Inference-Time Scaling of Diffusion Language Models with Particle Gibbs Sampling

• URL: http://arxiv.org/abs/2507.08390v2
• Date: Mon, 06 Oct 2025 05:26:50 GMT
• Title: Inference-Time Scaling of Diffusion Language Models with Particle Gibbs Sampling
• Authors: Meihua Dang, Jiaqi Han, Minkai Xu, Kai Xu, Akash Srivastava, Stefano Ermon,
• Abstract summary: We study how to steer generation toward desired rewards without retraining the models.<n>Prior methods typically resample or filter within a single denoising trajectory, optimizing rewards step-by-step without trajectory-level refinement.<n>We introduce particle Gibbs sampling for diffusion language models (PG-DLM), a novel inference-time algorithm enabling trajectory-level refinement while preserving generation perplexity.
• Score: 70.8832906871441
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Discrete diffusion models have recently emerged as strong alternatives to autoregressive language models, matching their performance through large-scale training. However, inference-time control remains relatively underexplored. In this work, we study how to steer generation toward desired rewards without retraining the models. Prior methods typically resample or filter within a single denoising trajectory, optimizing rewards step-by-step without trajectory-level refinement. We introduce particle Gibbs sampling for diffusion language models (PG-DLM), a novel inference-time algorithm enabling trajectory-level refinement while preserving generation perplexity under reward optimization. PG-DLM constructs a Markov chain over full denoising trajectories and applies a conditional sequential Monte Carlo kernel to resample them. We derive theoretical guarantees for convergence, including asymptotic consistency and variance bounds. Within this framework, we further analyze trade-offs across four key axes for inference-time scaling under fixed budgets: iterations, samples, denoising steps, and reward estimation. Our analysis shows scaling iterations achieves the best reward-perplexity trade-off. Empirically, PG-DLM consistently outperforms prior methods using MDLM and LLaDA-8B as base models across a wide range of compute budgets for reward-guided generation tasks including toxicity and sentiment control as well as linguistic acceptability.

Related papers

• Sharp Convergence Rates for Masked Diffusion Models [53.117058231393834]
  We develop a total-variation based analysis for the Euler method that overcomes limitations.<n>Our results relax assumptions on score estimation, improve parameter dependencies, and establish convergence guarantees.<n>Overall, our analysis introduces a direct TV-based error decomposition along the CTMC trajectory and a decoupling-based path-wise analysis for FHS.
  arXiv  Detail & Related papers  (2026-02-26T00:47:51Z)
• Self-Rewarding Sequential Monte Carlo for Masked Diffusion Language Models [58.946955321428845]
  This work presents self-rewarding sequential Monte Carlo (SMC)<n>Our algorithm stems from the observation that most existing MDLMs rely on a confidence-based sampling strategy.<n>We introduce the trajectory-level confidence as a self-rewarding signal for assigning particle importance weights.
  arXiv  Detail & Related papers  (2026-02-02T09:21:45Z)
• Discrete Feynman-Kac Correctors [47.62319930071118]
  We propose a framework that allows for controlling the generated distribution of discrete masked diffusion models at inference time.<n>We derive Sequential Monte Carlo (SMC) algorithms that, given a trained discrete diffusion model, control the temperature of the sampled distribution.<n>We illustrate the utility of our framework in several applications including: efficient sampling from the Boltzmann distribution of the Ising model, improving the performance of language models for code generation and amortized learning, as well as reward-tilted protein sequence generation.
  arXiv  Detail & Related papers  (2026-01-15T13:55:38Z)
• G$^2$RPO: Granular GRPO for Precise Reward in Flow Models [74.21206048155669]
  We propose a novel Granular-GRPO (G$2$RPO) framework that achieves precise and comprehensive reward assessments of sampling directions.<n>We introduce a Multi-Granularity Advantage Integration module that aggregates advantages computed at multiple diffusion scales.<n>Our G$2$RPO significantly outperforms existing flow-based GRPO baselines.
  arXiv  Detail & Related papers  (2025-10-02T12:57:12Z)
• Coefficients-Preserving Sampling for Reinforcement Learning with Flow Matching [6.238027696245818]
  Reinforcement Learning (RL) has emerged as a powerful technique for improving image and video generation in Diffusion and Flow Matching models.<n>Our investigation reveals a significant drawback to this approach: SDE-based sampling introduces pronounced noise artifacts in the generated images.<n>Our proposed method, Coefficients-Preserving Sampling (CPS) eliminates these noise artifacts.
  arXiv  Detail & Related papers  (2025-09-07T07:25:00Z)
• Self-Boost via Optimal Retraining: An Analysis via Approximate Message Passing [58.52119063742121]
  Retraining a model using its own predictions together with the original, potentially noisy labels is a well-known strategy for improving the model performance.<n>This paper addresses the question of how to optimally combine the model's predictions and the provided labels.<n>Our main contribution is the derivation of the Bayes optimal aggregator function to combine the current model's predictions and the given labels.
  arXiv  Detail & Related papers  (2025-05-21T07:16:44Z)
• Inference-Time Scaling for Flow Models via Stochastic Generation and Rollover Budget Forcing [10.542645300983878]
  We propose an inference-time scaling approach for pretrained flow models.<n>We show that SDE-based generation, particularly variance-preserving (VP) interpolant-based generation, improves particle sampling methods for inference-time scaling in flow models.
  arXiv  Detail & Related papers  (2025-03-25T06:30:45Z)
• Feynman-Kac Correctors in Diffusion: Annealing, Guidance, and Product of Experts [64.34482582690927]
  We provide an efficient and principled method for sampling from a sequence of annealed, geometric-averaged, or product distributions derived from pretrained score-based models.<n>We propose Sequential Monte Carlo (SMC) resampling algorithms that leverage inference-time scaling to improve sampling quality.
  arXiv  Detail & Related papers  (2025-03-04T17:46:51Z)
• Accelerated Diffusion Models via Speculative Sampling [89.43940130493233]
  Speculative sampling is a popular technique for accelerating inference in Large Language Models.<n>We extend speculative sampling to diffusion models, which generate samples via continuous, vector-valued Markov chains.<n>We propose various drafting strategies, including a simple and effective approach that does not require training a draft model.
  arXiv  Detail & Related papers  (2025-01-09T16:50:16Z)
• Convergence Analysis of Discrete Diffusion Model: Exact Implementation through Uniformization [17.535229185525353]
  We introduce an algorithm leveraging the uniformization of continuous Markov chains, implementing transitions on random time points. Our results align with state-of-the-art achievements for diffusion models in $mathbbRd$ and further underscore the advantages of discrete diffusion models in comparison to the $mathbbRd$ setting.
  arXiv  Detail & Related papers  (2024-02-12T22:26:52Z)
• Improved off-policy training of diffusion samplers [93.66433483772055]
  We study the problem of training diffusion models to sample from a distribution with an unnormalized density or energy function.<n>We benchmark several diffusion-structured inference methods, including simulation-based variational approaches and off-policy methods.<n>Our results shed light on the relative advantages of existing algorithms while bringing into question some claims from past work.
  arXiv  Detail & Related papers  (2024-02-07T18:51:49Z)
• Fast Sampling via Discrete Non-Markov Diffusion Models with Predetermined Transition Time [49.598085130313514]
  We propose discrete non-Markov diffusion models (DNDM), which naturally induce the predetermined transition time set.<n>This enables a training-free sampling algorithm that significantly reduces the number of function evaluations.<n>We study the transition from finite to infinite step sampling, offering new insights into bridging the gap between discrete and continuous-time processes.
  arXiv  Detail & Related papers  (2023-12-14T18:14:11Z)
• Post-training Quantization for Text-to-Image Diffusion Models with Progressive Calibration and Activation Relaxing [49.800746112114375]
  We propose a novel post-training quantization method (Progressive and Relaxing) for text-to-image diffusion models. We are the first to achieve quantization for Stable Diffusion XL while maintaining the performance.
  arXiv  Detail & Related papers  (2023-11-10T09:10:09Z)
• A Geometric Perspective on Diffusion Models [57.27857591493788]
  We inspect the ODE-based sampling of a popular variance-exploding SDE. We establish a theoretical relationship between the optimal ODE-based sampling and the classic mean-shift (mode-seeking) algorithm.
  arXiv  Detail & Related papers  (2023-05-31T15:33:16Z)
• Fast Inference in Denoising Diffusion Models via MMD Finetuning [23.779985842891705]
  We present MMD-DDM, a novel method for fast sampling of diffusion models. Our approach is based on the idea of using the Maximum Mean Discrepancy (MMD) to finetune the learned distribution with a given budget of timesteps. Our findings show that the proposed method is able to produce high-quality samples in a fraction of the time required by widely-used diffusion models.
  arXiv  Detail & Related papers  (2023-01-19T09:48:07Z)
• How Much is Enough? A Study on Diffusion Times in Score-based Generative Models [76.76860707897413]
  Current best practice advocates for a large T to ensure that the forward dynamics brings the diffusion sufficiently close to a known and simple noise distribution. We show how an auxiliary model can be used to bridge the gap between the ideal and the simulated forward dynamics, followed by a standard reverse diffusion process.
  arXiv  Detail & Related papers  (2022-06-10T15:09:46Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.