GFlowRL: Scaling Distribution-Matching RL to Large Language Models

• URL: http://arxiv.org/abs/2607.13394v1
• Date: Wed, 15 Jul 2026 02:43:16 GMT
• Title: GFlowRL: Scaling Distribution-Matching RL to Large Language Models
• Abstract summary: Generative Flow Networks (GFlowNets) offer a promising alternative to reward-maximizing reinforcement learning (RL) for large reasoning models.<n>Recent work shows promise on math and code, but scaling GFlowNet-style RL to modern post-training pipelines remains difficult.<n>We propose GFlowRL, a streamlined GFlowNet-style RL algorithm that removes the auxiliary partition network entirely.
• Score: 26.27023085358537
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Generative Flow Networks (GFlowNets) offer a promising alternative to reward-maximizing reinforcement learning (RL) for large reasoning models, encouraging diverse reasoning paths by matching reward distributions rather than collapsing to dominant modes. Recent work shows promise on math and code, but scaling GFlowNet-style RL to modern post-training pipelines remains difficult: as model size, rollout horizon, reward noise, and distributed-systems complexity grow together, a learned prompt-conditional partition function becomes a source of gradient instability and engineering overhead rather than a useful normalizer. Through systematic analysis, we find that the learned partition function, previously treated as essential, can be replaced by an in-batch Monte Carlo estimate computed from the rollout group already required for training. We propose GFlowRL, a streamlined GFlowNet-style RL algorithm that removes the auxiliary partition network entirely while preserving the reward-distribution-matching objective, completed by two stabilizers: importance-sampling correction for rollout/trainer drift and asymmetric flow-gap clipping for outlier residuals. GFlowRL exceeds all counterparts on math, code, and adversarial red-teaming benchmarks, reaching a Codeforces rating of 2048 at the 14B scale (within 25 Elo of o3-mini) and attaining the highest average ASR@1 on AdvBench and HarmBench, outperforming the previous SOTA multi-turn attacker in a regime where FlowRL, a prior GFlowNet-style method, diverges. The same recipe transfers to all evaluated MoE configurations up to 235B parameters, where FlowRL again fails to converge. To our knowledge, GFlowRL is the first GFlowNet-style RL algorithm to scale stably across both dense and sparse architectures. Code will be at: https://github.com/microsoft/gflowrl

Related papers

• Libra: Efficient Resource Management for Agentic RL Post-Training [11.701871372256205]
  Reinforcement learning (RL) has emerged as a standard post-training paradigm for shaping large language models (LLMs) into capable agents.<n>In agentic RL, the rollout stage generates trajectories while invoking tools, producing long-tailed and non-stationary workloads.<n>We present Libra, a resource management system to address both challenges via two core mechanisms.
  arXiv  Detail & Related papers  (2026-06-02T03:09:13Z)
• Boundary-Guided Policy Optimization for Memory-efficient RL of Diffusion Large Language Models [53.339700196282905]
  A key challenge in applying reinforcement learning to large language models (dLLMs) is the intractability of their likelihood functions.<n>We propose a memory-efficient RL algorithm that maximizes a specially constructed lower bound of the ELBO-based objective.<n> Experiments show that BGPO significantly outperforms previous RL algorithms for dLLMs in math problem solving, code generation, and planning tasks.
  arXiv  Detail & Related papers  (2025-10-13T17:47:50Z)
• Efficient Regression-Based Training of Normalizing Flows for Boltzmann Generators [85.25962679349551]
  Boltzmann Generators (BGs) offer efficient sampling and likelihoods, but their training via maximum likelihood is often unstable and computationally challenging.<n>We propose Regression Training of Normalizing Flows (RegFlow), a novel and scalable-based training objective that bypasses the numerical instability and computational challenge of conventional maximum likelihood training.
  arXiv  Detail & Related papers  (2025-06-01T20:32:27Z)
• Flow-GRPO: Training Flow Matching Models via Online RL [80.62659379624867]
  We propose Flow-GRPO, the first method to integrate online policy reinforcement learning into flow matching models.<n>Our approach uses two key strategies: (1) an ODE-to-SDE conversion that transforms a deterministic Ordinary Differential Equation into an equivalent Differential Equation (SDE) that matches the original model's marginal distribution at all timesteps; and (2) a Denoising Reduction strategy that reduces training denoising steps while retaining the original number of inference steps.
  arXiv  Detail & Related papers  (2025-05-08T17:58:45Z)
• Optimizing Backward Policies in GFlowNets via Trajectory Likelihood Maximization [4.158255103170876]
  GFlowNets are a family of generative models that learn to sample objects proportional to a given reward function.<n>Recent results show a close relationship between GFlowNet training and entropy-regularized reinforcement learning problems.<n>We introduce a simple backward policy optimization algorithm that involves direct sequentially of the value function in an entropy-regularized Markov Decision Process.
  arXiv  Detail & Related papers  (2024-10-20T19:12:14Z)
• Random Policy Evaluation Uncovers Policies of Generative Flow Networks [12.294107455811496]
  Generative Flow Network (GFlowNet) is a probabilistic framework in which an agent learns a policy and flow functions.<n>In this paper, we reveal a fundamental connection between GFlowNets and one RL's most basic components -- policy evaluation.<n>We introduce a rectified random policy evaluation algorithm, which achieves the same reward-matching effect as GFlowNets.
  arXiv  Detail & Related papers  (2024-06-04T11:11:53Z)
• Generative Flow Networks as Entropy-Regularized RL [4.857649518812728]
  generative flow networks (GFlowNets) are a method of training a policy to sample compositional objects with proportional probabilities to a given reward via a sequence of actions. We demonstrate how the task of learning a generative flow network can be efficiently as an entropy-regularized reinforcement learning problem. Contrary to previously reported results, we show that entropic RL approaches can be competitive against established GFlowNet training methods.
  arXiv  Detail & Related papers  (2023-10-19T17:31:40Z)
• Towards Understanding and Improving GFlowNet Training [71.85707593318297]
  We introduce an efficient evaluation strategy to compare the learned sampling distribution to the target reward distribution. We propose prioritized replay training of high-reward $x$, relative edge flow policy parametrization, and a novel guided trajectory balance objective.
  arXiv  Detail & Related papers  (2023-05-11T22:50:41Z)
• Better Training of GFlowNets with Local Credit and Incomplete Trajectories [81.14310509871935]
  We consider the case where the energy function can be applied not just to terminal states but also to intermediate states. This is for example achieved when the energy function is additive, with terms available along the trajectory. This enables a training objective that can be applied to update parameters even with incomplete trajectories.
  arXiv  Detail & Related papers  (2023-02-03T12:19:42Z)
• Learning GFlowNets from partial episodes for improved convergence and stability [56.99229746004125]
  Generative flow networks (GFlowNets) are algorithms for training a sequential sampler of discrete objects under an unnormalized target density. Existing training objectives for GFlowNets are either local to states or transitions, or propagate a reward signal over an entire sampling trajectory. Inspired by the TD($lambda$) algorithm in reinforcement learning, we introduce subtrajectory balance or SubTB($lambda$), a GFlowNet training objective that can learn from partial action subsequences of varying lengths.
  arXiv  Detail & Related papers  (2022-09-26T15:44:24Z)

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.