Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages

• URL: http://arxiv.org/abs/2603.12554v1
• Date: Fri, 13 Mar 2026 01:38:44 GMT
• Title: Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages
• Abstract summary: Reinforcement learning has been effective for post-training autoregressive (AR) language models, but extending these methods to diffusion language models (DLMs) is challenging due to intractable sequence-level likelihoods.<n>We formulate a diffusion-based sequence generation as a finite-horizon Markov decision process over the denoising trajectory and derive an exact, unbiased policy gradient.<n>Experiments on coding and logical reasoning benchmarks demonstrate state-of-the-art results, with strong competitive performance on mathematical reasoning.
• Score: 7.728607011076579
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Reinforcement learning (RL) has been effective for post-training autoregressive (AR) language models, but extending these methods to diffusion language models (DLMs) is challenging due to intractable sequence-level likelihoods. Existing approaches therefore rely on surrogate likelihoods or heuristic approximations, which can introduce bias and obscure the sequential structure of denoising. We formulate diffusion-based sequence generation as a finite-horizon Markov decision process over the denoising trajectory and derive an exact, unbiased policy gradient that decomposes over denoising steps and is expressed in terms of intermediate advantages, without requiring explicit evaluation of the sequence likelihood. To obtain a practical and compute-efficient estimator, we (i) select denoising steps for policy updates via an entropy-guided approximation bound, and (ii) estimate intermediate advantages using a one-step denoising reward naturally provided by the diffusion model, avoiding costly multi-step rollouts. Experiments on coding and logical reasoning benchmarks demonstrate state-of-the-art results, with strong competitive performance on mathematical reasoning, outperforming existing RL post-training approaches for DLMs. Code is available at https://github.com/vishnutez/egspo-dllm-rl.

Related papers

• dVLA-RL: Reinforcement Learning over Denoising Trajectories for Discrete Diffusion Vision-Language-Action Models [49.497309561043004]
  We propose textbfdVLA-RL, shifting the learning objective from the marginal action probability to the joint probability of the sampled generation path.<n>Our approach achieves a success rate of textbf99.7% on LIBERO.<n>It also establishes strong VLA-based results on RoboTwin 2.0 by delivering a textbf30.6% improvement over the SFT baseline.
  arXiv  Detail & Related papers  (2026-06-22T17:19:03Z)
• Uncertainty Quantification for Large Language Diffusion Models [35.87984313223733]
  Large Language Diffusion Models (LLDMs) are emerging as an alternative to autoregressive models.<n>Existing uncertainty quantification (UQ) methods assume autoregressive factorization or use expensive repeated sampling.<n>We present the first systematic study of UQ for LLDMs and propose lightweight, zero-shot uncertainty signals.
  arXiv  Detail & Related papers  (2026-05-14T08:39:56Z)
• Self-Distilled Trajectory-Aware Boltzmann Modeling: Bridging the Training-Inference Discrepancy in Diffusion Language Models [65.89572755202245]
  Diffusion Language Models (DLMs) offer stronger global awareness and highly parallel generation.<n>Post-training DLMs with standard Negative Evidence Lower Bound (NELBO)-based supervised fine-tuning remains inefficient.<n>We propose a self-distilled trajectory-based post-training framework that aligns training with the easy-to-hard structure of inference.
  arXiv  Detail & Related papers  (2026-05-12T09:39:06Z)
• Step-level Denoising-time Diffusion Alignment with Multiple Objectives [26.715620800855962]
  Reinforcement learning (RL) has emerged as a powerful tool for aligning diffusion models with human preferences.<n>In practice, however, human preferences are inherently pluralistic, and aligned models must balance multiple downstream objectives.<n>We propose Multi-objective Step-level Denoising-time Diffusion Alignment (MSDDA), a retraining-free framework for aligning diffusion models with multiple objectives.
  arXiv  Detail & Related papers  (2026-04-15T19:52:05Z)
• DiffuReason: Bridging Latent Reasoning and Generative Refinement for Sequential Recommendation [20.756497463882763]
  We propose DiffuReason, a unified "Think-then-Diffuse" framework for sequential recommendation.<n>It integrates multi-step Thinking Tokens for latent reasoning, diffusion-based refinement for denoising intermediate representations, and end-to-end Group Relative Policy Optimization.<n>Experiments on four benchmarks demonstrate that DiffuReason consistently improves diverse backbone architectures.
  arXiv  Detail & Related papers  (2026-02-10T12:55:30Z)
• Principled RL for Diffusion LLMs Emerges from a Sequence-Level Perspective [85.06838178922791]
  Reinforcement Learning (RL) has proven highly effective for autoregressive language models.<n>But adapting these methods to diffusion large language models (dLLMs) presents fundamental challenges.<n>We propose a principled RL framework that treats entire sequence generation as a single action and uses the ELBO as a tractable sequence-level likelihood proxy.
  arXiv  Detail & Related papers  (2025-12-03T13:05:32Z)
• MRO: Enhancing Reasoning in Diffusion Language Models via Multi-Reward Optimization [66.82303841930752]
  diffusion language models (DLMs) have presented a promising alternative to traditional autoregressive large language models (LLMs)<n>DLMs still lag behind LLMs in reasoning performance, especially as the number of denoising steps decreases.<n>We propose a Multi-Reward Optimization (MRO) approach, which encourages DLMs to consider the token correlation during the denoising process.
  arXiv  Detail & Related papers  (2025-10-24T13:57:59Z)
• Inference-Time Scaling of Diffusion Language Models with Particle Gibbs Sampling [70.8832906871441]
  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.
  arXiv  Detail & Related papers  (2025-07-11T08:00:47Z)
• Training-free Diffusion Model Alignment with Sampling Demons [15.400553977713914]
  We propose an optimization approach, dubbed Demon, to guide the denoising process at inference time without backpropagation through reward functions or model retraining.<n>Our approach works by controlling noise distribution in denoising steps to concentrate density on regions corresponding to high rewards through optimization.<n>Our experiments show that the proposed approach significantly improves the average aesthetics scores text-to-image generation.
  arXiv  Detail & Related papers  (2024-10-08T07:33:49Z)
• Unconditional Truthfulness: Learning Unconditional Uncertainty of Large Language Models [104.55763564037831]
  We train a regression model that leverages attention maps, probabilities on the current generation step, and recurrently computed uncertainty scores from previously generated tokens.<n>Our evaluation shows that the proposed method is highly effective for selective generation, achieving substantial improvements over rivaling unsupervised and supervised approaches.
  arXiv  Detail & Related papers  (2024-08-20T09:42:26Z)
• Amortizing intractable inference in large language models [56.92471123778389]
  We use amortized Bayesian inference to sample from intractable posterior distributions. We empirically demonstrate that this distribution-matching paradigm of LLM fine-tuning can serve as an effective alternative to maximum-likelihood training. As an important application, we interpret chain-of-thought reasoning as a latent variable modeling problem.
  arXiv  Detail & Related papers  (2023-10-06T16:36:08Z)
• Observation-Guided Diffusion Probabilistic Models [41.749374023639156]
  We propose a novel diffusion-based image generation method called the observation-guided diffusion probabilistic model (OGDM) Our approach reestablishes the training objective by integrating the guidance of the observation process with the Markov chain. We demonstrate the effectiveness of our training algorithm using diverse inference techniques on strong diffusion model baselines.
  arXiv  Detail & Related papers  (2023-10-06T06:29:06Z)

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.