Multimodal Large Language Models (MLLMs) have experienced rapid development in recent years. However, in the financial domain, there is a notable lack of effective and specialized multimodal evaluation datasets. To advance the development of MLLMs in the finance domain, we introduce FinMME, encompassing more than 11,000 high-quality financial research samples across 18 financial domains and 6 asset classes, featuring 10 major chart types and 21 subtypes. We ensure data quality through 20 annotators and carefully designed validation mechanisms. Additionally, we develop FinScore, an evaluation system incorporating hallucination penalties and multi-dimensional capability assessment to provide an unbiased evaluation. Extensive experimental results demonstrate that even state-of-the-art models like GPT-4o exhibit unsatisfactory performance on FinMME, highlighting its challenging nature. The benchmark exhibits high robustness with prediction variations under different prompts remaining below 1%, demonstrating superior reliability compared to existing datasets. Our dataset and evaluation protocol are available at https://huggingface.co/datasets/luojunyu/FinMME and https://github.com/luo-junyu/FinMME.
论文检索
输入标题、作者或关键词，从 3,086 篇学术成果中精准定位
This paper introduces a pioneering methodology, termed StructTuning, to efficiently transform foundation Large Language Models (LLMs) into domain specialists. It significantly reduces the training corpus needs to a mere 5% while achieving an impressive 100% of traditional knowledge injection performance. Motivated by structured human education, we propose a novel two-stage strategy for knowledge injection and alignment: Structure-aware Continual Pre-Training (SCPT) and Structure-aware Supervised Fine-Tuning (SSFT). In the SCPT phase, we automatically extract the domain knowledge taxonomy and reorganize the training corpora, enabling LLMs to effectively link textual segments to targeted knowledge points within the taxonomy. In the SSFT phase, we explicitly prompt models to elucidate the underlying knowledge structure in their outputs, leveraging the structured domain insight to address practical problems. Our ultimate method was extensively evaluated across model architectures and scales on LongBench and MMedBench datasets, demonstrating superior performance against other knowledge injection methods. We also explored our method’s scalability across different training corpus sizes, laying the foundation to enhance domain-specific LLMs with better data utilization.
Class Distillation with Mahalanobis Contrast: An Efficient Training Paradigm for Pragmatic Language Understanding Tasks
PDF ↗Detecting deviant language such as sexism, or nuanced language such as metaphors or sarcasm, is crucial for enhancing the safety, clarity, and interpretation of social interactions. While existing classifiers deliver strong results on these tasks, they often come with significant computational cost and high data demands. In this work, we propose Class Distillation (ClaD), a novel training paradigm that targets the core challenge: distilling a small, well-defined target class from a highly diverse and heterogeneous background. ClaD integrates two key innovations: (i) a loss function informed by the structural properties of class distributions, based on Mahalanobis distance, and (ii) an interpretable decision algorithm optimized for class separation. Across three benchmark detection tasks – sexism, metaphor, and sarcasm – ClaD outperforms competitive baselines, and even with smaller language models and orders of magnitude fewer parameters, achieves performance comparable to several large language models. These results demonstrate ClaD as an efficient tool for pragmatic language understanding tasks that require gleaning a small target class from a larger heterogeneous background.
Generating Wikipedia articles autonomously is a challenging task requiring the integration of accurate, comprehensive, and well-structured information from diverse sources. This paper introduces the Memory Organization-based Generation (MOG) framework, a novel approach to address these challenges by leveraging a hierarchical memory architecture. MOG extracts fine-grained memory units from web documents, recursively organizes them into a Wikipedia-style hierarchical structure, and uses this structure to guide the generation process. This ensures alignment between memory and the article outline, improving both informativeness and verifiability while minimizing hallucinations. Additionally, a citation module is implemented to enhance traceability by linking every generated sentence to specific memory units. Evaluations on our newly created WikiStart dataset demonstrate that MOG outperforms baseline methods in producing informative and reliable articles, making it particularly robust in real-world scenarios.
Tree-of-Debate: Multi-Persona Debate Trees Elicit Critical Thinking for Scientific Comparative Analysis
PDF ↗With the exponential growth of research facilitated by modern technology and improved accessibility, scientific discoveries have become increasingly fragmented within and across fields. This makes it challenging to assess the significance, novelty, incremental findings, and equivalent ideas between related works, particularly those from different research communities. Large language models (LLMs) have recently demonstrated strong quantitative and qualitative reasoning abilities, and multi-agent LLM debates have shown promise in handling complex reasoning tasks by exploring diverse perspectives and reasoning paths. Inspired by this, we introduce Tree-of-Debate (ToD), a framework which converts scientific papers into LLM personas that debate their respective novelties. To emphasize structured, critical reasoning rather than focusing solely on outcomes, ToD dynamically constructs a debate tree, enabling fine-grained analysis of independent novelty arguments within scholarly articles. Through experiments on scientific literature across various domains, evaluated by expert researchers, we demonstrate that ToD generates informative arguments, effectively contrasts papers, and supports researchers in their literature review.
Bypass Back-propagation: Optimization-based Structural Pruning for Large Language Models via Policy Gradient
PDF ↗Recent Large-Language Models (LLMs) pruning methods typically operate at the post-training phase without the expensive weight finetuning, however, their pruning criteria often rely on **heuristically hand-crafted metrics**, potentially leading to suboptimal performance. We instead propose a novel **optimization-based structural pruning** that learns the pruning masks in a probabilistic space directly by optimizing the loss of the pruned model. To preserve the efficiency, our method **eliminates the back-propagation** through the LLM *per se* during the optimization, requiring only **the forward pass of the LLM**. We achieve this by learning an underlying Bernoulli distribution to sample binary pruning masks, where we decouple the Bernoulli parameters from the LLM loss, thus facilitating an efficient optimization via *policy gradient estimator* without back-propagation. As a result, our method is able to 1) *support global and heterogeneous pruning* (*i.e.*, our method automatically determines different redundancy for different layers), and 2) *optionally initialize with a metric-based method* (for our Bernoulli distributions). Extensive experiments conducted on LLaMA, LLaMA-2, LLaMA-3, Vicuna, and Mistral models using the C4 and WikiText2 datasets demonstrate the promising performance of our method in efficiency and effectiveness.
Large Language Models (LLMs) are increasingly utilized in scientific research assessment, particularly in automated paper review. However, existing LLM-based review systems face significant challenges, including limited domain expertise, hallucinated reasoning, and a lack of structured evaluation. To address these limitations, we introduce DeepReview, a multi-stage framework designed to emulate expert reviewers by incorporating structured analysis, literature retrieval, and evidence-based argumentation. Using DeepReview-13K, a curated dataset with structured annotations, we train DeepReviewer-14B, which outperforms CycleReviewer-70B with fewer tokens. In its best mode, DeepReviewer-14B achieves win rates of 88.21% and 80.20% against GPT-o1 and DeepSeek-R1 in evaluations. Our work sets a new benchmark for LLM-based paper review, with all resources publicly available.
Evaluating machine-generated text remains a significant challenge in NLP, especially for non-English languages. Current methodologies, including automated metrics, human assessments, and LLM-based evaluations, predominantly focus on English, revealing a significant gap in multilingual evaluation frameworks. We introduce the Cross Lingual Auto Evaluation (CIA) Suite, an extensible framework that includes evaluator LLMs (Hercule) and a novel test set (Recon) specifically designed for multilingual evaluation. Our test set features 500 human-annotated instructions spanning various task capabilities along with human judgment scores across six languages. This would enable benchmarking of general-purpose multilingual LLMs and facilitate meta-evaluation of Evaluator LLMs. The proposed model, Hercule, is a cross-lingual evaluation model that addresses the scarcity of reference answers in the target language by learning to assign scores to responses based on easily available reference answers in English. Our experiments demonstrate that Hercule aligns more closely with human judgments compared to proprietary models, demonstrating the effectiveness of such cross-lingual evaluation in low resource scenarios. Further, it is also effective in zero-shot evaluation on unseen languages. This study is the first comprehensive examination of cross-lingual evaluation using LLMs, presenting a scalable and effective approach for multilingual assessment. All code, datasets, and models will be publicly available to enable further research in this important area.
Multi-document (MD) processing is crucial for LLMs to handle real-world tasks such as summarization and question-answering across large sets of documents. While LLMs have improved at processing long inputs, MD contexts still present unique difficulties, including management of inter-document dependencies, redundancy, and incoherent structures. To address this challenge, we introduce MDCure, a scalable and effective instruction data generation framework to enhance the MD capabilities of LLMs without the computational cost of pre-training or reliance on human-annotated data. MDCure generates high-quality synthetic MD instruction data over sets of articles via targeted prompts. We also introduce MDCureRM, a cost-effective, MD-specific reward model to score and filter generated data based on their training utility for MD settings. MDCure is compatible with open- and closed-source models in addition to policy optimization methods such as PPO, enabling even small open- source models to surpass proprietary LLMs as strong generators of high-quality MD instruction data without further data filtering. With MDCure, we fine-tune a wide variety of LLMs up to 70B parameters in size from the FlanT5, Qwen2, and LLAMA3.1 model families. Extensive evaluations on a wide range of MD and long-context benchmarks spanning various tasks and domains show MDCure consistently improves performance over pre-trained baselines and base models by up to 75.1%.
Some statements have one well-defined continuation (e.g., “the Eiffel Tower is in [Paris]"), whereas others have a natural distribution over multiple options (e.g., “the weighted coin flip was [Heads/Tails].") We argue that language model (LM) outputs should capture these natural distributions. Our work specifically tests whether LM output probabilities are calibrated to numeric information within their textual contexts. For example, if the context (the prompt) concerns two equally likely options (e.g., heads or tails for a fair coin), the LM output probabilities should also be equal. Likewise, in a context with nonuniformly likely events (e.g., rolling a pair with two dice) an LM should output proportionate probabilities. However, we find that even in simple settings, the best LMs (1) are poorly calibrated and (2) have systematic biases: artifacts like word identity, word order, and word frequency all impact calibration. For example, ‘gpt-4o-mini‘ often picks the first of two options presented in the prompt regardless of the options’ implied likelihoods, whereas ‘Llama-3.1-8B‘ picks the second. Models do not allocate probability mass among valid options in a calibrated manner.
With a combination of quantitative experiments, human judgments, and qualitative analyses, we evaluate the quantity and quality of African American Language (AAL) representation in 12 predominantly English, open-source pretraining corpora. We specifically focus on the sources, variation, and naturalness of included AAL texts representing the AAL speaking community. We find that AAL is underrepresented in all evaluated pretraining corpora compared to US demographics, constituting as few as 0.007% and at most 0.18% of documents. We also find that more than 25% of AAL texts in C4 may be perceived as inappropriate for LLMs to generate and to reinforce harmful stereotypes. Finally, we find that most automated filters are more likely to conserve White Mainstream English (WME) texts over AAL in pretraining corpora.
How LLMs Comprehend Temporal Meaning in Narratives: A Case Study in Cognitive Evaluation of LLMs
PDF ↗Large language models (LLMs) exihibit increasingly sophisticated linguistic capabilities, yet the extent to which these behaviors reflect human-like cognition versus advanced pattern recognition remains an open question.In this study, we investigate how LLMs process the temporal meaning of linguistic aspect in narratives that were previously used in human studies. Using an Expert-in-the-Loop probing pipeline, we conduct a series of targeted experiments to assess whether LLMs construct semantic representations and pragmatic inferences in a human-like manner.Our findings show that LLMs over-rely on prototypicality, produce inconsistent aspectual judgments, and struggle with causal reasoning derived from aspect, raising concerns about their ability to fully comprehend narratives.These results suggest that LLMs process aspect fundamentally differently from humans and lack robust narrative understanding.Beyond these empirical findings, we develop a standardized experimental framework for the reliable assessment of LLMs’ cognitive and linguistic capabilities.
DRPruning: Efficient Large Language Model Pruning through Distributionally Robust Optimization
PDF ↗Large language models (LLMs) deliver impressive results but face challenges from increasing model sizes and computational costs. Structured pruning reduces model size and speeds up inference but often causes uneven degradation across domains, leading to biased performance. To address this, we propose *DRPruning*, a method that dynamically adjusts the data distribution during training to restore balanced performance across heterogeneous and multi-tasking data. Experiments in monolingual and multilingual settings show that DRPruning surpasses similarly sized models in both pruning and continued pretraining over perplexity, downstream tasks, and instruction tuning. Further analysis demonstrates the robustness of DRPruning towards various domains and distribution shifts. Furthermore, DRPruning can determine optimal reference losses and data ratios automatically, suggesting potential for broader applications. Code and scripts are available at https://github.com/hexuandeng/DRPruning.
Motion instruction is a crucial task that helps athletes refine their technique by analyzing movements and providing corrective guidance. Although recent advances in multimodal models have improved motion understanding,generating precise and sport-specific instruction remains challenging due to the highly domain-specific nature of sports and the need for informative guidance. We propose CoachMe, a reference-based model that analyzes the differences between a learner’s motion and a reference under temporal and physical aspects. This approach enables both domain-knowledge learning and the acquisition of a coach-like thinking process that identifies movement errors effectively and provides feedback to explain how to improve. In this paper, weillustrate how CoachMe adapts well to specific sports such as skating and boxing by learning from general movements and then leveraging limited data. Experiments show that CoachMe provides high-quality instructions instead of directions merely in the tone of a coach but without critical information. CoachMe outperforms GPT-4o by 31.6% in G-Eval on figure skating and by 58.3% on boxing. Analysisfurther confirms that it elaborates on errors and their corresponding improvement methods in the generated instructions. You can find CoachMe here: https://motionxperts.github.io/
Open-source Large Language Models (LLMs) often employ safety alignment methods to resist harmful instructions. However, recent research shows that maliciously fine-tuning these LLMs on harmful data can easily bypass these safeguards. To counter this, we theoretically uncover why malicious fine-tuning succeeds and identify potential defense strategies. Building on the theoretical analysis, we introduce the Self-Degraded Defense (SDD) framework. SDD encourages LLMs to produce high-quality but irrelevant responses to harmful prompts. When attackers attempt malicious fine-tuning, the general capability of the LLM aligned by SDD will significantly decrease, rendering it incapable of following harmful instructions. Our experimental results confirm SDD’s effectiveness against such attacks.Our code is available at https://github.com/ZeroNLP/SDD.
Vision-language models (VLMs) excel at extracting and reasoning about information from images. Yet, their capacity to leverage internal knowledge about specific entities remains underexplored. This work investigates the disparity in model performance when answering factual questions about an entity described in text versus depicted in an image. Our results reveal a significant accuracy drop — reaching 18% for some models — when the entity is presented visually instead of textually. To study this gap we present PopVQA, a dataset which allows separating entity recognition and question answering, and use it to benchmark several models. We hypothesize that this decline arises from limitations in how information flows from image tokens to query tokens. Thus, we use mechanistic interpretability tools to reveal that, although image tokens are preprocessed by the vision encoder, meaningful information flow from these tokens occurs only in the much deeper layers. Furthermore, critical image processing happens in the language model’s middle layers, allowing few layers for consecutive reasoning, highlighting a potential inefficiency in how the model utilizes its layers for reasoning. These insights shed light on the internal mechanics of VLMs and offer pathways for enhancing their reasoning capabilities. PopVQA can be found at https://huggingface.co/datasets/idoco/PopVQA.
Diversity Explains Inference Scaling Laws: Through a Case Study of Minimum Bayes Risk Decoding
PDF ↗Inference methods play an important role in eliciting the performance of large language models (LLMs). Currently, LLMs use inference methods utilizing generated multiple samples, which can be derived from Minimum Bayes Risk (MBR) Decoding. Previous studies have conducted empirical analyses to clarify the improvements in generation performance achieved by MBR decoding and have reported various observations. However, the theoretical underpinnings of these findings remain uncertain. To address this, we offer a new theoretical interpretation of MBR decoding from the perspective of bias–diversity decomposition. In this interpretation, the error in the quality estimation of hypotheses by MBR decoding is decomposed into two main factors: bias, which considers the closeness between the utility function and human evaluation, and diversity, which represents the variability in the quality estimation of the utility function. The theoretical analysis reveals the difficulty of simultaneously improving bias and diversity, confirming the validity of enhancing MBR decoding performance by increasing diversity. Furthermore, we reveal that diversity can explain one aspect of inference scaling laws that describe performance improvement by increasing sample size. Moreover, experiments across multiple NLP tasks yielded results consistent with these theoretical characteristics. Our code is available at https://github.com/naist-nlp/mbr-bias-diversity.
Language Models, Graph Searching, and Supervision Adulteration: When More Supervision is Less and How to Make More More
PDF ↗This work concerns the path-star task, a minimal example of searching over a graph. The graph, G, is star-shaped with D arms radiating from a start node, s. A language model (LM) is given G, s, and a target node, t, which ends one of the arms and is tasked with generating the arm containing t. The minimal nature of this task means only a single choice needs to be made: which of the arms contains?Decoder-only LMs fail to solve this elementary task above 1/D chance due to a learned shortcut that absorbs training supervision. We show how this pathology is caused by excess supervision and present a series of solutions demonstrating that the task is solvable via decoder-only LMs. We find that the task’s minimal nature causes its difficulty, as it prevents task decomposition. Our solutions provide insight into the pathology and its implications for LMs trained via next-token prediction.
Towards Better Value Principles for Large Language Model Alignment: A Systematic Evaluation and Enhancement
PDF ↗As Large Language Models (LLMs) advance, aligning them with human values is critical for their responsible development. Value principles serve as the foundation for clarifying alignment goals.Multiple sets of value principles have been proposed, such as HHH (helpful, honest, harmless) and instructions for data synthesis in reinforcement learning from AI feedback (RLAIF). However, most of them are heuristically crafted, without consideration of three primary challenges in practical LLM alignment: 1) Comprehensiveness to deal with diverse and even unforeseen scenarios in which LLMs could be applied; 2) Precision to provide LLMs with clear and actionable guidance in specific scenarios; and 3) Compatability to avoid internal contracts between principles.In this paper, we formalize quantitative metrics to evaluate value principles along the three desirable properties. Building on these metrics, we propose the Hierarchical Value Principle framework (HiVaP), which constructs a hierarchical principle set and retrieves principles tailored to each scenario in a cascading way, addressing above challenges.Experimental results validate that the three metrics capture the effectiveness of value principles for LLM alignment, and our HiVaP framework that enhances these metrics leads to superior alignment. Warning: This paper contains several toxic and offensive statements.
Attributed text generation aims to enhance the reliability of content generated from large language models by providing citations for each claim, which thereby enables users to easily verify the correctness of the responses.However, the scarcity of high-quality training samples presents a significant challenge in aligning large language models to generate texts with citations, revealing considerable room for improvement in existing attribution systems.Besides, existing approaches of aligning large language models to follow user instructions can lead to an undue emphasis on irrelevant documents, which in turn reduces the quality of responses.To address the above problems, we propose Automatic Alignment Framework for Attributed Text Generation ( \mathcal{A}^3), a novel framework designed to automatically generate high-quality attributed query-response pairs for both supervised fine-tuning and preference optimization stages without human annotation.With the help of \mathcal{A}^3, Mistral-7B can achieve a citation recall of 84.4 and a precision of 87.0 precision on ASQA, which notably surpasses GPT-4’s citation recall of 73.0 and precision of 76.5.