Knowledge editing aims to efficiently modify the internal knowledge of large language models (LLMs) without compromising their other capabilities. The prevailing editing paradigm, which appends an update matrix to the original parameter matrix, has been shown by some studies to damage key numerical stability indicators (such as condition number and norm), thereby reducing editing performance and general abilities, especially in sequential editing scenario. Although subsequent methods have made some improvements, they remain within the additive framework and have not fundamentally addressed this limitation. To solve this problem, we analyze it from both statistical and mathematical perspectives and conclude that multiplying the original matrix by an orthogonal matrix does not change the numerical stability of the matrix. Inspired by this, different from the previous additive editing paradigm, a multiplicative editing paradigm termed Multiplicative Orthogonal Sequential Editing (MOSE) is proposed. Specifically, we first derive the matrix update in the multiplicative form, the new knowledge is then incorporated into an orthogonal matrix, which is multiplied by the original parameter matrix. In this way, the numerical stability of the edited matrix is unchanged, thereby maintaining editing performance and general abilities. We compared MOSE with several current knowledge editing methods, systematically evaluating their impact on both editing performance and the general abilities across three different LLMs. Experimental results show that MOSE effectively limits deviations in the edited parameter matrix and maintains its numerical stability. Compared to current methods, MOSE achieves a 12.08% improvement in sequential editing performance, while retaining 95.73% of general abilities across downstream tasks.
论文检索
输入标题、作者或关键词，从 100,903 篇学术成果中精准定位
Information retrieval (IR) systems play a critical role in navigating information overload across various applications. Existing IR benchmarks primarily focus on simple queries that are semantically analogous to single- and multi-hop relations, overlooking complex logical queries involving first-order logic operations such as conjunction (∧), disjunction (∨), and negation (¬). Thus, these benchmarks can not be used to sufficiently evaluate the performance of IR models on complex queries in real-world scenarios. To address this problem, we propose a novel method leveraging large language models (LLMs) to construct a new IR dataset ComLQ for Complex Logical Queries, which comprises 2,909 queries and 11,251 candidate passages. A key challenge in constructing the dataset lies in capturing the underlying logical structures within unstructured text. Therefore, by designing the subgraph-guided prompt with the subgraph indicator, an LLM (such as GPT-4o) is guided to generate queries with specific logical structures based on selected passages. All query-passage pairs in ComLQ are ensured structure conformity and evidence distribution through expert annotation. To better evaluate whether retrievers can handle queries with negation, we further propose a new evaluation metric, Log-Scaled Negation Consistency (LSNC@K). As a supplement to standard relevance-based metrics (such as nDCG and mAP), LSNC@K measures whether top-K retrieved passages violate negation conditions in queries. Our experimental results under zero-shot settings demonstrate existing retrieval models' limited performance on complex logical queries, especially on queries with negation, exposing their inferior capabilities of modeling exclusion. In summary, our ComLQ offers a comprehensive and fine-grained exploration, paving the way for future research on complex logical queries in IR.
MedMKEB: A Comprehensive Knowledge Editing Benchmark for Medical Multimodal Large Language Models
PDF ↗Recent advances in multimodal large language models (MLLMs) have significantly improved medical AI, enabling it to unify the understanding of visual and textual information. However, as medical knowledge continues to evolve, it is critical to allow these models to efficiently update outdated or incorrect information without retraining from scratch. Although textual knowledge editing has been widely studied, there is still a lack of systematic benchmarks for multimodal medical knowledge editing involving image and text modalities. To fill this gap, we present MedMKEB, the first comprehensive benchmark designed to evaluate the reliability, generality, locality, portability, and robustness of knowledge editing in medical multimodal large language models. MedMKEB is built on a high-quality medical visual question-answering dataset and enriched with carefully constructed editing tasks, including counterfactual correction, semantic generalization, knowledge transfer, and adversarial robustness. We incorporate human expert validation to ensure the accuracy and reliability of the benchmark. Extensive experiments on state-of-the-art general and medical MLLMs demonstrate the limitations of existing knowledge editing methods in the medical domain, highlighting the need to develop specialized editing strategies.
Large Language Models (LLMs) are prone to generating incorrect or outdated information, thereby necessitating efficient and precise mechanisms for knowledge updates. Existing knowledge editing approaches, however, often encounter conflicts between two competing objectives: maintaining existing knowledge (preservation) and incorporating new information (editing). During gradient-based optimization, these conflicting objectives can lead to imbalanced update directions, where one gradient dominates, ultimately resulting in suboptimal learning dynamics. To address this challenge, we propose a balanced knowledge editing framework inspired by Nash bargaining theory. Our method guides the optimization process toward a Pareto stationary point, ensuring an equilibrium solution wherein any deviation from the final state would degrade the overall performance with respect to both objectives. This guarantees optimality in preserving prior knowledge while integrating new information. We empirically validate the effectiveness of our approach across a range of evaluation metrics on standard benchmark datasets. Extensive experiments show that our method consistently outperforms state-of-the-art techniques, achieving a superior balance between knowledge preservation and update accuracy.
OPLoRA: Orthogonal Projection LoRA Prevents Catastrophic Forgetting During Parameter-Efficient Fine-Tuning
PDF ↗Low-Rank Adaptation (LoRA) enables efficient fine-tuning of large language models but suffers from catastrophic forgetting when learned updates interfere with the dominant singular directions that encode essential pre-trained knowledge. We propose Orthogonal Projection LoRA (OPLoRA), a theoretically grounded approach that prevents this interference through double-sided orthogonal projections. By decomposing frozen weights via SVD, OPLoRA constrains LoRA updates to lie entirely within the orthogonal complement of the top-k singular subspace using projections PL = I − Uk Ukᵀ and PR = I − Vk Vkᵀ. We prove that this construction exactly preserves the top-k singular triples, providing mathematical guarantees for knowledge retention. To quantify subspace interference, we introduce ρk, a metric measuring update alignment with dominant directions. Extensive experiments across commonsense reasoning, mathematics, and code generation demonstrate that OPLoRA significantly reduces forgetting while maintaining competitive task-specific performance on LLaMA-2 7B and Qwen2.5 7B, establishing orthogonal projection as an effective mechanism for knowledge preservation in parameter-efficient fine-tuning.
Black-box tuning is an emerging paradigm for adapting large language models (LLMs) to better achieve desired behaviors, particularly when direct access to model parameters is unavailable. Current strategies, however, often present a dilemma of suboptimal extremes: either separately train a small proxy model and then use it to shift the predictions of the foundation model, offering notable efficiency but often yielding limited improvement; or making API calls in each tuning iteration to the foundation model, which entails prohibitive computational costs. In this paper, we argue that a more reasonable way for black-box tuning is to train the proxy model with limited API calls. The underlying intuition is based on two key observations: first, the training samples may exhibit correlations and redundancies, suggesting that the foundation model’s predictions can be estimated from previous calls; second, foundation models frequently demonstrate low accuracy on downstream tasks. Therefore, we propose a novel advanced black-box tuning method for LLMs with limited API calls. Our core strategy involves training a Gaussian Process (GP) surrogate model with "LogitMap Pairs" derived from querying the foundation model on a minimal but highly informative training subset. This surrogate can approximate the outputs of the foundation model to guide the training of the proxy model, thereby effectively reducing the need for direct queries to the foundation model. Extensive experiments verify that our approach elevates pre-trained language model accuracy from 55.92% to 86.85%, reducing the frequency of API queries to merely 1.38%. This significantly outperforms offline approaches that operate entirely without API access. Notably, our method also achieves comparable or superior accuracy to query-intensive approaches, while significantly reducing API costs. This offers a robust and high-efficiency paradigm for language model adaptation.
Knowledge Distillation (KD) is a key technique for compressing Large-scale Language Models (LLMs), but prevailing logit-based methods employ static strategies misaligned with the student’s dynamic learning process. By treating all tokens indiscriminately with a fixed temperature, these methods result in suboptimal knowledge transfer. To address this, we propose LLM-oriented token-Adaptive Knowledge Distillation (AdaKD), a framework that adapts the distillation process to each token’s real-time learning state. AdaKD consists of two synergistic modules driven by a unified token difficulty metric. First, the Loss-driven Adaptive Token Focusing (LATF) module dynamically concentrates distillation on valuable tokens by monitoring the student’s learning stability. Second, Inverse Difficulty Temperature Scaling (IDTS) introduces a counterintuitive token-level temperature: low for difficult tokens to target error correction, and high for easy tokens to learn the teacher’s smooth output distribution for better generalization. As a plug-and-play framework, AdaKD consistently improves performance across diverse distillation methods, model architectures, and benchmarks.
CroPS: Improving Dense Retrieval with Cross-Perspective Positive Samples in Short-Video Search
PDF ↗Dense retrieval has become a foundational paradigm in modern search systems, especially on short-video platforms. However, most industrial systems adopt a self-reinforcing training pipeline that relies on historically exposed user interactions for supervision. This paradigm inevitably leads to a filter bubble effect, where potentially relevant but previously unseen content is excluded from the training signal, biasing the model toward narrow and conservative retrieval. In this paper, we present CroPS (Cross-Perspective Positive Samples), a novel retrieval data engine designed to alleviate this problem by introducing diverse and semantically meaningful positive examples from multiple perspectives. CroPS enhances training with positive signals derived from user query reformulation behavior (query-level), engagement data in recommendation streams (system-level), and world knowledge synthesized by large language models (knowledge-level). To effectively utilize these heterogeneous signals, we introduce a Hierarchical Label Assignment (HLA) strategy and a corresponding H-InfoNCE loss that together enable fine-grained, relevance-aware optimization. Extensive experiments conducted on Kuaishou Search, a large-scale commercial short-video search platform, demonstrate that CroPS significantly outperforms strong baselines both offline and in live A/B tests, achieving superior retrieval performance and reducing query reformulation rates. CroPS is now fully deployed in Kuaishou Search, serving hundreds of millions of users daily.
Optimization modeling plays a critical role in supporting optimal decision-making across various domains. Previous works have demonstrated that large language models (LLMs) tailored for optimization modeling have significantly automated and simplified this process. However, these models typically employ a straightforward input-output paradigm and struggle with challenging instances. In contrast, recent advances in general-purpose reasoning LLMs (RLLMs), such as DeepSeek-R1, have shown impressive capabilities in complex domains like mathematics and coding. In this paper, we introduce DeepOR, the first RLLM specifically designed for optimization modeling. Instead of directly outputting solutions, DeepOR explicitly performs multiple intermediate reasoning steps. To adapt a base LLM into an RLLM, we begin by synthesizing long chain-of-thought (CoT) data guided by a flowchart, which is automatically generated using a self-exploration algorithm. Once the training data are prepared, we employ supervised fine-tuning on the base LLM to endow it with reasoning capabilities tailored for optimization modeling. To fully leverage the model's reasoning potential, we further apply reinforcement learning with reward-shaping derived from solver feedback. Experimental results on benchmarks confirm that DeepOR consistently and significantly outperforms existing state-of-the-art approaches.
Large Language Models (LLMs) have achieved remarkable success across a wide range of natural language tasks, but often exhibit overconfidence and generate plausible yet incorrect answers. This overconfidence, especially in models undergone Reinforcement Learning from Human Feedback (RLHF), poses significant challenges for reliable uncertainty estimation and safe deployment. In this paper, we propose EAGLE (Expectation of AGgregated internaL bEief), a novel self-evaluation-based calibration method that leverages the internal hidden states of LLMs to derive more accurate confidence scores. Instead of relying on the model's final output, our approach extracts internal beliefs from multiple intermediate layers during self-evaluation. By aggregating these layer-wise beliefs and calculating the expectation over the resulting confidence score distribution, EAGLE produces a refined confidence score that more faithfully reflects the model's internal certainty. Extensive experiments on diverse datasets and LLMs demonstrate that EAGLE significantly improves calibration performance over existing baselines. We also provide an in-depth analysis of EAGLE, including a layer-wise examination of uncertainty patterns, a study of the impact of self-evaluation prompts, and an analysis of the effect of self-evaluation score range.
SCALE: Selective Resource Allocation for Overcoming Performance Bottlenecks in Mathematical Test-time Scaling
PDF ↗Test-time compute scaling has emerged as a powerful paradigm for enhancing mathematical reasoning in large language models (LLMs) by allocating additional computational resources during inference. However, current methods employ uniform resource distribution across all reasoning sub-problems, creating fundamental bottlenecks where challenging sub-problems receive insufficient attention while routine operations consume disproportionate resources. This uniform allocation creates performance bottlenecks where additional computational resources yield diminishing returns. Inspired by dual-process theory, we propose SCALE (Selective Resource Allocation), a framework that selectively allocates computational resources based on sub-problem difficulty. SCALE operates through four stages: (1) problem decomposition into sequential reasoning sub-problems, (2) difficulty assessment of each sub-problem to distinguish between routine operations and computationally challenging sub-problems, (3) selective processing mode assignment between System 1 for simple sub-problems and System 2 for complex ones, and (4) sequential execution with context propagation. By concentrating resources on challenging sub-problems while processing routine operations efficiently, SCALE achieves substantial performance improvements with superior resource utilization. Extensive experiments demonstrate that SCALE significantly outperforms uniform scaling baselines, achieving accuracy improvements of up to 13.75 percentage points (57.50% to 71.25% on AIME25) while reducing computational costs by 33-53%, representing a major advance in test-time scaling that addresses fundamental limitations of current approaches.
With the rapid advancement of large language models (LLMs), their deployment in real-world applications has become increasingly widespread. LLMs are expected to deliver robust performance across diverse tasks, user preferences, and practical scenarios. However, as demands grow, ensuring that LLMs produce responses aligned with human intent remains a foundational challenge. In particular, aligning model behavior effectively and efficiently during inference, without costly retraining or extensive supervision, is both a critical requirement and a non-trivial technical endeavor. To address the challenge, we propose SDA (Steering-Driven Distribution Alignment), a training-free and model-agnostic alignment framework designed for open-source LLMs. SDA dynamically redistributes model output probabilities based on user-defined alignment instructions, enhancing alignment between model behavior and human intents without fine-tuning. The method is lightweight, resource-efficient, and compatible with a wide range of open-source LLMs. It can function independently during inference or be integrated with training-based alignment strategies. Moreover, SDA supports personalized preference alignment, enabling flexible control over the model’s response behavior. Empirical results demonstrate that SDA consistently improves alignment performance across 8 open-source LLMs with varying scales and diverse origins, evaluated on three key alignment dimensions, helpfulness, harmlessness, and honesty (3H). Specifically, SDA achieves average gains of 64.4% in helpfulness, 30% in honesty and 11.5% in harmlessness across the tested models, indicating its effectiveness and generalization across diverse models and application scenarios.
We introduce KALL-E, a novel autoregressive (AR) language model for text-to-speech (TTS) synthesis that operates by predicting the next distribution of continuous speech frames. Unlike existing methods, KALL-E directly models the continuous speech distribution conditioned on text, eliminating the need for any diffusion-based components. Specifically, we utilize a Flow-VAE to extract a continuous latent speech representation from waveforms, instead of relying on discrete speech tokens. A single AR Transformer is then trained to predict these continuous speech distributions from text, optimizing a Kullback–Leibler divergence loss as its objective. Experimental results demonstrate that KALL-E achieves superior speech synthesis quality and can even adapt to a target speaker from just a single sample. Importantly, KALL-E provides a more direct and effective approach for utilizing continuous speech representations in TTS.
MetaAct-RL: Training Language Models for Reasoning Through Meta-Action-Based Reinforcement Learning
PDF ↗Outcome-based reinforcement learning has made notable advances in training language models (LMs) for reasoning. However, without explicit incentives and controls, this paradigm has limitations and instability in eliciting high-quality reasoning trajectories with diverse actions—particularly for models whose pretraining lacked extensive reasoning-related data. To this end, we introduce MetaAct-RL, a new RL framework that frames LMs’ thinking as sequential decision making over meta-actions. In this framework, the model chooses and executes a high-level action at each step—such as forward reasoning, critique, or refinement—to gradually reach the correct answer. To encourage deeper exploration, richer action diversity, and to improve sampling efficiency in the RL optimization process, MetaAct-RL incorporates appropriate length-based reward and regularization, and a key-state restart mechanism. Extensive experiments across six benchmarks show that MetaAct-RL improves reasoning performance by 7.99 on Llama3.2-1B and 7.17 on Llama3.1-8B relative to vanilla RL method. Moreover, on the challenging AIME-2024, our method outperforms the vanilla RL by 7.5 with Qwen2.5-1.5B.
GlitchMiner: Mining Glitch Tokens in Large Language Models via Gradient-based Discrete Optimization
PDF ↗Glitch tokens—inputs that trigger unpredictable or anomalous behavior in Large Language Models (LLMs)—pose significant challenges to model reliability and safety. Existing detection methods primarily rely on heuristic embedding patterns or statistical anomalies within internal representations, limiting their generalizability across different model architectures and potentially missing anomalies that deviate from observed patterns. We introduce GlitchMiner, an behavior-driven framework designed to identify glitch tokens by maximizing predictive entropy. Leveraging a gradient-guided local search strategy, GlitchMiner efficiently explores the discrete token space without relying on model-specific heuristics or large-batch sampling. Extensive experiments across ten LLMs from five major model families demonstrate that GlitchMiner consistently outperforms existing approaches in detection accuracy and query efficiency, providing a generalizable and scalable solution for effective glitch token discovery.
Graphical user interface (GUI) agents have recently emerged as an intriguing paradigm for human-computer interaction, capable of automatically executing user instructions to operate intelligent terminal devices. However, when encountering out-of-distribution (OOD) instructions that violate environmental constraints or exceed the current capabilities of agents, GUI agents may suffer task breakdowns or even pose security threats. Therefore, effective OOD detection for GUI agents is essential. Traditional OOD detection methods perform suboptimally in this domain due to the complex embedding space and evolving GUI environments. In this work, we observe that the in-distribution input semantic space of GUI agents exhibits a clustering pattern with respect to the distance from the centroid. Based on the finding, we propose GEM, a novel method based on fitting a Gaussian mixture model over input embedding distances extracted from the GUI Agent that reflect its capability boundary. Evaluated on eight datasets spanning smartphones, computers, and web browsers, our method achieves an average accuracy improvement of 23.70% over the best-performing baseline while only increasing training time by 4.9% and testing time by 6.5%. We also experimentally demonstrate that GEM can improve the step-wise success rate by 9.40% by requesting assistance from the cloud model when encountering OOD samples. Analysis verifies the generalization ability of our method through experiments on nine different backbones.
StepFun-Formalizer: Unlocking the Autoformalization Potential of LLMs Through Knowledge-Reasoning Fusion
PDF ↗Autoformalization aims to translate natural-language mathematical statements into a formal language. While LLMs have accelerated progress in this area, existing methods still suffer from low accuracy. We identify two key abilities for effective autoformalization: comprehensive mastery of formal-language domain knowledge, and reasoning capability of natural language problem understanding and informal-formal alignment. Without the former, a model cannot identify the correct formal objects; without the latter, it struggles to interpret real-world contexts and map them precisely into formal expressions. To address these gaps, we introduce ThinkingF, a data synthesis and training pipeline that improves both abilities. First, we construct two datasets: one by distilling and selecting large-scale examples rich in formal knowledge, and another by generating informal-to-formal reasoning trajectories guided by expert-designed templates. We then apply SFT and RLVR with these datasets to further fuse and refine the two abilities. The resulting 7B and 32B models exhibit both comprehensive formal knowledge and strong informal-to-formal reasoning. Notably, StepFun-Formalizer-32B achieves SOTA BEq@1 scores of 40.5% on FormalMATH-Lite and 26.7% on ProverBench, surpassing all prior general-purpose and specialized models.
Finding the Translation Switch: Discovering and Exploiting the Task-Initiation Features in LLMs
PDF ↗Large Language Models (LLMs) frequently exhibit strong translation abilities, even without task-specific fine-tuning. However, the internal mechanisms governing this innate capability remain largely opaque. To demystify this process, we leverage Sparse Autoencoders (SAEs) and introduce a novel framework for identifying task-specific features. Our method first recalls features that are frequently co-activated on translation inputs and then filters them for functional coherence using a PCA-based consistency metric. This framework successfully isolates a small set of "translation initiation" features. Causal interventions demonstrate that amplifying these features steers the model towards correct translation, while ablating them induces hallucinations and off-task outputs, confirming they represent a core component of the model's innate translation competency. Moving from analysis to application, we leverage this mechanistic insight to propose a new data selection strategy for efficient fine-tuning. Specifically, we prioritize training on "mechanistically hard" samples—those that fail to naturally activate the translation initiation features. Experiments show this approach significantly improves data efficiency and suppresses hallucinations. Furthermore, we find these mechanisms are transferable to larger models of the same family. Our work not only decodes a core component of the translation mechanism in LLMs but also provides a blueprint for using internal model mechanism to create more robust and efficient models.
Audio-Thinker: Guiding Large Audio Language Model When and How to Think via Reinforcement Learning
PDF ↗Recent advancements in large language models, multimodal large language models, and large audio language models (LALMs) have significantly improved their reasoning capabilities through reinforcement learning utilizing rule-based rewards. However, the explicit reasoning process has not yet yielded substantial benefits for audio question answering, and effectively leveraging deep reasoning remains an open challenge, with LALMs still falling short of achieving human-level auditory-language reasoning. To address these limitations, we propose Audio-Thinker, a reinforcement learning framework designed to enhance the reasoning capabilities of LALMs through improved adaptability, consistency, and effectiveness. Our approach introduces an adaptive think accuracy reward, enabling the model to adjust its reasoning strategies based on task complexity. Furthermore, we incorporate an external reward model to evaluate the overall consistency and quality of the reasoning process, complemented by think-based rewards that assist the model in distinguishing between valid and flawed reasoning paths during training. Experimental results demonstrate that Audio-Thinker models outperform existing reasoning-oriented LALMs across various benchmark tasks, exhibiting superior reasoning and generalization capabilities.
Radiology report generation from chest X-rays is an important task in artificial intelligence with the potential to greatly reduce radiologists' workload and shorten patient wait times. Despite recent advances, existing approaches often lack sufficient disease-awareness in visual representations and adequate vision-language alignment to meet the specialized requirements of medical image analysis. As a result, these models usually overlook critical pathological features on chest X-rays and struggle to generate clinically accurate reports. To address these limitations, we propose a novel dual-stage disease-aware framework for chest X-ray report generation. In Stage~1, our model learns Disease-Aware Semantic Tokens (DASTs) corresponding to specific pathology categories through cross-attention mechanisms and multi-label classification, while simultaneously aligning vision and language representations via contrastive learning. In Stage~2, we introduce a Disease-Visual Attention Fusion (DVAF) module to integrate disease-aware representations with visual features, along with a Dual-Modal Similarity Retrieval (DMSR) mechanism that combines visual and disease-specific similarities to retrieve relevant exemplars, providing contextual guidance during report generation. Extensive experiments on benchmark datasets (i.e., CheXpert Plus, IU X-ray, and MIMIC-CXR) demonstrate that our disease-aware framework achieves state-of-the-art performance in chest X-ray report generation, with significant improvements in clinical accuracy and linguistic quality.