order solvers for diffusion models, employing a proportional‑integral (PI) step‑size controller based on a diffusion noise‑normalised error estimator. Unlike existing adaptive methods that react only to the current error, the PI solver incorporates past error information, producing smoother step adaptations. The authors also demonstrate that per‑sample trajectories share structure, allowing aggregation into a fixed schedule that largely preserves adaptive benefits. Experiments on natural‑image and language datasets show that the fixed schedule outperforms the commonly used EDM schedule for image sampling with the stochastic Heun sampler, and that the PI adaptive solver achieves lower FID than most baselines, though it does not surpass the EDM‑churn sampler at low neural‑network‑evaluation counts. In language diffusion, the PI solver improves perplexity over EDM and entropy schedules at low‑to‑medium NFE, albeit with reduced token entropy. The study concludes that per‑sample adaptivity is highly beneficial for simple 1D toy problems but only marginally useful for image and language data, where an average schedule can sometimes outperform the PI‑adaptive approach.
Diffusion models rely on numerical solvers requiring time-discretization, which has a large influence on the tradeoff between sampling cost and quality. However, the computational difficulty of the re...
arXiv:2602.12624v2 Announce Type: replace Abstract: Diffusion-based generative models have achieved remarkable performance across various domains, yet their practical deployment is often limited by h...
GeoSPRINT is a training‑free framework that constructs non‑uniform sampling schedules for diffusion model inference by detecting geometrically redundant steps in denoising trajectories. It uses a hyperplanarity test in latent space, implemented via QR factorization, to allocate more steps to high‑curvature regions, and introduces the trajectory projection score α_traj as a model‑free diagnostic for flow quality. Across CIFAR‑10, LSUN Church, and Stable Diffusion v1.5, GeoSPRINT consistently outperforms uniform DDIM schedules at matched NFE budgets, improving FID scores by up to 1.93 points.
Forward-Free Diffusion Language Models with BPTT-Free Looped Refinement (FReDA) removes the need for a hand‑designed forward process in diffusion language modeling by treating model‑generated drafts as implicit intermediate states and refining them recursively. The approach detaches earlier refinement passes, backpropagating only through the final pass, and supports both self‑refinement and Best‑of‑N candidate selection. In sub‑8B experiments, FReDA‑4B surpasses larger diffusion baselines on reasoning and coding tasks, achieving up to 15% absolute gains and a 1.5‑1.8× speedup while scaling well with additional refinement steps.
arXiv:2607.02137v3 Announce Type: replace-cross Abstract: We study timestep allocation for score-based diffusion sampling, where a learned reverse-time dynamics is discretized on a finite grid. Unifo...
arXiv:2605. 00161v2 Announce Type: replace Abstract: Diffusion language models (DLMs) are an attractive alternative to autoregressive models because they promise sublinear-time, parallel generation, yet practical gains remain elusive as high-quality samples still demand hundreds of refinement steps.
arXiv:2507. 08390v5 Announce Type: replace Abstract: Discrete diffusion models have recently emerged as strong alternatives to autoregressive language models, matching their performance through large-scale training.
arXiv:2607. 00773v1 Announce Type: new Abstract: Discrete diffusion models are widely used for learning and generating discrete distributions.
arXiv:2602. 15008v2 Announce Type: replace Abstract: Diffusion models over discrete spaces have recently shown striking empirical success, yet their theoretical foundations remain incomplete.
arXiv:2607. 20540v1 Announce Type: cross Abstract: How should a diffusion model decide which noise levels to train on, and how much?
arXiv:2609.37974v1 Announce Type: cross Abstract: Masked diffusion models (MDMs) generate text by unmasking several tokens per step, but they are trained and sampled under different conditions. The m...
The paper introduces the first controlled benchmark for optimizers in discrete diffusion models, evaluating seven optimizers (AdamW, Lion, Muon, SOAP, MARS, MARS‑M, Schedule‑Free) across four diffusion formulations: masked diffusion on text8, uniform diffusion on QM9 and LM1B, and Gaussian diffusion on CelebA‑64. Each optimizer undergoes the same search protocol and is retrained with full budget and multiple seeds, revealing that AdamW, while strong, is not universally optimal and that optimizers validated on autoregressive language models (Muon, MARS‑M, SOAP) can outperform tuned AdamW on certain tasks.