RL Fine-Tuning

RL: Reinforcement Learning

Follow this topic Ask about this topic

Momentum

18 papers in the last four weeks, up 260% on the four weeks before. 0.2% of all new papers.

Jul 13Week of Sep 28

Latest papers 162

May 8, 2026cs.LG

DUET: Optimize Token-Budget Allocation for Reinforcement Learning with Verifiable Rewards

Reinforcement learning with verifiable rewards (RLVR) generates hundreds of thousands of tokens per training step, with rollout generation dominating the computational cost. The overall token budget can be controlled along two main dimensions: (i) deciding which prompts to allocate rollouts to, and (ii) deciding how long each rollout should be. Prior work has generally controlled only one of these dimensions at a time. We show that jointly tuning both decisions under a shared compute budget improves both reasoning quality and wall-clock training time. We instantiate this view as \textbf{DU}al-controlled tok\textbf{E}n alloca\textbf{T}ion (DUET), a computationally efficient layer over GRPO that uses a lightweight pre-rollout surrogate of prompt informativeness to set how many rollouts each prompt receives, and a marker-gated abort rule with importance reweighting to set when to stop them. On Qwen3-1.7B trained on MATH, DUET outperforms full-budget GRPO and the other three budget-aware baseline methods. DUET's advantage further generalizes to other benchmarks across math and coding, and is on par with the best baseline on the scientific Q&A domain, while also achieving a

1.62\times

wall-clock speedup. More notably, using only 50% of the token budget, DUET still outperforms all baseline methods at their full budget, achieving an even higher

2.51\times

speedup over full-budget GRPO. We verify the high performance of DUET on other backbone LLMs, including Qwen3-4B and Llama-3.2-3B-Instruct. Notably, the gap between DUET and the strongest baseline \emph{widens} as the budget tightens, contrary to the usual pattern in which efficient methods trade off quality as compute decreases. More broadly, these results suggest that DUET budget-aware control strategies are valuable not only for accelerating training, but also for improving the quality of the learning signal.

Haoyu Hu, Xuandong Zhao, Xuhai "Orson'' Xu +1

Cornell University · University of California, Berkeley · Columbia University

Reinforcement Learning Group Relative Policy Optimization Reinforcement Learning with Verifiable Rewards RL Fine-Tuning

May 8, 2026cs.AI

On Distinguishing Capability Elicitation from Capability Creation in Post-Training: A Free-Energy Perspective

Debates about large language model post-training often treat supervised fine-tuning (SFT) as imitation and reinforcement learning (RL) as discovery. But this distinction is too coarse. What matters is whether a training procedure increases the probability of behaviors the pretrained model could already produce, or whether it changes what the model can practically reach. We argue that post-training research should distinguish between capability elicitation and capability creation. We make this distinction operational by introducing the notion of accessible support: the set of behaviors that a model can practically produce under finite budgets. Post-training that reweights behaviors within this support is capability elicitation; whereas changing the support itself corresponds to capability creation. We develop this argument through a free-energy view of post-training. SFT and RL can both be seen as reweighting a pretrained reference distribution, only with different external signals. Demonstration signals define low-energy behavior for SFT, and reward signals define low-energy behavior for RL. When the update remains close to the base model, the main effect is local reweighting, not capability creation. Within this framework, the central question is no longer whether post-training is framed as SFT or RL, but whether it reweights behaviors already within reach, or instead expands the model's reachable behavioral space through search, interaction, tool use, or the incorporation of new information.

Yuhao Li, Shengchao Liu

Department of Computer Science and Engineering The Chinese University of Hong Kong

Supervised Fine-Tuning Free Energy Principle LLM Post-Training RL Fine-Tuning

May 8, 2026cs.AI

LiteGUI: Lightweight GUI Agents via Multi-Solution Guided Distillation and Dual-Level Reinforcement Learning

We present LiteGUI, a new framework for building lightweight GUI agents. GUI interaction poses unique challenges due to the long-horizon nature of complex tasks and the existence of multiple valid interaction paths, which are difficult for lightweight GUI agents to handle effectively. To address these challenges, LiteGUI introduces a two-stage post-training framework operating on logged GUI states. First, we propose Guided On-Policy Distillation, which provides training-time privileged teacher guidance at each GUI state by selecting the most-matched valid action from human-verified multi-solution action annotations, while leaving the student's on-policy rollout unchanged. Second, we develop Multi-Solution Dual-Level GRPO, which combines action-level supervision with history-conditioned, per-state planning-quality supervision while accounting for multiple valid actions at each logged GUI state. Together, these stages support multi-step GUI tasks without requiring the student to reproduce fixed demonstration trajectories. We further develop a scalable data generation pipeline and a corresponding multi-solution dataset, Lite-Dataset, to support the training and evaluation of GUI agents, addressing a gap in the existing literature. Extensive experiments across multiple GUI benchmarks demonstrate that LiteGUI substantially improves the performance of lightweight GUI agents over existing training paradigms, including SFT, on-policy distillation and GRPO, and achieves competitive or superior performance compared with substantially larger models.

Yubin Wu, Zicheng Cai, Liping Ning +4

Moore Threads AI

RL for GUI Agents GUI Agents Language Model Distillation LLM Agent Training RL Fine-Tuning

May 8, 2026cs.CL

Gradient-Based LoRA Rank Allocation Under GRPO: An Empirical Study

Adaptive rank allocation for LoRA, allocating more parameters to important layers and fewer to unimportant ones, consistently improves efficiency under supervised fine-tuning (SFT). We investigate whether this success transfers to reinforcement learning, specifically Group Relative Policy Optimization (GRPO). Using gradient-magnitude profiling on Qwen 2.5 1.5B with GSM8K, we find that it does not: proportional rank allocation degrades accuracy by 4.5 points compared to uniform allocation (70.0% vs. 74.5%), despite using identical parameter budgets. We identify two mechanisms behind this failure. First, the gradient landscape under GRPO is fundamentally flatter than under SFT, the max-to-min layer importance ratio is only 2.17x, compared to >10x reported in SFT literature. All layers carry meaningful gradient signal; none are truly idle. Second, we discover a gradient amplification effect: non-uniform allocation widens the importance spread from 2.17x to 3.00x, creating a positive feedback loop where high-rank layers absorb more gradient while low-rank layers are progressively silenced. Our results suggest that gradient importance does not predict capacity requirements under RL, and that naive transfer of SFT-era rank allocation to alignment training should be avoided.

Yash Ganpat Sawant

1Independent Researcher.

Group Relative Policy Optimization Low-Rank Adaptation Adaptive Rank Allocation RL Fine-Tuning

May 7, 2026cs.LG

Skip What You Can Predict: Predictive Repositioning for Policy Optimization for Efficient LLM Training

Reinforcement learning with verifiable rewards (RLVR) can improve the reasoning ability of large language models, but repeatedly updating a policy on the same rollout batch is expensive: every additional update requires another backward pass, and multi-step methods pay for all intermediate optimization steps. We introduce Predictive Repositioning for Policy Optimization (PrePO), which uses two observed optimizer transitions to estimate a farther point along the same-batch optimization trajectory, moves partway toward that point, evaluates the original objective there, and applies a corrective update. This gives PrePO a fixed active cost that does not grow with the virtual optimization depth. Our analysis gives finite-horizon error bounds for AdamW and Muon and shows that sufficiently accurate endpoint estimates preserve the usual descent and convergence behavior of smooth gradient descent. In RLVR experiments, PrePO reaches matched performance targets in fewer optimization steps and lower wall-clock time than the corresponding baselines. We further evaluate the same update mechanism in supervised fine-tuning on a different dataset, showing that its use is not restricted to the original RLVR setting. Together, these results suggest that PrePO provides a practical mechanism for approximating repeated same-batch optimization while avoiding the cost of explicitly executing every intermediate update.

Ismam Nur Swapnil, Aranya Saha, Tasneea Zahra +3

Bangladesh University of Engineering and Technology · University of Maryland, College Park · Illinois Institute of Technology +1

RL for Language Model Reasoning Reinforcement Learning with Verifiable Rewards Policy Gradient Methods RL Fine-Tuning

May 6, 2026cs.LG

Adaptive Policy Selection and Fine-Tuning under Interaction Budgets for Offline-to-Online Reinforcement Learning

In offline-to-online reinforcement learning (O2O-RL), policies are first safely trained offline using previously collected datasets and then further fine-tuned for tasks via limited online interactions. In a typical O2O-RL pipeline, candidate policies trained with offline RL are evaluated via either off-policy evaluation (OPE) or online evaluation (OE). The policy with the highest estimated value is then deployed and continually fine-tuned. However, this setup has two main issues. First, OPE can be unreliable, making it risky to deploy a policy based solely on those estimates, whereas OE may identify a viable policy with substantial online interaction, which could have been used for fine-tuning. Second--and more importantly--it is also often not possible to determine a priori whether a pretrained policy will improve with post-deployment fine-tuning, especially in non-stationary environments. As a result, procedures committing to a single deployed policy are impractical in many real-world settings. Moreover, a naive remedy that exhaustively fine-tunes all candidates would violate interaction budget constraints and is likewise infeasible. In this paper, we propose a novel adaptive approach for policy selection and fine-tuning under online interaction budgets in O2O-RL. Following the standard pipeline, we first train a set of candidate policies with different offline RL algorithms and hyperparameters; we then perform OPE to obtain initial performance estimates. We next adaptively select and fine-tune the policies based on their predicted performance via an upper-confidence-bound approach thereby making efficient use of online interactions. We demonstrate that our approach improves upon O2O-RL baselines with various benchmarks.

Alper Kamil Bozkurt, Xiaoan Xu, Shangtong Zhang +2

Virginia Commonwealth University, Richmond, VA · Duke University, Durham, NC · University of Virginia, Charlottesville, VA

Sequential Decision Making RL Fine-Tuning Sample-Efficient RL Offline-To-Online RL

May 6, 2026cs.LG

Reinforcement Learning for Compositional Generalization with Outcome-Level Optimization

Compositional generalization refers to correctly interpret novel combinations of known primitives, which remains a major challenge. Existing approaches often rely on supervised fine-tuning, which encourages models to imitate target outputs. This token-level training paradigm fails to capture the global compositional structure required for generalizing to unseen combinations. In this work, we investigate whether compositional generalization can instead be improved through outcome-level reinforcement learning. We adopt Group Relative Policy Optimization to optimize models based on feedback on their final outputs. Within this framework, we explore both a simple binary outcome reward and a composite reward that provides additional composition feedback. Experiments on multiple compositional benchmarks show that reinforcement learning improves compositional generalization compared to supervised fine-tuning. Further analysis reveals that supervised models tend to overfit frequent training compositions, whereas reinforcement learning improves compositional generalization by reshaping the output distribution, particularly for more complex composition types.

Xiyan Fu, Wei Liu

Nanyang Technological University · Independent Researcher

Reinforcement Learning RL for Generative Models RL Fine-Tuning Compositional Generalization

May 6, 2026cs.CL

Every Step Counts: Step-Level Credit Assignment for Tool-Integrated Text-to-SQL

Tool-integrated Text-to-SQL parsing has emerged as a promising paradigm, framing SQL generation as a sequential decision-making process interleaved with tool execution. However, existing reinforcement learning approaches mainly rely on coarse-grained outcome supervision, resulting in a fundamental credit assignment problem: models receive the same reward for any trajectory that yields the correct answer, even when intermediate steps are redundant, inefficient, or erroneous. Consequently, models are encouraged to explore suboptimal reasoning spaces, limiting both efficiency and generalization. To address this problem, we propose FineStep, a novel framework for step-level credit assignment in tool-augmented Text-to-SQL. First, we introduce a reward design with independent process rewards to alleviate the signal sparsity of outcome supervision. Next, we present a step-level credit assignment mechanism to precisely quantify the value of each reasoning step. Finally, we develop a policy optimization method based on step-level advantages for efficient updates. Extensive experiments on BIRD benchmarks show that FineStep achieves state-of-the-art performance and reduces redundant tool interactions, with a 3.25% average EX gain over GRPO at the 4B scale.

Yaxun Dai, Baolin Sun, Junying Wang +6

Institute of Computer Science and Technology, Soochow University, China · Ant Digital Technologies, Ant Group · School of Management, University of Science and Technology of China

Credit Assignment in RL Text-to-SQL Step-Level Credit Assignment RL Fine-Tuning RL for Tool Use

May 6, 2026cs.LG

CRAFT: Counterfactual-to-Interactive Reinforcement Fine-Tuning for Driving Policies

Open-loop imitation learning has advanced modern autonomous driving policy architectures, but closed-loop deployment remains vulnerable to policy-induced distribution shift. Existing post-training paradigms exhibit fundamental trade-offs: closed-loop RL fine-tuning provides grounded feedback from executed actions but is constrained by the sparsity of informative events, whereas counterfactual fine-tuning provides dense supervision over candidate futures but inherits bias from imperfect future estimates. We introduce Counterfactual-to-Interactive Reinforcement Fine-Tuning (CRAFT), an on-policy framework that formulates closed-loop post-training as proxy-residual optimization. CRAFT uses group-normalized counterfactual advantages as a dense proxy for real closed-loop advantages and aligns this proxy with the closed-loop world through grounded residual correction from interaction-critical events. To stabilize adaptation, CRAFT regularizes the online policy toward an EMA teacher via asymmetric KL self-distillation. Theoretically, CRAFT decomposes the real closed-loop policy gradient into proxy and residual terms under the same visited-state distribution, reducing residual variance with an aligned proxy while mitigating proxy bias through grounded residual approximation. Empirically, CRAFT achieves the strongest closed-loop gains on Bench2Drive across hierarchical planning, vision-language-action, and vocabulary-scoring architectures. Ablations, scaling behavior, stability analyses, and transfer results further validate the complementary roles of dense counterfactual proxy and grounded residual correction. Project page: https://currychen77.github.io/CRAFT.

Keyu Chen, Nanfei Ye, Yida Wang +4

School of Vehicle and Mobility, Tsinghua University · Li Auto Inc

Reinforcement Learning Autonomous Driving Planning RL Post-Training RL Fine-Tuning Counterfactual Learning

May 6, 2026cs.SE

Towards Robust LLM Post-Training: Automatic Failure Management for Reinforcement Fine-Tuning

Reinforcement fine-tuning (RFT) has become a core paradigm for post-training large language models, yet its training process remains highly fragile. Existing efforts mainly improve reliability at the system level or address specific issues in individual subproblems by modifying RFT algorithms. Despite their effectiveness, they largely overlook the problem of failure management at the training-process level. When training goes wrong, practitioners still rely heavily on expert-driven manual inspection and correction, and automatic failure management for RFT remains largely unexplored. In this paper, we take a first step toward systematic failure management for reinforcement fine-tuning. To understand the empirical structure of RFT failures, we first construct RFT-FaultBench, the first benchmark for fine-grained failures in reinforcement fine-tuning, covering 5 fault families, 16 fault types, 779 training runs, 22,549 train-step records, and 1,457,288 trajectory-level records. Based on this benchmark, we conduct a comprehensive empirical study showing that RFT failures are both observable from training dynamics and distinguishable through their empirical fault fingerprints. Building on these findings, we propose RFT-FM, an automatic failure management framework for reinforcement fine-tuning that unifies anomaly detection, failure diagnosis, and auto remediation in a closed loop. Experimental results show that RFT-FaultBench is neither trivial nor saturated: it exhibits clear anomaly structure while still posing substantial challenges, especially under subtle fault settings. Moreover, RFT-FM shows strong capability in detecting, diagnosing, and mitigating RFT failures.

Lingzhe Zhang, Tong Jia, Yunpeng Zhai +6

Peking University, Beijing, China · Alibaba Group, China · University of Illinois Chicago, United States

Reinforcement Learning RL Fine-Tuning

May 5, 2026cs.LG

DGPO: Distribution Guided Policy Optimization for Fine Grained Credit Assignment

Reinforcement learning is crucial for aligning large language models to perform complex reasoning tasks. However, current algorithms such as Group Relative Policy Optimization suffer from coarse grained, sequence level credit assignment, which severely struggles to isolate pivotal reasoning steps within long Chain of Thought generations. Furthermore, the standard unbounded Kullback Leibler divergence penalty induces severe gradient instability and mode seeking conservatism, ultimately stifling the discovery of novel reasoning trajectories. To overcome these limitations, we introduce Distribution Guided Policy Optimization, a novel critic free reinforcement learning framework that reinterprets distribution deviation as a guiding signal rather than a rigid penalty. DGPO replaces the volatile KL divergence with the bounded Hellinger distance to safely quantify token level exploration without the risk of gradient explosion. To effectively distinguish genuine reasoning breakthroughs from hallucinatory noise, we propose an entropy gating mechanism that scales this deviation by the policy`s epistemic uncertainty. By dynamically redistributing the coarse sequence-level advantage to individual tokens based on these gated scores, DGPO heavily incentivizes critical exploratory steps while suppressing unwarranted, low-entropy deviations. Consequently, DGPO completely eliminates the traditional token-level KL penalty and achieves fine-grained credit reallocation without the computational overhead of an additional value network. Extensive empirical evaluations demonstrate that DGPO sets a new state-of-the-art for critic free alignment. Notably, on the Qwen2.5-32B architecture, DGPO achieves 60.0% Avg@32 accuracy and 46.0% Avg@32 accuracy on the challenging AIME2024 and AIME2025 benchmarks respectively, substantially outperforming competitive baselines like DAPO.

Hongbo Jin, Rongpeng Zhu, Zhongjing Du +4

Peking University · SJTU · Tsinghua University

Token-Level Credit Assignment RL for Language Model Reasoning Credit Assignment in RL Step-Level Credit Assignment in RL RL Fine-Tuning

May 4, 2026cs.LG

OGPO: Sample Efficient Full-Finetuning of Generative Control Policies

Generative control policies (GCPs), such as diffusion- and flow-based control policies, have emerged as effective parameterizations for robot learning. This work introduces Off-policy Generative Policy Optimization (OGPO), a sample-efficient algorithm for finetuning GCPs that maintains off-policy critic networks to maximize data reuse and propagate policy gradients through the full generative process of the policy via a modified PPO objective, using critics as the terminal reward. OGPO achieves state-of-the-art performance on manipulation tasks spanning multi-task settings, high-precision insertion, and dexterous control. To our knowledge, it is also the only method that can fine-tune poorly-initialized behavior cloning policies to near full task-success with no expert data in the online replay buffer, and does so with few task-specific hyperparameter tuning. Through extensive empirical investigations, we demonstrate that OGPO drastically outperforms methods alternatives on policy steering and learning residual corrections, and identify the key mechanisms behind its performance. We further introduce practical stabilization tricks, including success-buffer regularization, two-sided conservative advantages, and Q-variance reduction, to mitigate critic over-exploitation across state- and pixel-based settings. Beyond proposing OGPO, we conduct a systematic empirical study of GCP finetuning, identifying the stabilizing mechanisms and failure modes that govern successful off-policy full-policy improvement.

Sarvesh Patil, Mitsuhiko Nakamoto, Manan Agarwal +14

Carnegie Mellon University · University of California, Berkeley · University of Washington +1

RL for Robotics RL for Generative Models Residual Policy Learning RL Fine-Tuning Off-Policy RL

May 2, 2026cs.CL

Injecting Distributional Awareness into MLLMs via Reinforcement Learning for Deep Imbalanced Regression

Multimodal large language models (MLLMs) struggle with numerical regression under long-tailed target distributions. Token-level supervised fine-tuning (SFT) and point-wise regression rewards bias learning toward high-density regions, leading to regression-to-the-mean behavior and poor tail performance. We identify the lack of cross-sample relational supervision as a key limitation of existing MLLM training paradigms. To address it, we propose a distribution-aware reinforcement learning framework based on Group Relative Policy Optimization, which introduces batch-level comparison-based supervision via the Concordance Correlation Coefficient-based reward to align predicted and ground-truth distributions in terms of correlation, scale, and mean. The framework is plug-and-play, requiring no architectural modification. Experiments on a unified suite of long-tailed regression benchmarks show consistent improvements over SFT and existing MLLM regression methods, with particularly strong gains in medium- and few-shot regimes.

Yao Du, Shanshan Song, Xiaomeng Li

The Hong Kong University of Science and Technology, Hong Kong SAR, China.

Multimodal Large Language Models RL Fine-Tuning

May 1, 2026cs.LG

TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination

Multi-agent LLM systems have shown promise for complex reasoning, yet recent evaluations reveal they often underperform single-model baselines. We identify a structural failure mode in sequential fine-tuning of shared-context teams: updating one agent shifts the team's context distribution, and when subsequent updates are evaluated on cached rollouts, this mismatch compounds. We formalize this as the compounding occupancy shift and prove that stale-occupancy evaluation incurs a penalty that scales quadratically with the number of agents. In contrast, intermediate-occupancy evaluation reduces this to linear scaling. We propose TeamTR, a trust-region framework that resamples trajectories after each component update and enforces per-agent divergence control, yielding rigorous per-update and per-stage improvement lower bounds. Experiments show that TeamTR outperforms single-agent and sequential baselines with 7.1% on average, mitigates coordination regressions, and supports plug-and-play component replacement. Code is available at https://github.com/Yydc/TeamTR.

Yi Xie, Siao Liu, Falong Fan +3

Department of Electrical & Computer Engineering, University of Arizona · Future Science and Engineering College, Soochow University · INSAIT, Sofia University ”St. Kliment Ohridski” +1

Multi-Agent System Optimization Multi-Agent Coordination Trust-Region Optimization Multi-Agent LLMs RL Fine-Tuning Multi-Agent Reinforcement Learning

May 1, 2026cs.AI

Iterative Finetuning is Mostly Idempotent

If a model has some behavioral tendency, such as sycophancy or misalignment, and it is trained on its own outputs, will the tendency be amplified in the next generation of models? We study this question by training a series of models where each model is finetuned on data generated by its predecessor, and the initial model is seeded with some persona or belief. We test three settings: supervised finetuning (SFT) on instruct models, synthetic document finetuning (SDF) on base models, and direct preference optimization (DPO). In the SFT and SDF settings, traits mostly decay or remain constant so that further finetuning cycles do nothing. In rare cases when amplification occurs, it generally comes at the cost of coherence. In the DPO setting, trait amplification can reliably occur when a model is continually trained with a preference for its own outputs, but vanishes when models are reinitialized at each cycle. Overall, our results suggest that amplification most likely comes from continual post-training, and limiting this stage may be an effective defense. For non-RL finetuning, trait amplification is rare and very sensitive to data quantity, making it significantly less likely to occur accidentally. Finally, the amplification-coherence tradeoff serves as a natural deterrent against trait amplification.

Zephaniah Roe, Jack Sanderson, Dang Nguyen +5

University of Chicago

Supervised Fine-Tuning LLM Post-Training Language Model Self-Improvement RL Fine-Tuning

Apr 29, 2026cs.LG

Distributional Alignment Games for Answer-Level Fine-Tuning

We focus on the problem of \emph{Answer-Level Fine-Tuning} (ALFT), where the goal is to optimize a language model based on the correctness or properties of its final answers, rather than the specific reasoning traces used to produce them. Directly optimizing answer-level objectives is computationally intractable due to the need to marginalize over the vast space of latent reasoning paths. To overcome this, we propose a general game-theoretical framework that lifts the problem to a \emph{Distributional Alignment Game}. We formulate ALFT as a two-player game between a Policy (the generator) and a Target (an auxiliary distribution). We prove that the Nash Equilibrium of this game corresponds exactly to the solution of the original answer-level optimization problem. This variational perspective transforms the intractable marginalization problem into a tractable projection problem. We demonstrate that this framework unifies recent approaches to diversity and self-improvement (coherence) and provide efficient algorithms compatible with Group Relative Policy Optimization (GRPO), such as Coherence-GRPO, yielding significant complexity gains in mathematical reasoning tasks.

Mehryar Mohri, Jon Schneider, Yifan Wu

Google Research · Microsoft Research

Fine-Tuning RL for Language Model Reasoning RL Fine-Tuning

Apr 29, 2026cs.CL

EXPO-SQL: Execution-based Clause-level Policy Optimization for Text-to-SQL

Text-to-SQL enables users to query databases using natural language by generating executable SQL queries. Recent methods have increasingly adopted Large Language Models based reinforcement learning (RL) to leverage execution feedback for training. However, existing RL methods assign uniform query-level rewards to all clauses in a SQL query, treating correct and incorrect clauses equally. This coarse-grained reward design leads to insufficient learning signals for correct SQL generation. To address this issue, we propose EXPO-SQL (EXecution-based clause-level Policy Optimization for Text-to-SQL) which provides fine-grained supervision through clause-level rewards. To assign clause-level rewards, our method identifies erroneous clauses by analyzing execution results, including error messages and clause-wise incremental execution. Experiments on widely-used Text-to-SQL benchmarks demonstrate that EXPO-SQL significantly outperforms existing supervised fine-tuning, prompting, and RL-based methods through fine-grained clause-level learning. Our code is available at https://github. com/jhn25/EXPO-SQL.

Jaehoon Lee, CheolWon Na, Suyoung Bae +4

College of Computing and Informatics Sungkyunkwan University, South Korea

Reward Shaping Text-to-SQL RL Fine-Tuning Fine-Grained Credit Assignment

Apr 28, 2026cs.LG

DGLight: DQN-Guided GRPO Fine-Tuning of Large Language Models for Traffic Signal Control

Traffic signal control (TSC) plays a central role in reducing congestion and maintaining urban mobility. This dissertation introduces DGLight, a critic-guided reinforcement-learning framework for adapting a pretrained large language model to TSC. DGLight first trains a CoLight-based Deep Q-Network critic to estimate traffic-aware action values from structured intersection states, then uses the frozen critic to score candidate language-model actions and optimize the policy with Group Relative Policy Optimization (GRPO). The resulting controller maps traffic states to interpretable reasoning traces and signal decisions while learning from dense per-state supervision rather than raw cumulative environment rewards. Experiments on TSC benchmarks covering Jinan and Hangzhou show that DGLight is the strongest overall method among the compared LLM-based controllers, remains competitive with strong RL baselines, and transfers well to city datasets not used to fit the critic. Qualitative examples further show that the model's generated reasoning is interpretable and aligned with the chosen signal phase. The project code is available

[here](https://github.com/yyccbb/FYP_LLMTSC)

.

Chenbo Yu

National University of Singapore Singapore

Language Model-Based Control Intelligent Transportation Systems Traffic Signal Control RL Fine-Tuning

Apr 26, 2026cs.LG

SFT-then-RL Outperforms Mixed-Policy Methods for LLM Reasoning

Recent mixed-policy optimization methods for LLM reasoning that interleave or blend supervised and reinforcement learning signals report improvements over the standard SFT-then-RL pipeline. We show that numerous recently published research papers rely on a faulty baseline caused by two distinct bugs: a CPU-offloaded optimizer bug in DeepSpeed that silently drops intermediate micro-batches during gradient accumulation (affecting multiple downstream frameworks including TRL, OpenRLHF and Llama-Factory), and a loss aggregation bug in OpenRLHF that incorrectly weights per-mini-batch losses. Together they suppress SFT performance, with the optimizer bug accounting for most of the gap and the loss aggregation bug contributing a smaller additional effect. Once corrected, the standard SFT-then-RL pipeline surpasses every published mixed-policy method we evaluate by +3.8 points on math benchmarks with Qwen2.5-Math-7B and by +22.2 points with Llama-3.1-8B. Even a truncated variant with just 50 RL steps outperforms mixed-policy methods on math benchmarks while using fewer FLOPs.

Alexis Limozin, Eduard Durech, Torsten Hoefler +2

ETH AI Center, ETH Zürich · EPFL · Allen Institute for AI

RL for Language Model Reasoning Policy Optimization LLM Post-Training RL Fine-Tuning LLM Reasoning

Apr 25, 2026cs.LG

GIFT: Global stabilisation via Intrinsic Fine Tuning

Deep reinforcement learning policies achieve strong performance in complex continuous control environments with nonlinear contact forces. However, these policies often produce chaotic state dynamics, with trivially small changes to the initial conditions significantly impacting the long-term behaviour of the control system. This high sensitivity to initial conditions limits the application of Deep RL to real-world control systems where performance and stability guarantees are often required. To address this issue, we propose Global stabilisation via Intrinsic Fine Tuning (GIFT), a general-purpose training framework which directly optimises the global stability of existing high-performing deep RL policies using a custom reward function. We demonstrate that GIFT increase the stability of the control interaction while maintaining comparable task performance, thereby improving the suitability of deep RL policies for real-world control systems.

Rory Young, Nicolas Pugeault

School of Computing Science University of Glasgow

RL Control Reinforcement Learning RL Fine-Tuning Robust RL

Apr 24, 2026cs.LG

RL Token: Bootstrapping Online RL with Vision-Language-Action Models

Vision-language-action (VLA) models can learn to perform diverse manipulation skills "out of the box," but achieving the precision and speed that real-world tasks demand requires further fine-tuning -- for example, via reinforcement learning (RL). We introduce a lightweight method that enables sample-efficient online RL fine-tuning of pretrained VLAs using just a few hours of real-world practice. We (1) adapt the VLA to expose an "RL token," a compact readout representation that preserves task-relevant pretrained knowledge while serving as an efficient interface for online RL, and (2) train a small actor-critic head on this RL token to refine the actions, while anchoring the learned policy to the VLA. Online RL with the RL token (RLT) makes it possible to fine-tune even large VLAs with RL quickly and efficiently. Across four real-robot tasks (screw installation, zip tie fastening, charger insertion, and Ethernet insertion), RLT improves the speed on the hardest part of the task by up to 3x and raises success rates significantly within minutes to a few hours of practice. It can even surpass the speed of human teleoperation on some of the tasks.

Charles Xu, Jost Tobias Springenberg, Michael Equi +4

Physical Intelligence

RL for Robotics Robot Skill Learning Robotic Manipulation Data-Efficient Robot Learning Vision-Language-Action Models RL Fine-Tuning

Apr 24, 2026cs.CR

Behavioral Canaries: Auditing Private Retrieved Context Usage in RL Fine-Tuning

In agentic workflows, LLMs frequently process retrieved contexts that are legally protected from further training. However, auditors currently lack a reliable way to verify if a provider has violated the terms of service by incorporating these data into post-training, especially through Reinforcement Learning (RL). While standard auditing relies on verbatim memorization and membership inference, these methods are ineffective for RL-trained models, as RL primarily influences a model's behavioral style rather than the retention of specific facts. To bridge this gap, we introduce Behavioral Canaries, a new auditing mechanism for RLFT pipelines. The framework instruments preference data by pairing document triggers with feedback that rewards a distinctive stylistic response, inducing a latent trigger-conditioned preference if such data are used in training. Empirical results show that these behavioral signals enable detection of unauthorized document-conditioned training, achieving a 67% detection rate at a 10% false-positive rate (AUROC = 0.756) at a 1% canary injection rate. More broadly, our results establish behavioral canaries as a new auditing mechanism for RLFT pipelines, enabling auditors to test for training-time influence even when such influence manifests as distributional behavioral change rather than memorization.

Chaoran Chen, Dayu Yuan, Peter Kairouz

Google

LLM Auditing RL Fine-Tuning

Apr 21, 2026cs.AI

Reinforcement Learning Improves LLM Accuracy and Reasoning in Disease Classification from Radiology Reports

Accurate disease classification from radiology reports is essential for many applications. While supervised fine-tuning (SFT) of lightweight LLMs improves accuracy, it can degrade reasoning. We propose a two-stage approach: SFT on disease labels followed by Group Relative Policy Optimization (GRPO) to refine predictions by optimizing accuracy and format without reasoning supervision. Across three radiologist-annotated datasets, SFT outperformed baselines and GRPO further improved classification and enhanced reasoning recall and comprehensiveness.

Yishu Wei, Yi Lin, Adam Flanders +2

Department of Population Health Sciences, Weill Cornell Medicine, New York, NY · Department of Radiology, Thomas Jefferson University, Philadelphia, PA · Department of Radiology, Weill Cornell Medicine, New York, NY

Healthcare Clinical Reasoning RL Fine-Tuning Medical LLMs

Apr 20, 2026cs.LG

Bounded Ratio Reinforcement Learning

Proximal Policy Optimization (PPO) has become the predominant algorithm for on-policy reinforcement learning due to its scalability and empirical robustness across domains. However, there is a significant disconnect between the underlying foundations of trust region methods and the heuristic clipped objective used in PPO. In this paper, we bridge this gap by introducing the Bounded Ratio Reinforcement Learning (BRRL) framework. We formulate a novel regularized and constrained policy optimization problem and derive its analytical optimal solution. We prove that this solution ensures monotonic performance improvement. To handle parameterized policy classes, we develop a policy optimization algorithm called Bounded Policy Optimization (BPO) that minimizes an advantage-weighted divergence between the policy and the analytic optimal solution from BRRL. We further establish a lower bound on the expected performance of the resulting policy in terms of the BPO loss function. Notably, our framework also provides a new theoretical lens to interpret the success of the PPO loss, and connects trust region policy optimization and the Cross-Entropy Method (CEM). We additionally extend BPO to Group-relative BPO (GBPO) for LLM fine-tuning. Empirical evaluations of BPO across MuJoCo, Atari, and complex IsaacLab environments (e.g., Humanoid locomotion), and of GBPO for LLM fine-tuning tasks, demonstrate that BPO and GBPO generally match or outperform PPO and GRPO in stability and final performance.

Yunke Ao, Le Chen, Bruce D. Lee +5

ETH Zurich · MPI for Intelligent Systems · University of Alberta +2

Reinforcement Learning Group Relative Policy Optimization Policy Optimization Proximal Policy Optimization Trust Region Policy Optimization RL Fine-Tuning+1

Apr 19, 2026cs.LG

Internalizing Outcome Supervision into Process Supervision: A New Paradigm for Reinforcement Learning for Reasoning

The central challenge of reinforcement learning for reasoning lies not only in the sparsity of outcome-level supervision, but more fundamentally in how to transform feedback provided only at the end of a sequence into fine-grained learning signals that can guide intermediate reasoning steps. Existing approaches either rely on outcome-level rewards for sequence-level optimization, which makes precise credit assignment difficult, or depend on externally constructed process supervision, which is costly and difficult to scale sustainably. To address this, we propose a new perspective: reinforcement learning for reasoning can be understood as the problem of internalizing outcome supervision into process supervision. From this perspective, we introduce a supervision-internalization method for reinforcement learning for reasoning, enabling the model to automatically extract process-level learning signals through identifying, correcting, and reusing failed reasoning trajectories, thereby achieving finer-grained policy optimization under outcome-only supervision. We further abstract this idea into a new training paradigm, in which the model continually generates and refines its own internal process supervision during reinforcement learning, opening a new path for fine-grained credit assignment in reinforcement learning for reasoning that differs from externally provided process supervision.

Fei Ding, Yongkang Zhang, Runhao Liu +4

Alibaba Group · Tsinghua University

RL for Language Model Reasoning Credit Assignment in RL Step-Level Credit Assignment in RL Process Supervision RL Fine-Tuning

Apr 18, 2026eess.SP

TimeRFT: Stimulating Generalizable Time Series Forecasting for TSFMs via Reinforcement Finetuning

Time Series Foundation Models (TSFMs) advance generalization and data efficiency in time series forecasting by unified large-scale pretraining. But TSFMs remain lacking when adapting to specific downstream forecasting tasks for two reasons. First, the non-stationary and uncertain nature of time series data lead to inevitable temporal distribution shifts between historical training and future testing data, while current Supervised FineTuning (SFT)-based methods are prone to overfitting and may degrade generalization. Second, training data availability varies across forecasting tasks, requiring TSFMs to generalize well under diverse data regimes. To address these challenges, we introduce the Time series Reinforcement Finetuning (TimeRFT) paradigm for TSFM downstream adaptation, which consists of two task-specific training recipes: i) A forecasting quality-based temporal reward mechanism that conducts a multi-faceted evaluation of the contribution of each prediction step to overall forecasting accuracy. ii) A forecasting difficulty-based data selection strategy to identify time series samples with generalizable predictive patterns and informative training signals. Extensive experiments demonstrate TimeRFT can consistently outperform SFT-based adaptation methods across various real-world forecasting tasks and training data regimes, enhancing prediction accuracy and generalization against unforeseen distribution shifts.

Siyang Li, Yize Chen, Zijie Zhu +4

HKUST(GZ) Guangzhou, China · University of Alberta Edmonton, Canada · Alibaba Cloud Guangzhou, China +3

Time Series Forecasting OOD Generalization Non-Stationary Time Series Forecasting RL Fine-Tuning Time-Series Foundation Models

Apr 18, 2026cs.LG

AutoOR: Scalably Post-training LLMs to Autoformalize Operations Research Problems

Optimization problems are central to decision-making in manufacturing, logistics, scheduling, and other industrial settings. Translating complicated descriptions of these problems into solver-ready formulations requires specialized operations research (OR) expertise, making it hard to scale. We present AutoOR, a scalable synthetic data generation and reinforcement learning pipeline that trains LLMs to autoformalize optimization problems specified in natural language across linear, mixed-integer, and non-linear categories. AutoOR generates verified training data from standard optimization forms and uses solver execution feedback as the reward signal for RL post-training. AutoOR applied to an 8B model achieves state-of-the-art or competitive results across six established OR benchmarks, matching significantly larger frontier models. For a non-linear problem class involving physical dynamics, where frontier models score near 0%, we introduce a curriculum RL strategy that bootstraps from limited initial training data to make this class tractable for post-training. We believe that methods such as AutoOR can significantly accelerate industrial decision-making with AI.

Sumeet Ramesh Motwani, Chuan Du, Aleksander Petrov +4

1X · University of Oxford

LLM Post-Training RL Fine-Tuning Natural Language to Optimization Modeling

Apr 17, 2026cs.LG

Towards Robust Endogenous Reasoning: Unifying Drift Adaptation in Non-Stationary Tuning

Reinforcement Fine-Tuning (RFT) has established itself as a critical paradigm for the alignment of Multi-modal Large Language Models (MLLMs) with complex human values and domain-specific requirements. Nevertheless, current research primarily focuses on mitigating exogenous distribution shifts arising from data-centric factors, the non-stationarity inherent in the endogenous reasoning remains largely unexplored. In this work, a critical vulnerability is revealed within MLLMs: they are highly susceptible to endogenous reasoning drift, across both thinking and perception perspectives. It manifests as unpredictable distribution changes that emerge spontaneously during the autoregressive generation process, independent of external environmental perturbations. To adapt it, we first theoretically define endogenous reasoning drift within the RFT of MLLMs as the multi-modal concept drift. In this context, this paper proposes Counterfactual Preference Optimization ++ (CPO++), a comprehensive and autonomous framework adapted to the multi-modal concept drift. It integrates counterfactual reasoning with domain knowledge to execute controlled perturbations across thinking and perception, employing preference optimization to disentangle spurious correlations. Extensive empirical evaluations across two highly dynamic and safety-critical domains: medical diagnosis and autonomous driving. They demonstrate that the proposed framework achieves superior performance in reasoning coherence, decision-making precision, and inherent robustness against extreme interference. The methodology also exhibits exceptional zero-shot cross-domain generalization, providing a principled foundation for reliable multi-modal reasoning in safety-critical applications.

Xiaoyu Yang, En Yu, Wei Duan +1

Australian Artificial Intelligence Institute (AAII), Faculty of Engineering and Information Technology, University of Technology Sydney, Australia

Multimodal Robustness Multimodal Large Language Models Multimodal Reasoning RL Fine-Tuning

Apr 16, 2026cs.LG

Reward Weighted Classifier-Free Guidance as Policy Improvement in Autoregressive Models

Consider an auto-regressive model that produces outputs x (e.g., answers to questions, molecules) each of which can be summarized by an attribute vector y (e.g., helpfulness vs. harmlessness, or bio-availability vs. lipophilicity). An arbitrary reward function r(y) encodes tradeoffs between these properties. Typically, tilting the model's sampling distribution to increase this reward is done at training time via reinforcement learning. However, if the reward function changes, re-alignment requires re-training. In this paper, we show that a reward weighted classifier-free guidance (RCFG) can act as a policy improvement operator in this setting, approximating tilting the sampling distribution by the Q function. We apply RCFG to molecular generation, demonstrating that it can optimize novel reward functions at test time. Finally, we show that using RCFG as a teacher and distilling into the base policy to serve as a warm start significantly speeds up convergence for standard RL.

Alexander Peysakhovich, William Berman

∗Sutter Hill Ventures · †Sutter Hill Ventures

Classifier-Free Guidance Autoregressive Generation RL for Generative Models RL Fine-Tuning

Apr 16, 2026cs.LG

Beyond Importance Sampling: Rejection-Gated Policy Optimization

We propose a new perspective on policy optimization: rather than reweighting all samples by their importance ratios, an optimizer should select which samples are trustworthy enough to drive a policy update. Building on this view, we introduce Rejection-Gated Policy Optimization (RGPO), which replaces the importance sampling ratio r_theta = pi_theta / pi_old with a smooth, differentiable acceptance gate alpha_theta(s, a) = g(r_theta(s, a)) in the range [0, 1]. Unlike prior work that applies rejection sampling as a data-level heuristic before training, RGPO elevates rejection to an optimization principle: the gate participates directly in gradient computation and is implicitly updated alongside the policy. RGPO provides a unified framework: the policy gradients of TRPO, PPO, and REINFORCE all correspond to specific choices of the effective gradient weight w(r) = g'(r) * r. We prove that RGPO guarantees finite, bounded gradient variance even when importance sampling ratios are heavy-tailed (where IS variance diverges). We further show that RGPO incurs only a bounded, controllable bias and provides an approximate monotonic policy improvement guarantee analogous to TRPO. RGPO matches PPO in computational cost, requires no second-order optimization, and extends naturally to RLHF-style preference alignment. In online preference fine-tuning of Qwen2.5-1.5B-Instruct on Anthropic HH-RLHF (n = 3 seeds), RGPO uses a dual-ratio gate that anchors learning to both the previous policy and the reference model, achieving a Pareto-dominant outcome: the highest reward among online RL methods (+14.8% vs. PPO-RLHF) and the lowest KL divergence to the reference model (-16.0% vs. PPO-RLHF, -53.1% vs. GRPO).

Ziwu Sun, Zhen Gao, Jiyong Zhang +1

Importance Sampling Policy Optimization Policy Gradient Methods RL Fine-Tuning Policy Learning