Large language models (LLMs) have demonstrated remarkable reasoning capabilities, including in financial question answering (FQA). However, the performance in FQA remains limited, particularly in questions that require deep financial knowledge and complex numerical reasoning. While supervised fine-tuning and closed-source LLMs have shown promise, they are often constrained by high costs or computational inefficiency. In this paper, we propose a low-cost yet effective framework, named FinMAN (Financial multi-agent framework), that enables small LLMs (e.g., 8B) to perform complex reasoning tasks without relying on expensive models or task-specific fine-tuning. FinMAN improves formula selection, extraction, and calculation to help small-scale models solve FQA tasks more accurately, with a lightweight verification mechanism to correct common errors. Experimental results show that FinMAN outperforms the best open-source model on BizBench by 10.46% and achieves competitive performance to GPT-3.5 using significantly fewer parameters. Our code and data are publicly available at https://github.com/coenliu/MultiAgentFin.
论文检索
输入标题、作者或关键词，从 7,591 篇学术成果中精准定位
Prior research diverges on language diversity in LLM fine-tuning: Some studies report benefits while others find no advantages. Through controlled fine-tuning experiments across 132 translation directions, we systematically resolve these disparities. We find that expanding language diversity during fine-tuning improves translation quality for both unsupervised and—surprisingly—supervised pairs, despite less diverse models being fine-tuned exclusively on these supervised pairs. However, benefits plateau or decrease beyond a certain diversity threshold. We show that increased language diversity creates more language-agnostic representations. These representational adaptations help explain the improved performance in models fine-tuned with greater diversity.
Sequence Structure Aware Retriever for Procedural Document Retrieval: A New Dataset and Baseline
PDF ↗Execution failures are common in daily life when individuals perform procedural tasks, such as cooking or handicrafts making. Retrieving relevant procedural documents that align closely with both the content of steps and the overall execution sequence can help correct these failures with fewer modifications. However, existing retrieval methods, which primarily focus on declarative knowledge, often neglect the execution sequence structures inherent in procedural documents. To tackle this challenge, we introduce a new dataset Procedural Questions, and propose a retrieval model Graph-Fusion Procedural Document Retriever (GFPDR) which integrates procedural graphs with document representations. Extensive experiments demonstrate the effectiveness of GFPDR, highlighting its superior performance in procedural document retrieval compared to existing models.
ReLoop: “Seeing Twice and Thinking Backwards” via Closed-loop Training to Mitigate Hallucinations in Multimodal understanding
PDF ↗While Multimodal Large Language Models (MLLMs) have achieved remarkable progress in open-ended visual question answering, they remain vulnerable to hallucinations. These are outputs that contradict or misrepresent input semantics, posing a critical challenge to the reliability and factual consistency. Existing methods often rely on external verification or post-hoc correction, lacking an internal mechanism to validate outputs directly during training. To bridge this gap, we propose ReLoop, a unified closed-loop training framework that encourages multimodal consistency for cross-modal understanding in MLLMs. ReLoop adopts a ring-shaped structure that integrates three complementary consistency feedback mechanisms, obliging MLLMs to “seeing twice and thinking backwards”. Specifically, ReLoop employs the frozen Consistency Feedback Plugin (CFP), comprising semantic reconstruction, visual description, and an attention supervision module for attention alignment. These components collectively enforce semantic reversibility, visual consistency, and interpretable attention, enabling the model to correct its outputs during training. Extensive evaluations and analyses demonstrate the effectiveness of ReLoop in reducing hallucination rates across multiple benchmarks, establishing a robust method for hallucination mitigation in MLLMs. We will release our source code and data in the camera-ready version. The code is available at: https://github.com/ZiyanHuang11/Reloop-hallucinations.
FlexQuant: A Flexible and Efficient Dynamic Precision Switching Framework for LLM Quantization
PDF ↗The rapid advancement of large language models (LLMs) has exacerbated the memory bottleneck due to the widening gap between model parameter scaling and hardware capabilities. While post-training quantization techniques effectively reduce memory overhead, existing methods predominantly rely on static quantization strategies, which struggle to adapt to dynamic workloads. To address this, we propose FlexQuant, a dynamic precision-switching framework that optimizes the trade-off between inference speed and accuracy. Leveraging model perplexity entropy and Kullback-Leibler divergence, FlexQuant enables fine-grained, layer-wise mixed-precision quantization and dynamically adjusts bit-widths during each token generation. FlexQuant provides a comprehensive analysis of quantization strategies, introduces a precision requirement model for optimal switching, and implements efficient fine-grained precision management. Evaluations demonstrate that FlexQuant achieves a 1.3× end-to-end speedup across diverse language tasks with negligible accuracy loss introduced. This framework offers a flexible and adaptive solution for efficient LLM deployment.
ParetoRAG: Leveraging Sentence-Context Attention for Robust and Efficient Retrieval-Augmented Generation
PDF ↗While Retrieval-Augmented Generation systems enhance Large Language Models by incorporating external knowledge, they still face persistent challenges in retrieval inefficiency and the inability of LLMs to filter out irrelevant information. We presentParetoRAG, an unsupervised framework that optimizes RAG systems through sentence-level refinement guided by the Pareto principle. By decomposing paragraphs into sentences and dynamically re-weighting core content while preserving contextual coherence, ParetoRAG achieves dual improvements in retrieval precision and generation quality without requiring additional training or API resources, while using only 40% of the tokens compared to traditional RAG approaches. This framework has been empirically validated across various datasets, LLMs, and retrievers. Furthermore, we show that ParetoRAG’s architectural improvements are orthogonally compatible with adaptive noise-robust models, enabling retrieval-augmented optimization and robust training to enhance generation quality mutually. This highlights complementary architectural refinements and noise mitigation, offering insights for integrating retrieval augmentation with robustness enhancement.
UPLex: Fine-Grained Personality Control in Large Language Models via Unsupervised Lexical Modulation
PDF ↗Personality is a crucial factor that shapes human communication patterns, thereby regulating the personalities of large language models (LLMs) holds significant potential in enhancing their user experiences. Previous approaches either relied on fine-tuning LLMs on specific corpora or required manually crafted prompts to evoke specific personalities from LLMs. However, the former is inefficient and costly, while the latter cannot precisely manipulate personality traits at a fine-grained level. To address these challenges, we propose UPLex, a method that uses an Unsupervisedly-Built Personalized Lexicon (UPL) during the decoding phase to manipulate LLM’s personality traits. UPLex can be constructed from a newly built situational judgment test dataset in an unsupervised fashion and used to modulate the personality expression of LLMs by dynamically altering their predicted probability of upcoming words in a pluggable fashion. Extensive experimentation demonstrates the remarkable effectiveness and pluggability of our method for fine-grained manipulation of LLMs’ personalities.
Retrieval-augmented generation (RAG) is a powerful method for enhancing natural language generation by integrating external knowledge into a model’s output. While prior work has demonstrated the importance of improving knowledge retrieval for boosting generation quality, the role of knowledge selection, a.k.a. reranking or filtering, remains less clear. This paper empirically analyzes how knowledge selection influences downstream generation performance in RAG systems. By simulating different retrieval and selection conditions through a controlled mixture of gold and distractor knowledge, we assess the impact of these factors on generation outcomes. Our findings indicate that the downstream generator model’s capability, as well as the complexity of the task and dataset, significantly influence the impact of knowledge selection on the overall RAG system performance. In typical scenarios, improving the knowledge recall score is key to enhancing generation outcomes, with the knowledge selector providing limited benefit when a strong generator model is used on clear, well-defined tasks. For weaker generator models or more ambiguous tasks and datasets, the knowledge F1 score becomes a critical factor, and the knowledge selector plays a more prominent role in improving overall performance.
Sparkle: Mastering Basic Spatial Capabilities in Vision Language Models Elicits Generalization to Spatial Reasoning
PDF ↗Vision-language models (VLMs) excel in many downstream tasks but struggle with spatial reasoning, which is crucial for navigation and interaction with physical environments. Specifically, many spatial reasoning tasks rely on fundamental two-dimensional (2D) capabilities, yet our evaluation shows that state-of-the-art VLMs often produce implausible or incorrect solutions for composite spatial problems, including simple pathfinding tasks that humans solve effortlessly at a glance. To address this, we explore an effective approach to enhance 2D spatial reasoning in VLMs by training them solely on basic spatial capabilities. We first disentangle 2D spatial reasoning into three core components: direction comprehension, distance estimation, and localization. Our central hypothesis is that mastering these basic capabilities will significantly boost performance on more complex spatial tasks requiring advanced reasoning and combinatorial problem-solving, as well as generalize to real-world visual-spatial scenarios. To test this hypothesis, we introduce Sparkle, a framework that generates synthetic data to provide targeted supervision for VLMs across these three basic spatial capabilities, producing an instruction dataset for each capability. Our experiments demonstrate that VLMs fine-tuned with Sparkle achieve substantial improvements, not only on basic tasks but also in generalizing to composite and out-of-distribution real-world spatial reasoning tasks. These findings highlight that enhancing basic spatial capabilities through synthetic generalization effectively improves complex spatial reasoning, offering insights into systematic strategies for boosting VLMs’ spatial understanding. Source codes of Sparkle are available at https://github.com/YihongT/Sparkle.
StereoDetect: Detecting Stereotypes and Anti-stereotypes the Correct Way Using Social Psychological Underpinnings
PDF ↗Stereotypes are known to have very harmful effects, making their detection critically important. However, current research predominantly focuses on detecting and evaluating stereotypical biases, leaving the study of stereotypes in its early stages. Our study revealed that many works have failed to clearly distinguish between stereotypes and stereotypical biases, which has significantly slowed progress in advancing research in this area. Stereotype and Anti-stereotype detection is a problem that requires social knowledge; hence, it is one of the most difficult areas in Responsible AI. This work investigates this task, where we propose a five-tuple definition and provide precise terminologies disentangling stereotypes, anti‐stereotypes, stereotypical bias, and general bias. We provide a conceptual framework grounded in social psychology for reliable detection. We identify key shortcomings in existing benchmarks for this task of stereotype and anti-stereotype detection. To address these gaps, we developed *StereoDetect*, a well curated, definition‐aligned benchmark dataset designed for this task. We show that language models with fewer than 10 billion parameters frequently misclassify anti‐stereotypes and fail to recognize neutral overgeneralizations. We demonstrate StereoDetect’s effectiveness through multiple qualitative and quantitative comparisons with existing benchmarks and models fine-tuned on them.
Retrieval-Augmented Generation (RAG) compensates for the static knowledge limitations of Large Language Models (LLMs) by integrating external knowledge, producing responses with enhanced factual correctness and query-specific contextualization. However, it also introduces new attack surfaces such as corpus poisoning at the same time. Most of the existing defense methods rely on the internal knowledge of the model, which conflicts with the design concept of RAG. To bridge the gap, EcoSafeRAG uses sentence-level processing and bait-guided context diversity detection to identify malicious content by analyzing the context diversity of candidate documents without relying on LLM internal knowledge. Experiments show EcoSafeRAG delivers state-of-the-art security with plug-and-play deployment, simultaneously improving clean-scenario RAG performance while maintaining practical operational costs (relatively 1.2 \times latency, 48%-80% token reduction versus Vanilla RAG).
Medical question answering fundamentally relies on accurate clinical knowledge. The dominant paradigm, Retrieval-Augmented Generation (RAG), acquires expertise conceptual knowledge from large-scale medical corpus to guide general-purpose large language models (LLMs) in generating trustworthy answers. However, existing retrieval approaches often overlook the patient-specific factual knowledge embedded in Electronic Health Records (EHRs), which limits the contextual relevance of retrieved conceptual knowledge and hinders its effectiveness in vital clinical decision-making. This paper introduces RGAR, a recurrence generation-augmented retrieval framework that synergistically retrieves both factual and conceptual knowledge from dual sources (i.e., EHRs and the corpus), allowing mutual refinement through iterative interaction. Across three factual-aware medical QA benchmarks, RGAR establishes new state-of-the-art performance among medical RAG systems. Notably, RGAR enables the Llama-3.1-8B-Instruct model to surpass the considerably larger GPT-3.5 augmented with traditional RAG. Our findings demonstrate the benefit of explicitly mining patient-specific factual knowledge during retrieval, consistently improving generation quality and clinical relevance.
Unlocking Smarter Device Control: Foresighted Planning with a World Model-Driven Code Execution Approach
PDF ↗The automatic control of mobile devices is essential for efficiently performing complex tasks that involve multiple sequential steps. However, these tasks pose significant challenges due to the limited environmental information available at each step, primarily through visual observations. As a result, current approaches, which typically rely on reactive policies, focus solely on immediate observations and often lead to suboptimal decision-making. To address this problem, we propose Foresighted Planning with World Model-Driven Code Execution (FPWC),a framework that prioritizes natural language understanding and structured reasoning to enhance the agent’s global understanding of the environment by developing a task-oriented, refinable world model at the outset of the task. Foresighted actions are subsequently generated through iterative planning within this world model, executed in the form of executable code. Extensive experiments conducted in simulated environments and on real mobile devices demonstrate that our method outperforms previous approaches, particularly achieving a 44.4% relative improvement in task success rate compared to the state-of-the-art in the simulated environment.
Learning What to Remember: Adaptive Probabilistic Memory Retention for Memory-Efficient Language Models
PDF ↗Transformer attention scales quadratically with sequence length O(n^2), limiting long-context use. We propose Adaptive Retention, a probabilistic, layer-wise token selection mechanism that learns which representations to keep under a strict global budget M. Retention is modeled with Bernoulli gates trained via a Hard-Concrete/variational relaxation and enforced with a simple top-M rule at inference, making the method differentiable and drop-in for standard encoders. Across classification, extractive QA, and long-document summarization, keeping only 30–50% of tokens preserves \geq 95\% of full-model performance while cutting peak memory by \sim 35–45% and improving throughput by up to \sim 1.8\times. This architecture-agnostic approach delivers practical long-context efficiency without modifying base attention or task heads.
Exploring the Generalizability of Factual Hallucination Mitigation via Enhancing Precise Knowledge Utilization
PDF ↗Large Language Models (LLMs) often struggle to align their responses with objective facts, resulting in the issue of \textbf{factual hallucinations}, which can be difficult to detect and mislead users without relevant knowledge. Although post-training techniques have been employed to mitigate the issue, existing methods usually suffer from poor generalization and trade-offs in other different capabilities. In this paper, we propose to address these by directly augmenting LLM’s fundamental ability to precisely leverage its knowledge and introduce \textbf{PKUE} (\textbf{P}recise \textbf{K}nowledge \textbf{U}tilization \textbf{E}nhancement), which fine-tunes the model on self-generated responses to precise and simple factual questions through preference optimization. Furthermore, we construct \textbf{FactualBench}, a comprehensive and precise factual QA dataset containing 181k Chinese data spanning 21 domains, to facilitate both evaluation and training. Extensive experiments demonstrate that PKUE significantly improves LLM overall performance, with consistent enhancement across factual tasks of various forms, general tasks beyond factuality, and tasks in different language.
Recent advancements in large language models (LLMs) have enabled LLM-based agents to successfully tackle interactive planning tasks. However, despite their successes, existing approaches often suffer from planning hallucinations and require retraining for each new agent. To address these challenges, we propose the **M**eta **P**lan **O**ptimization (**MPO**) framework, , which enhances agent planning capabilities by directly incorporating explicit guidance. Unlike previous methods that rely on complex knowledge, which either require significant human effort or lack quality assurance, MPO leverages high-level general guidance through meta plans to assist agent planning and enables continuous optimization of the meta plans based on feedback from the agent’s task execution. Our experiments conducted on two representative tasks demonstrate that MPO significantly outperforms existing baselines. Moreover, our analysis indicates that MPO provides a plug-and-play solution that enhances both task completion efficiency and generalization capabilities in previous unseen scenarios.
Large Reason Models (LRMs) extend long reasoning process to solve complex tasks. However, due to the lack of fine-grained control, they often suffer from overthinking and erroneous reasoning problems, risking accuracy loss. To address this issue, we introduce Reasoning Direction Steering (RDS) to enable fine-grained control over LRMs’ reasoning behaviors by aligning reasoning trajectories with specific cognitive patterns. We develop a simple yet effective paradigm, Thinking Intervention, which explores two key dimensions - intervention positions and intervention styles - to achieve integration intervention throughout model reasoning processes. To validate the effectiveness of our approach, we conduct comprehensive experiments on multi-hop question answering tasks using state-of-the-art LRMs, including Qwen3-Series and R1-Series models. Experimental results demonstrate the efficacy of Thinking Intervention with 9.4% average improvement on R1-Series models and 1.9% improvement on Qwen3-Series models.
RouterEval: A Comprehensive Benchmark for Routing LLMs to Explore Model-level Scaling Up in LLMs
PDF ↗Routing large language models (LLMs) is a new paradigm that uses a router to recommend the best LLM from a pool of candidates for a given input. In this paper, our comprehensive analysis with more than 8,500 LLMs reveals a novel model-level scaling up phenomenon in Routing LLMs, i.e., a capable router can significantly enhance the performance of this paradigm as the number of candidates increases. This improvement can even surpass the performance of the best single model in the pool and many existing strong LLMs, confirming it a highly promising paradigm. However, the lack of comprehensive and open-source benchmarks for Routing LLMs has hindered the development of routers. In this paper, we introduce RouterEval, a benchmark tailored for router research, which includes over 200,000,000 performance records for 12 popular LLM evaluations across various areas such as commonsense reasoning, semantic understanding, etc., based on over 8,500 various LLMs. Using RouterEval, extensive evaluations of existing Routing LLM methods reveal that most still have significant room for improvement.
Hierarchical Text Classification is a challenging task which classifies texts into categories arranged in a hierarchy. Zero‐Shot Hierarchical Text Classification (ZS-HTC) further assumes only the availability of hierarchical taxonomy, without any training data. Existing works of ZS-HTC are typically built on the prototype-based framework by embedding the category names into prototypes, which, however, do not perform very well due to the ambiguity and impreciseness of category names. In this paper, we propose HierPrompt, a method that leverages hierarchy-aware prompts to instruct LLM to produce more representative and informative prototypes. Specifically, we first introduce Example Text Prototype (ETP), in conjunction with Category Name Prototype (CNP), to enrich the information contained in hierarchical prototypes. A Maximum Similarity Propagation (MSP) technique is also proposed to consider the hierarchy in similarity calculation. Then, the hierarchical prototype refinement module is utilized to (i) contextualize the category names for more accurate CNPs and (ii) produce detailed example texts for each leaf category to form ETPs. Experiments on three benchmark datasets demonstrate that HierPrompt substantially outperforms existing ZS‐HTC methods.
This paper addresses the challenge of handling unseen modalities and dynamic modality combinations at test time with our proposed text-centric alignment method. This training-free alignment approach unifies different input modalities into a single semantic text representation by leveraging in-context learning with Large Language Models and uni-modal foundation models. Our method significantly enhances the ability to manage unseen, diverse, and unpredictable modality combinations, making it suitable for both generative and discriminative models to adopt on top. Our extensive experiments primarily evaluate on discriminative tasks, demonstrating that our approach is essential for LLMs to achieve strong modality alignment performance. It also surpasses the limitations of traditional fixed-modality frameworks in embedding representations. This study contributes to the field by offering a flexible and effective solution for real-world applications where modality availability is dynamic and uncertain.