Near-Future Policy Optimization

Reinforcement learning with verifiable rewards (RLVR) has become a core post-training recipe. Introducing suitable off-policy trajectories into on-policy exploration accelerates RLVR convergence and raises the performance ceiling, yet finding a source of such trajectories remains the key challenge. Existing mixed-policy methods either import trajectories from external teachers (high-quality but distributionally far) or replay past training trajectories (close but capped in quality), and neither simultaneously satisfies the strong enough (higher $Q$ , more new knowledge to learn) and close enough (lower $V$ , more readily absorbed) conditions required to maximize the effective learning signal $\mathcal{S} = Q/V$. We propose \textbf{N}ear-Future \textbf{P}olicy \textbf{O}ptimization (\textbf{NPO}), a simple mixed-policy scheme that learns from a policy's own near-future self: a later checkpoint from the same training run is a natural source of auxiliary trajectories that is both stronger than the current policy and closer than any external source, directly balancing trajectory quality against variance cost. We validate NPO through two manual interventions, early-stage bootstrapping and late-stage plateau breakthrough, and further propose \textbf{AutoNPO},an adaptive variant that automatically triggers interventions from online training signals and selects the guide checkpoint that maximizes $S$. On Qwen3-VL-8B-Instruct with GRPO, NPO improves average performance from 57.88 to 62.84, and AutoNPO pushes it to 63.15, raising the final performance ceiling while accelerating convergence.

Self-Distilled RLVR

On-policy distillation (OPD) has become a popular training paradigm in the LLM community. This paradigm selects a larger model as the teacher to provide dense, fine-grained signals for each sampled trajectory, in contrast to reinforcement learning with verifiable rewards (RLVR), which only obtains sparse signals from verifiable outcomes in the environment. Recently, the community has explored on-policy self-distillation (OPSD), where the same model serves as both teacher and student, with the teacher receiving additional privileged information such as reference answers to enable self-evolution. This paper demonstrates that learning signals solely derived from the privileged teacher result in severe information leakage and unstable long-term training. Accordingly, we identify the optimal niche for self-distillation and propose \textbf{RLSD} (\textbf{RL}VR with \textbf{S}elf-\textbf{D}istillation). Specifically, we leverage self-distillation to obtain token-level policy differences for determining fine-grained update magnitudes, while continuing to use RLVR to derive reliable update directions from environmental feedback (e.g., response correctness). This enables RLSD to simultaneously harness the strengths of both RLVR and OPSD, achieving a higher convergence ceiling and superior training stability.

Diagnosing and Repairing Unsafe Channels in Vision-Language Models via Causal Discovery and Dual-Modal Safety Subspace Projection

Large Vision-Language Models (LVLMs) have achieved impressive performance across multimodal understanding and reasoning tasks, yet their internal safety mechanisms remain opaque and poorly controlled. In this work, we present a comprehensive framework for diagnosing and repairing unsafe channels within LVLMs (CARE). We first perform causal mediation analysis to identify neurons and layers that are causally responsible for unsafe behaviors. Based on these findings, we introduce a dual-modal safety subspace projection method that learns generalized safety subspaces for both visual and textual modalities through generalized eigen-decomposition between benign and malicious activations. During inference, activations are dynamically projected toward these safety subspaces via a hybrid fusion mechanism that adaptively balances visual and textual corrections, effectively suppressing unsafe features while preserving semantic fidelity. Extensive experiments on multiple safety benchmarks demonstrate that our causal-subspace repair framework significantly enhances safety robustness without degrading general multimodal capabilities, outperforming prior activation steering and alignment-based baselines. Additionally, our method exhibits good transferability, defending against unseen attacks.

A Closer Look into LLMs for Table Understanding

Despite the success of Large Language Models (LLMs) in table understanding, their internal mechanisms remain unclear. In this paper, we conduct an empirical study on 16 LLMs, covering general LLMs, specialist tabular LLMs, and Mixture-of-Experts (MoE) models, to explore how LLMs understand tabular data and perform downstream tasks. Our analysis focus on 4 dimensions including the attention dynamics, the effective layer depth, the expert activation, and the impacts of input designs. Key findings include: (1) LLMs follow a three-phase attention pattern -- early layers scan the table broadly, middle layers localize relevant cells, and late layers amplify their contributions; (2) tabular tasks require deeper layers than math reasoning to reach stable predictions; (3) MoE models activate table-specific experts in middle layers, with early and late layers sharing general-purpose experts; (4) Chain-of-Thought prompting increases table attention, further enhanced by table-tuning. We hope these findings and insights can facilitate interpretability and future research on table-related tasks.

Beyond the Covariance Trap: Unlocking Generalization in Same-Subject Knowledge Editing for Large Language Models

While locate-then-edit knowledge editing efficiently updates knowledge encoded within Large Language Models (LLMs), a critical generalization failure mode emerges in the practical same-subject knowledge editing scenario: models fail to recall the updated knowledge when following user instructions, despite successfully recalling it in the original edited form. This paper identifies the geometric root of this generalization collapse as a fundamental conflict where the inner activation drifts induced by prompt variations exceed the model's geometric tolerance for generalization after editing. We attribute this instability to a dual pathology: (1) The joint optimization with orthogonal gradients collapses solutions into sharp minima with narrow stability, and (2) the standard covariance constraint paradoxically acts as a Covariance Trap that amplifies input perturbations. To resolve this, we introduce RoSE (Robust Same-subject Editing), which employs Isotropic Geometric Alignment to minimize representational deviation and Hierarchical Knowledge Integration to smooth the optimization landscape. Extensive experiments demonstrate that RoSE significantly improves instruction-following capabilities, laying the foundation for robust interactive parametric memory of LLM agents.

HiMemVLN: Enhancing Reliability of Open-Source Zero-Shot Vision-and-Language Navigation with Hierarchical Memory System

LLM-based agents have demonstrated impressive zero-shot performance in vision-language navigation (VLN) tasks. However, most zero-shot methods primarily rely on closed-source LLMs as navigators, which face challenges related to high token costs and potential data leakage risks. Recent efforts have attempted to address this by using open-source LLMs combined with a spatiotemporal CoT framework, but they still fall far short compared to closed-source models. In this work, we identify a critical issue, Navigation Amnesia, through a detailed analysis of the navigation process. This issue leads to navigation failures and amplifies the gap between open-source and closed-source methods. To address this, we propose HiMemVLN, which incorporates a Hierarchical Memory System into a multimodal large model to enhance visual perception recall and long-term localization, mitigating the amnesia issue and improving the agent's navigation performance. Extensive experiments in both simulated and real-world environments demonstrate that HiMemVLN achieves nearly twice the performance of the open-source state-of-the-art method. The code is available at https://github.com/lvkailin0118/HiMemVLN.

System 1&2 Synergy via Dynamic Model Interpolation

Training a unified language model that adapts between intuitive System 1 and deliberative System 2 remains challenging due to interference between their cognitive modes. Recent studies have thus pursued making System 2 models more efficient. However, these approaches focused on output control, limiting what models produce. We argue that this paradigm is misaligned: output length is merely a symptom of the model's cognitive configuration, not the root cause. In this work, we shift the focus to capability control, which modulates \textit{how models think} rather than \textit{what they produce}. To realize this, we leverage existing Instruct and Thinking checkpoints through dynamic parameter interpolation, without additional training. Our pilot study establishes that linear interpolation yields a convex, monotonic Pareto frontier, underpinned by representation continuity and structural connectivity. Building on this, we propose \textbf{DAMI} (\textbf{D}yn\textbf{A}mic \textbf{M}odel \textbf{I}nterpolation), a framework that estimates a query-specific Reasoning Intensity $λ(q)$ to configure cognitive depth. For training-based estimation, we develop a preference learning method encoding accuracy and efficiency criteria. For zero-shot deployment, we introduce a confidence-based method leveraging inter-model cognitive discrepancy. Experiments on five mathematical reasoning benchmarks demonstrate that DAMI achieves higher accuracy than the Thinking model while remaining efficient, effectively combining the efficiency of System 1 with the reasoning depth of System 2.

Outcome-Grounded Advantage Reshaping for Fine-Grained Credit Assignment in Mathematical Reasoning

Group Relative Policy Optimization (GRPO) has emerged as a promising critic-free reinforcement learning paradigm for reasoning tasks. However, standard GRPO employs a coarse-grained credit assignment mechanism that propagates group-level rewards uniformly to to every token in a sequence, neglecting the varying contribution of individual reasoning steps. We address this limitation by introducing Outcome-grounded Advantage Reshaping (OAR), a fine-grained credit assignment mechanism that redistributes advantages based on how much each token influences the model's final answer. We instantiate OAR via two complementary strategies: (1) OAR-P, which estimates outcome sensitivity through counterfactual token perturbations, serving as a high-fidelity attribution signal; (2) OAR-G, which uses an input-gradient sensitivity proxy to approximate the influence signal with a single backward pass. These importance signals are integrated with a conservative Bi-Level advantage reshaping scheme that suppresses low-impact tokens and boosts pivotal ones while preserving the overall advantage mass. Empirical results on extensive mathematical reasoning benchmarks demonstrate that while OAR-P sets the performance upper bound, OAR-G achieves comparable gains with negligible computational overhead, both significantly outperforming a strong GRPO baseline, pushing the boundaries of critic-free LLM reasoning.

Breaking the Trade-Off Between Faithfulness and Expressiveness for Large Language Models

Grounding responses in external knowledge represents an effective strategy for mitigating hallucinations in Large Language Models (LLMs). However, current LLMs struggle to seamlessly integrate knowledge while simultaneously maintaining faithfulness (or fidelity) and expressiveness, capabilities that humans naturally possess. This limitation results in outputs that either lack support from external knowledge, thereby compromising faithfulness, or appear overly verbose and unnatural, thus sacrificing expressiveness. In this work, to break the trade-off between faithfulness and expressiveness, we propose Collaborative Decoding (CoDe), a novel approach that dynamically integrates output probabilities generated with and without external knowledge. This integration is guided by distribution divergence and model confidence, enabling the selective activation of relevant and reliable expressions from the model's internal parameters. Furthermore, we introduce a knowledge-aware reranking mechanism that prevents over-reliance on prior parametric knowledge while ensuring proper utilization of provided external information. Through comprehensive experiments, our plug-and-play CoDe framework demonstrates superior performance in enhancing faithfulness without compromising expressiveness across diverse LLMs and evaluation metrics, validating both its effectiveness and generalizability.

Stable Reinforcement Learning for Efficient Reasoning

The success of Deepseek-R1 has drawn the LLM community's attention to reinforcement learning (RL) methods like GRPO. However, such rule-based 0/1 outcome reward methods lack the capability to regulate the intermediate reasoning processes during chain-of-thought (CoT) generation, leading to severe overthinking phenomena. In response, recent studies have designed reward functions to reinforce models' behaviors in producing shorter yet correct completions. Nevertheless, we observe that these length-penalty reward functions exacerbate RL training instability: as the completion length decreases, model accuracy abruptly collapses, often occurring early in training. To address this issue, we propose a simple yet effective solution GRPO-$\lambda$, an efficient and stabilized variant of GRPO, which dynamically adjusts the reward strategy by monitoring the correctness ratio among completions within each query-sampled group. A low correctness ratio indicates the need to avoid length penalty that compromises CoT quality, triggering a switch to length-agnostic 0/1 rewards that prioritize reasoning capability. A high ratio maintains length penalties to boost efficiency. Experimental results show that our approach avoids training instability caused by length penalty while maintaining the optimal accuracy-efficiency trade-off. On the GSM8K, GPQA, MATH-500, AMC 2023, and AIME 2024 benchmarks, it improves average accuracy by 1.48% while reducing CoT sequence length by 47.3%.