Min Zhang

Xinyu Ma, Mingzhou Xu, Xuebo Liu et al.

Recent advancements in Reinforcement Learning with Verifiable Rewards (RLVR) have significantly improved Large Language Model (LLM) reasoning, yet models often struggle to explore novel trajectories beyond their initial latent space. While offline teacher guidance and entropy-driven strategies have been proposed to address this, they often lack deep integration or are constrained by the model's inherent capacity. In this paper, we propose OGER, a novel framework that unifies offline teacher guidance and online reinforcement learning through a specialized reward modeling lens. OGER employs multi-teacher collaborative training and constructs an auxiliary exploration reward that leverages both offline trajectories and the model's own entropy to incentivize autonomous exploration. Extensive experiments across mathematical and general reasoning benchmarks demonstrate that OGER significantly outperforms competitive baselines, achieving substantial gains in mathematical reasoning while maintaining robust generalization to out-of-domain tasks. We provide a comprehensive analysis of training dynamics and conduct detailed ablation studies to validate the effectiveness of our entropy-aware reward modulation. Our code is available at https://github.com/ecoli-hit/OGER.git.

Junlin Li, Shuangyong Song, Guodong Du et al.

Supervised Fine-Tuning (SFT) accelerates taskspecific large language models (LLMs) development, but the resulting proliferation of finetuned models incurs substantial memory overhead. Delta compression addresses this by retaining a single pre-trained LLM with multiple compressed delta weights. However, existing methods fail on models fine-tuned with largescale datasets. We find that larger SFT data scale amplifies delta parameter magnitude, singular values, and entropy, exacerbating compression errors. To tackle this, we propose DQRELO (Delta Compression via Quantization and Residual Low-Rank), a novel training- and data-free delta compression method. It combines coarse-grained one-bit quantization to capture the dominant structure of the delta, followed by compensated residual low-rank approximation to recover fine-grained details from the smaller residual error. Experiments on various LLMs spanning dense and MoE architectures across multiple domains under this challenging setting demonstrate that DQRELO outperforms existing methods. Moreover, we establish key design principles for delta compression through extensive empirical analysis, demonstrating how task difficulty, architecture, and layer positioning create predictable patterns that can guide optimal compression strategies in production systems.

Yu Zhang, Chuyang Sun, Kehai Chen et al.

Large Vision-Language Models (LVLMs) still struggle with vision hallucination, where generated responses are inconsistent with the visual input. Existing methods either rely on large-scale annotated data for fine-tuning, which incurs massive computational overhead, or employ static post-hoc strategies that overlook the dynamic nature of hallucination emergence. To address these, we introduce a new self-rewarding framework, enabling dynamic hallucination mitigation at inference time without external supervision. On the empirical side, we reveal that visual hallucination exhibits phase-wise dynamic patterns, peaking at the onset of each semantic phase. Drawing on these insights, we propose \textbf{PSRD} (\textbf{Phase-wise \textbf{S}elf-\textbf{R}eward \textbf{D}ecoding) for online hallucination correction guided by phase-wise self-reward signals. To reduce the cost of repeated self-evaluation during decoding, we distill the hallucination guidance signal from LVLMs into a lightweight reward model. The reward model subsequently provides on-the-fly guidance for targeted intervention during the decoding process, enabling precise hallucination suppression. The proposed PSRD significantly reduces the hallucination rate of LLaVA-1.5-7B by 50.0% and consistently outperforms existing post-hoc methods across five hallucination evaluation benchmarks for four LVLMs. Further analysis confirms that PSRD effectively mitigates hallucination propagation and achieves a highly controllable trade-off between strong performance and inference efficiency.

Benteng Chen, Weida Wang, Shufei Zhang et al.

Large reasoning models that use long chain-of-thought excel at problem-solving yet waste compute on redundant checks. Curbing this overthinking is hard: training-time length penalties can cripple ability, while inference-time early-exit adds system overhead. To bridge this gap, we propose Step-GRPO, a novel post-training framework that internalizes dynamic early-exit capabilities directly into the model. Step-GRPO shifts the optimization objective from raw tokens to semantic steps by utilizing linguistic markers to structure reasoning. We introduce a Dynamic Truncated Rollout mechanism that exposes the model to concise high-confidence trajectories during exploration, synergized with a Step-Aware Relative Reward that dynamically penalizes redundancy based on group-level baselines. Extensive experiments across three model sizes on diverse benchmarks demonstrate that Step-GRPO achieves a superior accuracy-efficiency trade-off. On Qwen3-8B, our method reduces token consumption by 32.0\% compared to the vanilla model while avoiding the accuracy degradation observed in traditional length-penalty methods.

Zhenyu Liu, Yunxin Li, Baotian Hu et al.

To improve Multimodal Large Language Models' (MLLMs) ability to process images and complex instructions, researchers predominantly curate large-scale visual instruction tuning datasets, which are either sourced from existing vision tasks or synthetically generated using LLMs and image descriptions. However, they often suffer from critical flaws, including misaligned instruction-image pairs and low-quality images. Such issues hinder training efficiency and limit performance improvements, as models waste resources on noisy or irrelevant data with minimal benefit to overall capability. To address this issue, we propose a \textbf{Vi}sual-Centric \textbf{S}election approach via \textbf{A}gents Collaboration (ViSA), which centers on image quality assessment and image-instruction relevance evaluation. Specifically, our approach consists of 1) an image information quantification method via visual agents collaboration to select images with rich visual information, and 2) a visual-centric instruction quality assessment method to select high-quality instruction data related to high-quality images. Finally, we reorganize 80K instruction data from large open-source datasets. Extensive experiments demonstrate that ViSA outperforms or is comparable to current state-of-the-art models on seven benchmarks, using only 2.5\% of the original data, highlighting the efficiency of our data selection approach. Moreover, we conduct ablation studies to validate the effectiveness of each component of our method. The code is available at https://github.com/HITsz-TMG/ViSA.

Tao Wu, Jingyuan Chen, Wang Lin et al.

Distractors, incorrect but plausible answer choices in multiple-choice questions (MCQs), play a critical role in educational assessment by diagnosing student misconceptions. Recent work has leveraged large language models (LLMs) to generate shared, group-level distractors by learning common error patterns across large student populations. However, such distractors often fail to capture the diverse reasoning errors of individual students, limiting their diagnostic effectiveness. To address this limitation, we introduce the task of personalized distractor generation, which aims to generate tailored distractors based on individual misconceptions inferred from each student's past question-answering (QA) records, ensuring every student receives options that effectively exposes their specific reasoning errors. While promising, this task is challenging because each student typically has only a few QA records, which often lack the student's underlying reasoning processes, making training-based group-level approaches infeasible. To overcome this, we propose a training-free two-stage framework. In the first stage, we construct a student-specific misconception prototype by applying Monte Carlo Tree Search (MCTS) to recover the student's reasoning trajectories from past incorrect answers. In the second stage, this prototype guides the simulation of the student's reasoning on new questions, enabling the generation of personalized distractors that align with the student's recurring misconceptions. Experiments show that our approach achieves the best performance in generating plausible, personalized distractors for 140 students, and also effectively generalizes to group-level settings, highlighting its robustness and adaptability.

Yichen Tang, Weihang Su, Yujia Zhou et al.

Multi-agent techniques such as role playing or multi-turn debates have been shown to be effective in improving the performance of large language models (LLMs) in downstream tasks. Despite their differences in workflows, existing multi-agent systems constructed from a single base LLM mostly use natural language for agent communication. While this is appealing for its simplicity and interpretability, it also introduces inevitable information loss as one model must down sample its continuous state vectors to discrete tokens before transferring them to the other model. Such losses are particularly significant when the information to transfer is not simple facts, but reasoning logics or abstractive thoughts. To tackle this problem, we propose a new communication protocol that transfers both natural language tokens and token-wise state transition trajectory from one agent to another. Particularly, compared to the actual state value, we find that the sequence of state changes in LLMs after generating each token can better reflect the information hidden behind the inference process. We propose a State Delta Encoding (SDE) method to represent state transition trajectories. The experimental results show that multi-agent systems with SDE achieve SOTA performance compared to other communication protocols, particularly in tasks that involve complex reasoning.