RL for Language Models

RL: Reinforcement Learning

Follow this topic Ask about this topic

This field studies how feedback can shape a language model’s behavior after pretraining, including how it answers, reasons, uses tools, and handles uncertainty. Common approaches include preference-based reinforcement learning from human or AI feedback, reinforcement learning with verifiable rewards for tasks such as math and coding, and policy methods such as PPO and GRPO; rubrics and step- or turn-level rewards provide finer guidance. Researchers evaluate these methods on reasoning benchmarks as well as dialogue, tool use, multilingual tasks, and domain-specific applications. A central tension is that sparse or imperfect rewards can be costly to obtain and easy to exploit, while reliable credit assignment, exploration, and generalization beyond the training setting remain difficult.

Latest papers

445 papers63 papers in the last four weeks, up 17% on the four weeks before. 0.4% of all new papers.

Apr 3, 2026

JoyAI-LLM Flash: Advancing Mid-Scale LLMs with Token Efficiency

We introduce JoyAI-LLM Flash, an efficient Mixture-of-Experts (MoE) language model designed to redefine the trade-off between strong performance and token efficiency in the sub-50B parameter regime. JoyAI-LLM Flash is pretrained on a massive corpus of 20 trillion tokens and further optimized through a rigorous post-training pipeline, including supervised fine-tuning (SFT), Direct Preference Optimization (DPO), and large-scale reinforcement learning (RL) across diverse environments. To improve token efficiency, JoyAI-LLM Flash strategically balances thinking and non-thinking cognitive modes and introduces FiberPO, a novel RL algorithm inspired by fibration theory that decomposes trust-region maintenance into global and local components, providing unified multi-scale stability control for LLM policy optimization. To enhance architectural sparsity, the model comprises 48B total parameters while activating only 2.7B parameters per forward pass, achieving a substantially higher sparsity ratio than contemporary industry leading models of comparable scale. To further improve inference throughput, we adopt a joint training-inference co-design that incorporates dense Multi-Token Prediction (MTP) and Quantization-Aware Training (QAT). We release the checkpoints for both JoyAI-LLM-48B-A3B Base and its post-trained variants on Hugging Face to support the open-source community.

Aichen Cai, Anmeng Zhang, Anyu Li +66

Language Modeling Mixture-of-Experts Language Models RL for Language Models LLM Inference Acceleration

Apr 3, 2026

KARL: Mitigating Hallucinations in LLMs via Knowledge-Boundary-Aware Reinforcement Learning

Enabling large language models (LLMs) to appropriately abstain from answering questions beyond their knowledge is crucial for mitigating hallucinations. While existing reinforcement learning methods foster autonomous abstention, they often compromise answer accuracy because their static reward mechanisms, agnostic to models' knowledge boundaries, drive models toward excessive caution. In this work, we propose KARL, a novel framework that continuously aligns an LLM's abstention behavior with its evolving knowledge boundary. KARL introduces two core innovations: a Knowledge-Boundary-Aware Reward that performs online knowledge boundary estimation using within-group response statistics, dynamically rewarding correct answers or guided abstention; and a Two-Stage RL Training Strategy that first explores the knowledge boundary and bypasses the "abstention trap", and subsequently converts incorrect answers beyond the knowledge boundary into abstentions without sacrificing accuracy. Extensive experiments on multiple benchmarks demonstrate that KARL achieves a superior accuracy-hallucination trade-off, effectively suppressing hallucinations while maintaining high accuracy across both in-distribution and out-of-distribution scenarios.

Cheng Gao, Cheng Huang, Kangyang Luo +5

Tsinghua University

LLM Hallucination Mitigation LLM Abstention RL for Language Models

Apr 3, 2026

Rubrics to Tokens: Bridging Response-level Rubrics and Token-level Rewards in Instruction Following Tasks

Rubric-based Reinforcement Learning (RL) has emerged as a promising approach for aligning Large Language Models (LLMs) with complex, open-domain instruction following tasks. However, existing methods predominantly rely on response-level rewards, introducing severe reward sparsity and reward ambiguity problems. To address these issues, we propose Rubrics to Tokens (RTT), a novel rubric-based RL framework that bridges coarse response-level scores and fine-grained token-level credit assignment. RTT introduces a Token-Level Relevance Discriminator to predict which tokens in the response are responsible for a specific constraint, and optimizes the policy model via RTT-GRPO, which integrates response-level and token-level advantages within a unified framework. Furthermore, when transitioning from one-dimensional, outcome-level reward to three-dimensional reward space in the token-level rubric-based RL, we propose a novel group normalization method, called Intra-sample Token Group Normalization, to accommodate this shift. Extensive experiments and benchmarks demonstrate that RTT consistently outperforms other baselines in both instruction- and rubric-level accuracy across different models.

Tianze Xu, Yanzhao Zheng, Pengrui Lu +11

Shanghai Jiao Tong University · GAIR · Alibaba Group +2

RL for Language Models Rubric-Based RL Instruction Following Reward Design for RL Step-Level Credit Assignment in RL

Apr 3, 2026

Generalization Limits of Reinforcement Learning Alignment

The safety of large language models (LLMs) relies on alignment techniques such as reinforcement learning from human feedback (RLHF). However, recent theoretical analyses suggest that reinforcement learning-based training does not acquire new capabilities but merely redistributes the utilization probabilities of existing ones. In this study, we propose ``compound jailbreaks'' targeting OpenAI gpt-oss-20b, which exploit the generalization failures of alignment. This approach combines multiple attack techniques -- each individually defended against -- to saturate the instruction hierarchy maintenance process. Our evaluation shows that the attack success rate (ASR) increased from 14.3% with individual methods to 71.4% with the combined approach. These results provide empirical evidence for the hypothesis that safety training does not generalize as broadly as model capabilities, highlighting the need for multifaceted safety evaluations using compound attack scenarios.

Haruhi Shida, Koo Imai, Keigo Kansa

Aladdin Security Inc.

LLM Alignment Adversarial Attacks on LLMs Jailbreak Attacks Language Model Safety Evaluation RL for Language Models

Apr 1, 2026

Policy Improvement Reinforcement Learning

Reinforcement learning has become a central post-training paradigm for improving LLM and agent capabilities. Yet existing RL post-training methods share a common blind spot: they construct local learning signals from sampled trajectories, rewards, or feedback-conditioned targets, then update the policy without explicitly verifying whether the resulting policy outperforms its predecessor. Optimizing these local signals does not necessarily produce a better policy, while finite sampling, generation stochasticity and feedback noise can further widen this gap. We argue that the missing ingredient is policy improvement feedback: the ability to measure progress across policy iterations. We introduce Policy Improvement Reinforcement Learning (PIRL), which formulates inter-iteration performance gain as an explicit objective structurally aligned with final task performance. Building on PIRL, we propose Policy Improvement Policy Optimization (PIPO), a plug-in closed-loop framework that verifies the previous update against a sliding-window historical performance anchor. PIPO uses this improvement feedback to modulate the local learning signal of the base policy optimization algorithm, reinforcing updates associated with measured progress and suppressing those associated with performance drops. We provide theoretical evidence that PIPO locally aligns policy updates with the PIRL improvement objective. Experiments on mathematical reasoning, code, tool-use, and self-distillation settings show that PIPO yields consistent gains across PPO, group-relative, and self-distillation policy optimization families.

Huaiyang Wang, Xiaojie Li, Xiaohan Wang +10

Beihang University · Meituan · Peking University

RL for Language Models RL for Language Model Reasoning Policy Optimization Proximal Policy Optimization RL Post-Training Policy Learning

Mar 31, 2026

REM-CTX: Automated Peer Review via Reinforcement Learning with Auxiliary Context

Most automated peer review systems rely on textual manuscript content alone, leaving visual elements such as figures and external scholarly signals underutilized. We introduce REM-CTX, a reinforcement-learning system that incorporates auxiliary context into the review generation process via correspondence-aware reward functions. REM-CTX trains an 8B-parameter language model with Group Relative Policy Optimization (GRPO) and combines a multi-aspect quality reward with two correspondence rewards that explicitly encourage alignment with auxiliary context. Experiments on manuscripts across Computer, Biological, and Physical Sciences show that REM-CTX achieves the highest overall review quality among six baselines, outperforming other systems with substantially larger commercial models, and surpassing the next-best RL baseline across both quality and contextual grounding metrics. Ablation studies confirm that the two correspondence rewards are complementary: each selectively improves its targeted correspondence reward while preserving all quality dimensions, and the full model outperforms all partial variants. Analysis of training dynamics reveals that the criticism aspect is negatively correlated with other metrics during training, suggesting that future studies should group multi-dimension rewards for review generation.

Pawin Taechoyotin, Daniel E. Acuna

Department of Computer Science, University of Colorado Boulder, Boulder, CO 80309

Automated Peer Review RL for Language Models Rubric-Based RL

Mar 30, 2026

ERPO: Token-Level Entropy-Regulated Policy Optimization for Large Reasoning Models

Reinforcement learning from verifiable rewards has significantly advanced the reasoning capabilities of large language models. However, Group Relative Policy Optimization (GRPO) typically assigns a uniform, sequence-level advantage to all tokens, thereby overlooking the intrinsic information heterogeneity along reasoning chains. We show that this coarse-grained credit assignment leads to premature entropy collapse and encourages the model to generate redundant, low-quality reasoning paths. Through systematic empirical analysis, we identify Critical Decision Pivots (CDPs): transient high-entropy states where the policy's trajectory is most sensitive to perturbations. These pivots represent the "forks in the road" where effective multi-path exploration is most crucial yet often suppressed by uniform advantage signals. Building on these insights, we propose Entropy-Regulated Policy Optimization (ERPO), which transitions the optimization focus from coarse sequences to fine-grained token dynamics. ERPO introduces three synergistic components: (i) Entropy-aware Gating, which adaptively amplifies exploration at CDPs to facilitate diverse path discovery; (ii) Bucket-based Implicit Normalization, which mitigates difficulty bias by aligning token progress windows; and (iii) Result-anchored Advantage Synthesis, which re-weights token-level signals via outcome-driven anchors. Extensive experiments on competitive mathematical benchmarks demonstrate that ERPO significantly outperforms GRPO. Notably, ERPO not only boosts reasoning accuracy but also yields significantly more concise and robust derivation paths, while achieving performance comparable to large models with orders of magnitude more parameters.

Song Yu, Li Li, Wenwen Zhao +1

School of Computer and Information Science, Southwest University, Chongqing, 400715, China

RL for Language Model Reasoning RL for Language Models LLM Reasoning Large Reasoning Models

Mar 26, 2026

TAPO: Translation Augmented Policy Optimization for Multilingual Mathematical Reasoning

Large Language Models (LLMs) have demonstrated remarkable proficiency in English mathematical reasoning, yet a significant performance disparity persists in multilingual contexts, largely attributed to deficiencies in language understanding. To bridge this gap, we introduce Translation-Augmented Policy Optimization (TAPO), a novel reinforcement learning framework built upon GRPO. TAPO enforces an explicit alignment strategy where the model leverages English as a pivot and follows an understand-then-reason paradigm. Crucially, we employ a step-level relative advantage mechanism that decouples understanding from reasoning, allowing the integration of translation quality rewards without introducing optimization conflicts. Extensive experiments reveal that TAPO effectively synergizes language understanding with reasoning capabilities and is compatible with various models. It outperforms baseline methods in both multilingual mathematical reasoning and translation tasks, while generalizing well to unseen languages and out-of-domain tasks.

Xu Huang, Zhejian Lai, Zixian Huang +2

National Key Laboratory for Novel Software Technology, Nanjing University · Shanghai Artificial Intelligence Laboratory

Cross-Lingual Reasoning in Language Models RL for Language Model Reasoning RL for Language Models Numerical Reasoning in Language Models

Mar 26, 2026

Train at Moving Edge: Online-Verified Prompt Selection for Efficient RL Training of Large Reasoning Model

Reinforcement learning (RL) has become essential for post-training large language models (LLMs) in reasoning tasks. While scaling rollouts can stabilize training and enhance performance, the computational overhead is a critical issue. In algorithms like GRPO, multiple rollouts per prompt incur prohibitive costs, as a large portion of prompts provide negligible gradients and are thus of low utility. To address this problem, we investigate how to select high-utility prompts before the rollout phase. Our experimental analysis reveals that sample utility is non-uniform and evolving: the strongest learning signals concentrate at the ``learning edge", the intersection of intermediate difficulty and high uncertainty, which shifts as training proceeds. Motivated by this, we propose HIVE (History-Informed and online-VErified prompt selection), a dual-stage framework for data-efficient RL. HIVE utilizes historical reward trajectories for coarse selection and employs prompt entropy as a real-time proxy to prune instances with stale utility. By evaluating HIVE across multiple math reasoning benchmarks and models, we show that HIVE yields significant rollout efficiency without compromising performance.

Jiahao Wu, Ning Lu, Shengcai Liu +6

Southern University of Science and Technology · The Hong Kong Polytechnic University · The Hong Kong University of Science and Technology +3

Prompt Selection for RL RL for Language Models Sample-Efficient RL Uncertainty Sampling

Mar 24, 2026

Reciprocal Co-Training (RCT): Coupling Gradient-Based and Non-Differentiable Models via Reinforcement Learning

Language models (LMs) and classical machine learning methods offer complementary strengths for predictive modeling, yet their fundamentally different representations and training paradigms hinder effective integration: LMs rely on gradient-based optimization over textual data, whereas models such as Random Forests (RF) employ non-differentiable feature partitioning. This work introduces a reciprocal co-training framework that couples an LM with an RF classifier via reinforcement learning, creating an iterative feedback loop in which each model improves using signals from the other. Tabular data are reformulated into standardized textual representations for the LM, whose embeddings augment the RF feature space, while calibrated RF probability estimates provide feedback signals that guide reinforcement learning updates of the LM. Experiments across three medical datasets, evaluated with both a domain-adapted clinical encoder (ClinicalBERT) and a larger instruction-tuned language model (Qwen2-7B-Instruct), demonstrate consistent performance gains for both model components. Ablation analyses indicate that iterative refinement, hybrid reward design, and dimensionality control jointly contribute to these gains. SHAP analysis further confirms that LM-derived representations are among the most important inputs to the RF predictions. The proposed framework provides a general mechanism that allows incompatible model families to leverage each other's strengths through bidirectional adaptation.

Yunshuo Tian, Akayou Kitessa, Tanuja Chitnis +1

Department of Computer and Information Science, Fordham University, New York, NY · Department of Neurology, Mass General Brigham, Boston, MA

RL for Language Models

Mar 24, 2026

SortedRL: Accelerating RL Training for LLMs through Online Length-Aware Scheduling

Scaling reinforcement learning (RL) has shown strong promise for enhancing the reasoning abilities of large language models (LLMs), particularly in tasks requiring long chain-of-thought generation. However, RL training efficiency is often bottlenecked by the rollout phase, which can account for up to 70% of total training time when generating long trajectories (e.g., 16k tokens), due to slow autoregressive generation and synchronization overhead between rollout and policy updates. We propose SortedRL, an online length-aware scheduling strategy designed to address this bottleneck by improving rollout efficiency and maintaining training stability. SortedRL reorders rollout samples based on output lengths, prioritizing short samples forming groups for early updates. This enables large rollout batches, flexible update batches, and near on-policy micro-curriculum construction simultaneously. To further accelerate the pipeline, SortedRL incorporates a mechanism to control the degree of off-policy training through a cache-based mechanism, and is supported by a dedicated RL infrastructure that manages rollout and update via a stateful controller and rollout buffer. Experiments using LLaMA-3.1-8B and Qwen-2.5-32B on diverse tasks, including logical puzzles, and math challenges like AIME 24, Math 500, and Minerval, show that SortedRL reduces RL training bubble ratios by over 50%, while attaining 3.9% to 18.4% superior performance over baseline given same amount of data.

Yiqi Zhang, Huiqiang Jiang, Xufang Luo +7

National University of Singapore · Microsoft Research Asia · The Chinese University of Hong Kong

RL for Language Models RL for Language Model Reasoning RL Post-Training

Mar 23, 2026

TAMTRL: Teacher-Aligned Reward Reshaping for Multi-Turn Reinforcement Learning in Long-Context Compression

The rapid progress of large language models (LLMs) has led to remarkable performance gains across a wide range of tasks. However, when handling long documents that exceed the model's context window limit, the entire context cannot be processed in a single pass, making chunk-wise processing necessary. This requires multiple turns to read different chunks and update memory. However, supervision is typically provided only by the final outcome, which makes it difficult to evaluate the quality of memory updates at each turn in the multi-turn training setting. This introduces a temporal credit assignment challenge. Existing approaches, such as LLM-as-a-judge or process reward models, incur substantial computational overhead and suffer from estimation noise. To better address the credit assignment problem in multi-turn memory training, we propose Teacher-Aligned Reward Reshaping for Multi-Turn Reinforcement Learning (TAMTRL). TAMTRL leverages relevant documents as teacher signals by aligning them with each turn of model input and assigns rewards through normalized probabilities in a self-supervised manner. This provides fine-grained learning signals for each memory update and improves long-context processing. Experiments with multiple models of varying scales across seven long-context benchmarks show that TAMTRL consistently outperforms strong baselines, demonstrating its effectiveness. Our code is available at https://anonymous.4open.science/r/TAMTRL-F1F8.

Li Wang, Yandong Wang, Xin Yu +3

School of Artificial Intelligence, Beihang University, Beijing, 100191, China · Hangzhou International Innovation Institute, Beihang University, Hangzhou, China · Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing, Beihang University, Beijing, China

RL for Language Models Long-Context Language Modeling Reward Shaping Memory-Augmented Language Models

Mar 21, 2026

Implicit Turn-Wise Policy Optimization for Proactive User-LLM Interaction

Multi-turn human-AI collaboration is fundamental to deploying interactive services such as adaptive tutoring, conversational recommendation, and professional consultation. However, optimizing these interactions via reinforcement learning is hindered by the sparsity of verifiable intermediate rewards and the high stochasticity of user responses. To address these challenges, we introduce Implicit Turn-wise Policy Optimization (ITPO). ITPO leverages an implicit process reward model to derive fine-grained, turn-wise process rewards from sparse outcome signals. Unlike volatile token-level rewards, these turn-level signals exhibit superior robustness and may utilize a normalization mechanism to further enhance training stability. We evaluate ITPO across three representative multi-turn collaborative tasks: math tutoring, document writing, and medical recommendation. Empirical results demonstrate that ITPO, when combined with PPO, GRPO, or RLOO, consistently achieves improved convergence than existing baselines. Elaborate trajectory analysis confirms that ITPO infers turn-wise preferences that are semantically aligned with human judgment. Code is publicly available at https://github.com/Graph-COM/ITPO.

Haoyu Wang, Yuxin Chen, Liang Luo +3

Georgia Institute of Technology · Meta AI

RL for Language Models Process Reward Models Reward Modeling

Mar 20, 2026

FIPO: Eliciting Deep Reasoning with Future-KL Influenced Policy Optimization

We present Future-KL Influenced Policy Optimization (FIPO), a reinforcement learning algorithm designed to overcome reasoning bottlenecks in large language models. While GRPO style training scales effectively, it typically relies on outcome-based rewards (ORM) that distribute a global advantage uniformly across every token in a trajectory. We argue that this coarse-grained credit assignment imposes a performance ceiling by failing to distinguish critical logical pivots from trivial tokens. FIPO addresses this by incorporating discounted future-KL divergence into the policy update, creating a dense advantage formulation that re-weights tokens based on their influence on subsequent trajectory behavior. Empirically, FIPO enables models to break through the length stagnation seen in standard baselines. Evaluated on Qwen2.5-32B, FIPO extends the average chain-of-thought length from roughly 4,000 to over 10,000 tokens and increases AIME 2024 Pass@1 accuracy from 50.0% to a peak of 58.0% (converging at approximately 56.0%). This outperforms both DeepSeek-R1-Zero-Math-32B (around 47.0%) and o1-mini (approximately 56.0%). Our results suggest that establishing dense advantage formulations is a vital path for evolving ORM-based algorithms to unlock the full reasoning potential of base models. We open-source our training system, built on the verl framework.

Chiyu Ma, Shuo Yang, Kexin Huang +7

Qwen Pilot Team, Alibaba Group

RL for Language Model Reasoning RL for Language Models LLM Reasoning

Mar 20, 2026

LoopRPT: Reinforcement Pre-Training for Looped Language Models

Looped language models (LoopLMs) perform iterative latent computation to refine internal representations, offering an alternative to explicit chain-of-thought (CoT) reasoning. However, existing reinforcement learning (RL) paradigms primarily target output tokens, creating a structural mismatch with looped architectures whose reasoning unfolds implicitly. In this work, we propose LoopRPT, a reinforcement pre-training framework tailored for LoopLMs. By reframing next-token prediction as a next-token reasoning task, LoopRPT assigns reinforcement signals directly to latent steps using an EMA teacher reference and noisy latent rollouts. This formulation enables RL to directly shape intermediate representations, compressing effective reasoning into fewer iterations. We instantiate LoopRPT on Ouro across multiple model scales. Results show that LoopRPT improves per-step representation quality and achieves Pareto dominance in accuracy-computation trade-offs. Large gains on hard tokens indicate that LoopRPT enhances early-stage reasoning rather than merely encouraging premature exits. Our findings highlight reinforcement pre-training as a principled paradigm for learning efficient latent reasoning in LoopLMs.

Guo Tang, Shixin Jiang, Heng Chang +7

Harbin Institute of Technology, Harbin, China · Tsinghua University, Beijing, China · The Hong Kong University of Science and Technology, Guangzhou, China

Efficient Language Model Reasoning Implicit Reasoning in Language Models RL for Language Models Recursive Language Models

Mar 19, 2026

Adaptive Layerwise Perturbation: Unifying Off-Policy Corrections for LLM RL

Off-policy problems such as policy staleness and training--inference mismatch have become a major bottleneck for training stability and further exploration in LLM RL. The distribution gap between the inference and updated policies grows because of the techniques to enhance inference efficiency, leading to heavy-tailed importance ratios. Heavy-tailed ratios arise when the policy is locally sharp, which further inflates gradients and can push updates outside the trust region. To address this, we propose Adaptive Layerwise Perturbation (ALP), which injects small learnable perturbations into the input hidden states of each layer during updates and uses the resulting perturbed policy as the numerator of the importance ratio against the unchanged inference policy in the objective. Intuitively, by adding controlled noise to intermediate representations, ALP prevents the updated policy from deviating too sharply from the inference policy and enlarges the policy family to cover inference-time mismatch noise. Hence, the flattened distribution can naturally tighten the gap between the updated and inference policies and reduce the tail of importance ratios, thus maintaining training stability. This is further validated empirically. Experiments on single-turn math and multi-turn tool-integrated reasoning tasks show that ALP not only improves final performance, but also avoids blow-up in the importance-ratio tail and KL spikes during iterative training, along with boosted exploration. Ablations show that representation-level perturbations across all layers are most effective, substantially outperforming partial-layer and logits-only variants.

Chenlu Ye, Xuanchang Zhang, Yifan Hao +6

University of Illinois Urbana-Champaign · Amazon

Importance Sampling for RL RL for Language Models Reinforcement Learning Off-Policy RL RL for Language Model Reasoning Policy Optimization

Mar 19, 2026

VEPO: Variable Entropy Policy Optimization for Low-Resource Language Foundation Models

Large language models frequently exhibit suboptimal performance on low resource languages, primarily due to inefficient subword segmentation and systemic training data imbalances. In this paper, we propose Variable Entropy Policy Optimization (VEPO), which leverages Reinforcement Learning with Verifiable Rewards to incorporate deterministic structural constraints into the policy alignment process. This framework ensures prescribed sequence length, robust format consistency, and rigorous linguistic well formedness, all enforced during training. Central to our approach is a variable entropy mechanism that enables the model to dynamically calibrate the equilibrium between literal fidelity and semantic naturalness by modulating the exploration exploitation manifold. By integrating entropy tempered advantage estimation with asymmetric clipping, VEPO sustains robust exploration while mitigating policy collapse. Empirical evaluations across 90 FLORES-200, COMET-22, chrF directions demonstrate that VEPO yields substantial improvements in both tokenization efficiency and translation quality, bridging the performance gap for underrepresented languages.

Chonghan Liu, Yimin Du, Qi An +8

1Qiyuan Tech.

RL for Language Models Low-Resource MT Multilingual Language Models Maximum Entropy RL

Mar 19, 2026

Learning to Self-Evolve

We introduce Learning to Self-Evolve (LSE), a reinforcement learning framework that trains large language models (LLMs) to improve their own contexts at test time. We situate LSE in the setting of test-time self-evolution, where a model iteratively refines its context from feedback on seen problems to perform better on new ones. Existing approaches rely entirely on the inherent reasoning ability of the model and never explicitly train it for this task. LSE reduces the multi-step evolution problem to a single-step RL objective, where each context edit is rewarded by the improvement in downstream performance. We pair this objective with a tree-guided evolution loop. On Text-to-SQL generation (BIRD) and general question answering (MMLU-Redux), a 4B-parameter model trained with LSE outperforms self-evolving policies powered by GPT-5 and Claude Sonnet 4.5, as well as prompt optimization methods including GEPA and TextGrad, and transfers to guide other models without additional training. Our results highlight the effectiveness of treating self-evolution as a learnable skill.

Xiaoyin Chen, Canwen Xu, Yite Wang +3

Mila – Quebec AI Institute · University of Montreal · Snowflake

Language Model Self-Improvement RL for Language Models

Mar 19, 2026

Adaptive Decoding via Test-Time Policy Learning for Self-Improving Generation

Decoding strategies largely determine the quality of Large Language Model (LLM) outputs, yet widely used heuristics such as greedy or fixed temperature/top-p decoding are static and often task-agnostic, leading to suboptimal or inconsistent generation quality across domains that demand stylistic or structural flexibility. We introduce a reinforcement learning-based decoder sampler that treats decoding as sequential decision-making and learns a lightweight policy to adjust sampling parameters at test-time while keeping LLM weights frozen. We evaluated summarization datasets including BookSum, arXiv, and WikiHow using Granite-3.3-2B and Qwen-2.5-0.5B. Our policy sampler consistently outperforms greedy and static baselines, achieving relative gains of up to +88% (BookSum, Granite) and +79% (WikiHow, Qwen). Reward ablations show that overlap-only objectives underperform compared to composite rewards, while structured shaping terms (length, coverage, repetition, completeness) enable stable and sustained improvements. These findings highlight reinforcement learning as a practical mechanism for test-time adaptation in decoding, enabling domain-aware and user-controllable generation without retraining large models.

Asmita Bhardwaj, Yuya Jeremy Ong, Eelaaf Zahid +1

UC San Diego San Diego, CA · Plastic Labs San Jose, CA · IBM Research Cambridge, MA +1

Language Model Decoding Test-Time Adaptation RL for Language Models

Mar 19, 2026

PowerFlow: Unlocking the Dual Nature of LLMs via Principled Distribution Matching

Unsupervised Reinforcement Learning from Internal Feedback (RLIF) has emerged as a promising paradigm for eliciting the latent capabilities of Large Language Models (LLMs) without external supervision. However, current methods rely on heuristic intrinsic rewards, which often lack a well-defined theoretical optimization target and are prone to degenerative biases. In this work, we introduce PowerFlow, a principled framework that reformulates unsupervised fine-tuning as a distribution matching problem. By casting GFlowNet as an amortized variational sampler for unnormalized densities, we propose a length-aware Trajectory-Balance objective that explicitly neutralizes the structural length biases inherent in autoregressive generation. By targeting

α

-power distributions, PowerFlow enables the directional elicitation of the dual nature of LLMs: sharpening the distribution (

α> 1

) to intensify logical reasoning, or flattening it (

α< 1

) to unlock expressive creativity. Extensive experiments demonstrate that PowerFlow consistently outperforms existing RLIF methods, matching or even exceeding supervised GRPO. Furthermore, by mitigating over-sharpening in aligned models, our approach achieves simultaneous gains in diversity and quality, shifting the Pareto frontier in creative tasks.

Ruishuo Chen, Yu Chen, Zhuoran Li +1

Institute for Interdisciplinary Information Sciences, Tsinghua University, Beijing, China

RL for Language Models Reinforcement Learning RL for Language Model Reasoning Text Generation Generative Flow Networks

Mar 18, 2026

Enhancing Reinforcement Learning Fine-Tuning with an Online Refiner

Constraints are essential for stabilizing reinforcement learning fine-tuning (RFT) and preventing degenerate outputs, yet they inherently conflict with the optimization objective because stronger constraints limit the ability of a fine-tuned model to discover better solutions. We propose dynamic constraints that resolve this tension by adapting to the evolving capabilities of the fine-tuned model based on the insight that constraints should only intervene when degenerate outputs occur. We implement this by using a reference model as an online refiner that takes the response from the fine-tuned model and generates a minimally corrected version which preserves correct content verbatim while fixing errors. A supervised fine-tuning loss then trains the fine-tuned model to produce the refined output. This mechanism yields a constraint that automatically strengthens or relaxes based on output quality. Experiments on dialogue and code generation show that dynamic constraints outperform both KL regularization and unconstrained baselines, achieving substantially higher task rewards while maintaining training stability.

Hao Ma, Zhiqiang Pu, Yang Liu +1

School of Artificial Intelligence, University of Chinese Academy of Sciences · Institute of Automation, Chinese Academy of Sciences

RL Fine-Tuning RL for Language Models Constrained RL RL Stability

Mar 17, 2026

REAL: Regression-Aware Reinforcement Learning for LLM-as-a-Judge

Large language models (LLMs) are increasingly deployed as automated evaluators that assign numeric scores to model outputs, a paradigm known as LLM-as-a-Judge. However, standard Reinforcement Learning (RL) methods typically rely on binary rewards (e.g., 0-1 accuracy), thereby ignoring the ordinal structure inherent in regression tasks; for instance, they fail to recognize that predicting 4 is significantly better than predicting 1 when the ground truth is 5. Conversely, existing regression-aware approaches are often confined to Supervised Fine-Tuning (SFT), limiting their ability to explore optimal reasoning paths. To bridge this gap, we propose REAL (REgression-Aware Reinforcement Learning), a principled RL framework designed to optimize regression rewards, and also proven to be optimal for correlation metrics. A key technical challenge is that the regression objective is explicitly policy-dependent, thus invalidating standard policy gradient methods. To address this, we employ the generalized policy gradient estimator, which naturally decomposes optimization into two complementary components: (1) exploration over Chain-of-Thought (CoT) trajectory, and (2) regression-aware prediction refinement of the final score. Extensive experiments across model scales (8B to 32B) demonstrate that REAL consistently outperforms both regression-aware SFT baselines and standard RL methods, exhibiting significantly better generalization on out-of-domain benchmarks. On Qwen3-32B specifically, we achieve gains of +8.40 Pearson and +7.20 Spearman correlation over the SFT baseline, and +18.30/+11.20 over the base model. These findings highlight the critical value of integrating regression objectives into RL exploration for accurate LLM evaluation.

Yasi Zhang, Tianyu Chen, Mingyuan Zhou +3

University of California, Los Angeles · The University of Texas at Austin · Google Research

RL for Language Models Reinforcement Learning LLM-as-a-Judge LLM Evaluation

Mar 17, 2026

HIPO: Instruction Hierarchy via Constrained Reinforcement Learning

Hierarchical Instruction Following (HIF) refers to the problem of prompting large language models with a priority-ordered stack of instructions. Standard methods like RLHF and DPO typically fail in this problem since they mainly optimize for a single objective, failing to explicitly enforce system prompt compliance. Meanwhile, supervised fine-tuning relies on mimicking filtered, compliant data, which fails to establish the priority asymmetry at the algorithmic level. In this paper, we introduce HIPO, a novel alignment framework that formulates HIF as a Constrained Markov Decision Process. HIPO elevates system prompts from mere input context to strict algorithmic boundaries. Using a primal-dual safe reinforcement learning approach, the algorithm dynamically enforces system prompt compliance as an explicit constraint, maximizing user utility strictly within this feasible region. Extensive evaluations across diverse model architectures (e.g., Qwen, Phi, Llama) demonstrate that HIPO significantly improves both system compliance and user utility. Furthermore, mechanistic analysis reveals that this constrained optimization autonomously drives the model to shift its attention toward long-range system tokens, providing a principled foundation for reliable LLM deployment in complex workflows.

Keru Chen, Jun Luo, Sen Lin +4

School of ECEE Arizona State University · Department of ECE The Ohio State University · Computer Science Department University of Houston +2

LLM Alignment RL for Language Models LLM Safety Alignment Constrained RL

Mar 16, 2026

Aligning Paralinguistic Understanding and Generation in Speech LLMs via Multi-Task Reinforcement Learning

Speech large language models (LLMs) observe paralinguistic cues such as prosody, emotion, and non-verbal sounds--crucial for intent understanding. However, leveraging these cues faces challenges: limited training data, annotation difficulty, and models exploiting lexical shortcuts over paralinguistic signals. We propose multi-task reinforcement learning (RL) with chain-of-thought prompting that elicits explicit affective reasoning. To address data scarcity, we introduce a paralinguistics-aware speech LLM (PALLM) that jointly optimizes sentiment classification from audio and paralinguistics-aware response generation via a two-stage pipeline. Experiments demonstrate that our approach improves paralinguistics understanding over both supervised baselines and strong proprietary models (Gemini-2.5-Pro, GPT-4o-audio) by 8-12% on Expresso, IEMOCAP, and RAVDESS. The results show that modeling paralinguistic reasoning with multi-task RL is crucial for building emotionally intelligent speech LLMs.

Jingxiang Chen, Minseok Kim, Seong-Gyun Leem +13

Meta Reality Labs

Speech Language Models Emotion Representation in Language Models Speech Emotion Recognition RL for Language Models

Mar 16, 2026

Writer-R1: Dynamic Rubric Mining for Reinforcement Learning of Open-Ended Writing

Training models for open-ended writing requires diverse prompts and sufficiently precise specifications of response quality. Constructing such specifications at scale entails recovering task requirements, consolidating overlapping criteria, and defining observable differences between performance levels. We introduce Writer-R1, a writing training pipeline built around Inductive Rubric Generation (IRG), which integrates these operations into an automated rubric construction procedure. The pipeline expands an open writing corpus and uses the resulting prompt-specific rubrics throughout supervised training and reinforcement learning (RL). Each rubric is concatenated with the user request and also defines the reward criteria, making the requirements used for assessment available to the writer before generation. A compact model, RubricGen-4B, learns to supply rubrics for new requests. Writer-R1 improves WritingBench scores by 0.182 and 0.158 over the 4B and 8B backbones, respectively, with gains also on HelloBench. Training ablations favor IRG over direct rubric generation, and further analyses support rubric transfer and the robustness of writer improvements across evaluators. These results support using inductively mined rubrics as both generation guidance and reward specifications for open-ended writing.

Jihao Zhao, Shuaishuai Zu, Zhiyuan Ji +2

School of Information, Renmin University of China, Beijing, China · School of Information, BRAIN, Renmin University of China, Beijing, China

Memory-Augmented RL RL for Language Models Language Model Generation Evaluation Rubric-Based RL Reinforcement Learning

Mar 13, 2026

Mending the Holes: Mitigating Reward Hacking in Reinforcement Learning for Multilingual Translation

Large Language Models (LLMs) have demonstrated remarkable capability in machine translation on high-resource language pairs, yet their performance on low-resource translation still lags behind. Existing post-training methods rely heavily on high-quality parallel data, which are often scarce or unavailable for low-resource languages. In this paper, we introduce WALAR, a reinforcement training method using only monolingual text to elevate LLMs' translation capabilities on massive low-resource languages while retaining their performance on high-resource languages. Our key insight is based on the observation of failure modes (or "holes") in existing source-based multilingual quality estimation (QE) models. Reinforcement learning (RL) using these QE models tends to amplify such holes, resulting in poorer multilingual LLMs. We develop techniques including word alignment and language alignment to mitigate such holes in WALAR's reward for RL training. We continually trained an LLM supporting translation of 101 languages using WALAR. The experiments show that our new model outperforms LLaMAX, one of the strongest open-source multilingual LLMs by a large margin on 1400 language directions on Flores-101 dataset.

Yifeng Liu, Siqi Ouyang, Yatish Hosmane Revanasiddappa +1

Carnegie Mellon University

Low-Resource MT Multilingual Language Models RL for Language Models Reward Hacking

Mar 11, 2026

Tackling Length Inflation Without Trade-offs: Group Relative Reward Rescaling for Reinforcement Learning

Reinforcement learning significantly enhances LLM capabilities but suffers from a critical issue: length inflation, where models adopt verbosity or inefficient reasoning to maximize rewards. Prior approaches struggle to address this challenge in a general and lossless manner, primarily because additive penalties introduce a compensatory effect that creates optimization shortcuts, while heuristic gating strategies lack generality beyond binary feedback. To bridge this gap, we present Group Relative Reward Rescaling (GR

^3

), which reframes length control as a multiplicative rescaling paradigm, effectively establishing a generalized, continuous, and reward-dependent gating mechanism. To further ensure lossless optimization, we incorporate group-relative regularization and advantage-aware calibration, which dynamically adapt length budgets to instance difficulty and preserve the advantage signal of high-quality trajectories. Empirically, across both RLHF and RLVR settings, GR

^3

~maintains training dynamics and downstream performance comparable to standard GRPO while significantly mitigating length inflation, outperforming state-of-the-art length-regularized baselines.

Zichao Li, Jie Lou, Fangchen Dong +8

Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences · University of Chinese Academy of Sciences · Xiaohongshu Inc.

RL for Language Models RL Post-Training Reward Shaping

Mar 10, 2026

Learning Adaptive LLM Decoding

Decoding from large language models (LLMs) typically relies on fixed sampling hyperparameters (e.g., temperature, top-p), despite substantial variation in task difficulty and uncertainty across prompts and individual decoding steps. We propose to learn adaptive decoding policies that dynamically select sampling strategies at inference time, conditioned on available compute resources. Rather than fine-tuning the language model itself, we introduce lightweight decoding adapters trained with reinforcement learning and verifiable terminal rewards (e.g. correctness on math and coding tasks). At the sequence level, we frame decoding as a contextual bandit problem: a policy selects a decoding strategy (e.g. greedy, top-k, min-p) for each prompt, conditioned on the prompt embedding and a parallel sampling budget. At the token level, we model decoding as a partially observable Markov decision process (POMDP), where a policy selects sampling actions at each token step based on internal model features and the remaining token budget. Experiments on the MATH and CodeContests benchmarks show that the learned adapters improve the accuracy-budget tradeoff: on MATH, the token-level adapter improves Pass@1 accuracy by up to 10.2% over the best static baseline under a fixed token budget, while the sequence-level adapter yields 2-3% gains under fixed parallel sampling. Ablation analyses support the contribution of both sequence- and token-level adaptation.

Chloe H. Su, Zhe Ye, Samuel Tenka +3

Department of Computer Science, Harvard University, Boston, USA · Kempner Institute, Harvard University, Cambridge, MA, USA · UC Berkeley, CA, USA +1

Language Model Decoding Inference-Time Adaptation of Language Models RL for Language Models

Mar 9, 2026

MAPLE: Elevating Medical Reasoning from Statistical Consensus to Process-Led Alignment

Recent advances in medical large language models have explored Test-Time Reinforcement Learning (TTRL) to enhance reasoning. However, standard TTRL often relies on majority voting (MV) as a heuristic supervision signal, which can be unreliable in complex medical scenarios where the most frequent reasoning path is not necessarily the clinically correct one. In this work, we propose a novel and unified training paradigm that integrates medical process reward models with TTRL to bridge the gap between test-time scaling (TTS) and parametric model optimization. Specifically, we advance the TTRL framework by replacing the conventional MV with a fine-grained, expert-aligned supervision paradigm using Med-RPM. This integration ensures that reinforcement learning is guided by medical correctness rather than mere consensus, effectively distilling search-based intelligence into the model's parametric memory. Extensive evaluations on four different benchmarks have demonstrated that our developed method consistently and significantly outperforms current TTRL and standalone PRM selection. Our findings establish that transitioning from stochastic heuristics to structured, step-wise rewards is essential for developing reliable and scalable medical AI systems

Kailong Fan, Anqi Pu, Yichen Wu +7

Zhejiang University · Harvard Medical School · Harvard University

Medical LLMs RL for Language Models Process Reward Models

Mar 9, 2026

How Far Can Unsupervised RLVR Scale LLM Training?

Unsupervised reinforcement learning with verifiable rewards (URLVR) offers a pathway to scale LLM training beyond the supervision bottleneck by deriving rewards without ground truth labels. Recent works leverage model intrinsic signals, showing promising early gains, yet their potential and limitations remain unclear. In this work, we revisit URLVR and provide a comprehensive analysis spanning taxonomy, theory and extensive experiments. We first classify URLVR methods into intrinsic versus external based on reward sources, then establish a unified theoretical framework revealing that all intrinsic methods converge toward sharpening the model's initial distribution This sharpening mechanism succeeds when initial confidence aligns with correctness but fails catastrophically when misaligned. Through systematic experiments, we show intrinsic rewards consistently follow a rise-then-fall pattern across methods, with collapse timing determined by model prior rather than engineering choices. Despite these scaling limits, we find intrinsic rewards remain valuable in test-time training on small datasets, and propose Model Collapse Step to measure model prior, serving as a practical indicator for RL trainability. Finally, we explore external reward methods that ground verification in computational asymmetries, showing preliminary evidence they may escape the confidence-correctness ceiling. Our findings chart boundaries for intrinsic URLVR while motivating paths toward scalable alternatives.

Bingxiang He, Yuxin Zuo, Zeyuan Liu +18

Tsinghua University · Shanghai AI Lab · Xi’an Jiaotong University +3

RL for Language Models Reinforcement Learning with Verifiable Rewards Intrinsic Reward Methods