aiqbench analysis
This paper addresses the high computational cost of reinforcement learning (RL) for diffusion models by adapting RL objectives to efficient few-step samplers. The authors systematically transform, decompose, and adapt existing methods like FlowGRPO and DiffusionNFT to recover instantaneous velocity estimates from interval-based predictions. Their experiments demonstrate that reward-gradient-based approaches remain most effective when differentiable rewards are affordable, offering a modular framework for reducing training expenses without sacrificing performance.
FewStepRL: Fine-tuning Few-step Diffusion Models with Reinforcement Learning
arXiv:2611.08240v1 Announce Type: new Abstract: Reinforcement learning (RL) has become a standard post-training method for diffusion models, but most RL objectives roll out the current policy at every iteration, and with a many-step sampler these rollouts typically dominate the cost of training. Few-step models cut this cost, yet they model an average velocity over an interval, whereas existing RL objectives are built for an instantaneous velocity with a many-step sampler. We adapt RL objectives to few-step models systematically in three phases, namely a transformation, a decomposition and an adaptation. The transformation to $x$-space expresses each objective we consider through a common policy prediction, at the clean end by default and at an explicitly marked level for transition or flow-map objectives, so that objectives stated over densities, velocities or rewards become comparable. The decomposition writes the gradient of each of these objectives as the sum of a self-regression toward a reference and a reward augmentation, so that a method reduces, at the gradient level, to a choice of anchors and weights. The adaptation computes each anchor from a few-step model at the end of its transport or along its tangent, with the instantaneous velocity recovered from the network itself or from an auxiliary network. This yields few-step versions of FlowGRPO, AWM, DiffusionNFT, DiffusionOPSD and ReFL in one modular codebase. In our experiments, reward-gradient-based methods are the most effective when the reward is differentiable and its gradient is affordable; otherwise DiffusionNFT-style methods generally reach high rewards in relatively few iterations despite a sometimes higher per-iteration cost. Extra inference steps show no systematic aggregate benefit in our experiments, although selected outputs visibly improve with two to four times as many sampling steps as in training.
Abstract:Reinforcement learning (RL) has become a standard post-training method for diffusion models, but most RL objectives roll out the current policy at every iteration, and with a many-step sampler these rollouts typically dominate the cost of training. Few-step models cut this cost, yet they model an average velocity over an interval, whereas existing RL objectives are built for an instantaneous velocity with a many-step sampler. We adapt RL objectives to few-step models systematically in three phases, namely a transformation, a decomposition and an adaptation. The transformation to $x$-space expresses each objective we consider through a common policy prediction, at the clean end by default and at an explicitly marked level for transition or flow-map objectives, so that objectives stated over densities, velocities or rewards become comparable. The decomposition writes the gradient of each of these objectives as the sum of a self-regression toward a reference and a reward augmentation, so that a method reduces, at the gradient level, to a choice of anchors and weights. The adaptation computes each anchor from a few-step model at the end of its transport or along its tangent, with the instantaneous velocity recovered from the network itself or from an auxiliary network. This yields few-step versions of FlowGRPO, AWM, DiffusionNFT, DiffusionOPSD and ReFL in one modular codebase. In our experiments, reward-gradient-based methods are the most effective when the reward is differentiable and its gradient is affordable; otherwise DiffusionNFT-style methods generally reach high rewards in relatively few iterations despite a sometimes higher per-iteration cost. Extra inference steps show no systematic aggregate benefit in our experiments, although selected outputs visibly improve with two to four times as many sampling steps as in training.
Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2611.08240 [cs.LG]
(or arXiv:2611.08240v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2611.08240 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 377 words
Related coverage
TimeRFT: Stimulating Generalizable Time Series Forecasting for TSFMs via Reinforcement Finetuning
arXiv · Aug 3, 2026
Apertus 1.5: From Open Text Models to Open Multimodal Reasoners
arXiv · Oct 14, 2026
Kalman Meets Curriculum: Efficient Dynamic Prompt Selection for Adaptive RL Finetuning
arXiv · Jul 31, 2026
Eliciting Self-Verification in Multimodal Reasoning Agents with Reinforcement Learning
arXiv · Sep 9, 2026