Authors: Yan Li, Tianyi Zhang, Zechuan Li, Soyeon Caren Han
Abstract: Transformer-based Large Language Models (LLMs) struggle to process inputs exceeding their training context window, with performance degrading due to positional out-of-distribution (O.O.D.) that disrupt attention computations. Existing solutions, fine-tuning and training-free methods, are limited by computational inefficiency, attention logit outliers or loss of local positional information. To address this, we propose Greedy Attention Logit Interpolation (GALI), a training-free length extrapolation method that maximizes the utilization of pretrained positional intervals while avoiding attention logit outliers through attention logit interpolation. The result demonstrates that GALI consistently outperforms state-of-the-art training-free methods. Our findings reveal that LLMs interpret positional intervals unevenly within their training context window, suggesting that extrapolating within a smaller positional interval range yields superior results-even for short-context tasks. GALI represents a significant step toward resolving the positional O.O.D. challenge, enabling more reliable long-text understanding in LLMs. Our implementation of GALI, along with the experiments from our paper, is open-sourced at https://github.com/AcademyCityL/GALI.
Authors: Mayuka Jayawardhana (University of Maryland), Renbo Tu (University of Toronto), Samuel Dooley (Meta), Valeriia Cherepanova (Amazon), Andrew Gordon Wilson (New York University), Frank Hutter (University of Freiburg), Colin White (Abacus.AI), Tom Goldstein (University of Maryland), Micah Goldblum (Columbia University)
Abstract: Large language models (LLMs) perform remarkably well on tabular datasets in zero- and few-shot settings, since they can extract meaning from natural language column headers that describe features and labels. Similarly, TabPFN, a recent non-LLM transformer pretrained on numerous tables for in-context learning, has demonstrated excellent performance for dataset sizes up to a thousand samples. In contrast, gradient-boosted decision trees (GBDTs) are typically trained from scratch on each dataset without benefiting from pretraining data and must learn the relationships between columns from their entries alone since they lack natural language understanding. LLMs and TabPFN excel on small tabular datasets where a strong prior is essential, yet they are not competitive with GBDTs on medium or large datasets, since their context lengths are limited. In this paper, we propose a simple and lightweight approach for fusing large language models and TabPFN with gradient-boosted decision trees, which allows scalable GBDTs to benefit from the natural language capabilities and pretraining of transformers. We name our fusion methods LLM-Boost and PFN-Boost, respectively. While matching or surpassing the performance of the transformer at sufficiently small dataset sizes and GBDTs at sufficiently large sizes, LLM-Boost and PFN-Boost outperform both standalone components on a wide range of dataset sizes in between. We demonstrate state-of-the-art performance against numerous baselines and ensembling algorithms. We find that PFN-Boost achieves the best average performance among all methods we test for all but very small dataset sizes. We release our code at http://github.com/MayukaJ/LLM-Boost .
Authors: Michael Galarnyk, Agam Shah, Dipanwita Guhathakurta, Poojitha Nandigam, Sudheer Chava
Abstract: This paper discusses and contains offensive content. Language models (LMs) are used in decision-making systems and as interactive assistants. However, how well do these models making judgements align with the diversity of human values, particularly regarding social and moral norms? In this work, we investigate how inclusively LMs perceive norms across demographic groups (e.g., gender, age, and income). We prompt 11 LMs on rules-of-thumb (RoTs) and compare their outputs with the existing responses of 100 human annotators. We introduce the Absolute Distance Alignment Metric (ADA-Met) to quantify alignment on ordinal questions. We find notable disparities in LM responses, with younger, higher-income groups showing closer alignment, raising concerns about the representation of marginalized perspectives. Our findings highlight the importance of further efforts to make LMs more inclusive of diverse human values. The code and prompts are available on GitHub under the CC BY-NC 4.0 license.
Authors: Shenran Wang, Changbing Yang, Mike Parkhill, Chad Quinn, Christopher Hammerly, Jian Zhu
Abstract: We present lightweight flow matching multilingual text-to-speech (TTS) systems for Ojibwe, Mi'kmaq, and Maliseet, three Indigenous languages in North America. Our results show that training a multilingual TTS model on three typologically similar languages can improve the performance over monolingual models, especially when data are scarce. Attention-free architectures are highly competitive with self-attention architecture with higher memory efficiency. Our research not only advances technical development for the revitalization of low-resource languages but also highlights the cultural gap in human evaluation protocols, calling for a more community-centered approach to human evaluation.
Authors: Iker Garc\'ia-Ferrero
Abstract: Natural Language Processing (NLP) has seen remarkable advances in recent years, particularly with the emergence of Large Language Models that have achieved unprecedented performance across many tasks. However, these developments have mainly benefited a small number of high-resource languages such as English. The majority of languages still face significant challenges due to the scarcity of training data and computational resources. To address this issue, this thesis focuses on cross-lingual transfer learning, a research area aimed at leveraging data and models from high-resource languages to improve NLP performance for low-resource languages. Specifically, we focus on Sequence Labeling tasks such as Named Entity Recognition, Opinion Target Extraction, and Argument Mining. The research is structured around three main objectives: (1) advancing data-based cross-lingual transfer learning methods through improved translation and annotation projection techniques, (2) developing enhanced model-based transfer learning approaches utilizing state-of-the-art multilingual models, and (3) applying these methods to real-world problems while creating open-source resources that facilitate future research in low-resource NLP. More specifically, this thesis presents a new method to improve data-based transfer with T-Projection, a state-of-the-art annotation projection method that leverages text-to-text multilingual models and machine translation systems. T-Projection significantly outperforms previous annotation projection methods by a wide margin. For model-based transfer, we introduce a constrained decoding algorithm that enhances cross-lingual Sequence Labeling in zero-shot settings using text-to-text models. Finally, we develop Medical mT5, the first multilingual text-to-text medical model, demonstrating the practical impact of our research on real-world applications.
Authors: Loubna Ben Allal, Anton Lozhkov, Elie Bakouch, Gabriel Mart\'in Bl\'azquez, Guilherme Penedo, Lewis Tunstall, Andr\'es Marafioti, Hynek Kydl\'i\v{c}ek, Agust\'in Piqueres Lajar\'in, Vaibhav Srivastav, Joshua Lochner, Caleb Fahlgren, Xuan-Son Nguyen, Cl\'ementine Fourrier, Ben Burtenshaw, Hugo Larcher, Haojun Zhao, Cyril Zakka, Mathieu Morlon, Colin Raffel, Leandro von Werra, Thomas Wolf
Abstract: While large language models have facilitated breakthroughs in many applications of artificial intelligence, their inherent largeness makes them computationally expensive and challenging to deploy in resource-constrained settings. In this paper, we document the development of SmolLM2, a state-of-the-art "small" (1.7 billion parameter) language model (LM). To attain strong performance, we overtrain SmolLM2 on ~11 trillion tokens of data using a multi-stage training process that mixes web text with specialized math, code, and instruction-following data. We additionally introduce new specialized datasets (FineMath, Stack-Edu, and SmolTalk) at stages where we found existing datasets to be problematically small or low-quality. To inform our design decisions, we perform both small-scale ablations as well as a manual refinement process that updates the dataset mixing rates at each stage based on the performance at the previous stage. Ultimately, we demonstrate that SmolLM2 outperforms other recent small LMs including Qwen2.5-1.5B and Llama3.2-1B. To facilitate future research on LM development as well as applications of small LMs, we release both SmolLM2 as well as all of the datasets we prepared in the course of this project.
Authors: Amirhossein Dabiriaghdam, Lele Wang
Abstract: The rapid proliferation of large language models (LLMs) has created an urgent need for reliable methods to detect whether a text is generated by such models. In this paper, we propose SimMark, a posthoc watermarking algorithm that makes LLMs' outputs traceable without requiring access to the model's internal logits, enabling compatibility with a wide range of LLMs, including API-only models. By leveraging the similarity of semantic sentence embeddings and rejection sampling to impose detectable statistical patterns imperceptible to humans, and employing a soft counting mechanism, SimMark achieves robustness against paraphrasing attacks. Experimental results demonstrate that SimMark sets a new benchmark for robust watermarking of LLM-generated content, surpassing prior sentence-level watermarking techniques in robustness, sampling efficiency, and applicability across diverse domains, all while preserving the text quality.
Authors: Jingyu Liu, Beidi Chen, Ce Zhang
Abstract: Improving time-to-first-token (TTFT) is an essentially important objective in modern large language model (LLM) inference engines. Because optimizing TTFT directly results in higher maximal QPS and meets the requirements of many critical applications. However, boosting TTFT is notoriously challenging since it is purely compute-bounded and the performance bottleneck shifts from the self-attention to the MLP part. We present SpecPrefill, a training free framework that accelerates the inference TTFT for both long and medium context queries based on the following insight: LLMs are generalized enough to still preserve the quality given only a carefully chosen subset of prompt tokens. At its core, SpecPrefill leverages a lightweight model to speculate locally important tokens based on the context. These tokens, along with the necessary positional information, are then sent to the main model for processing. We evaluate SpecPrefill with a diverse set of tasks, followed by a comprehensive benchmarking of performance improvement both in a real end-to-end setting and ablation studies. SpecPrefill manages to serve Llama-3.1-405B-Instruct-FP8 with up to $7\times$ maximal end-to-end QPS on real downstream tasks and $7.66\times$ TTFT improvement during benchmarking.
Authors: Yizhe Yang, Palakorn Achananuparp, Heyan Huang, Jing Jiang, John Pinto, Jenny Giam, Kit Phey Leng, Nicholas Gabriel Lim, Cameron Tan Shi Ern, Ee-peng Lim
Abstract: Simulating human clients in mental health counseling is crucial for training and evaluating counselors (both human or simulated) in a scalable manner. Nevertheless, past research on client simulation did not focus on complex conversation tasks such as mental health counseling. In these tasks, the challenge is to ensure that the client's actions (i.e., interactions with the counselor) are consistent with with its stipulated profiles and negative behavior settings. In this paper, we propose a novel framework that supports consistent client simulation for mental health counseling. Our framework tracks the mental state of a simulated client, controls its state transitions, and generates for each state behaviors consistent with the client's motivation, beliefs, preferred plan to change, and receptivity. By varying the client profile and receptivity, we demonstrate that consistent simulated clients for different counseling scenarios can be effectively created. Both our automatic and expert evaluations on the generated counseling sessions also show that our client simulation method achieves higher consistency than previous methods.
Authors: Yizhe Yang, Palakorn Achananuparp, Heyan Huang, Jing Jiang, Kit Phey Leng, Nicholas Gabriel Lim, Cameron Tan Shi Ern, Ee-peng Lim
Abstract: Conversational counselor agents have become essential tools for addressing the rising demand for scalable and accessible mental health support. This paper introduces CAMI, a novel automated counselor agent grounded in Motivational Interviewing (MI) -- a client-centered counseling approach designed to address ambivalence and facilitate behavior change. CAMI employs a novel STAR framework, consisting of client's state inference, motivation topic exploration, and response generation modules, leveraging large language models (LLMs). These components work together to evoke change talk, aligning with MI principles and improving counseling outcomes for clients from diverse backgrounds. We evaluate CAMI's performance through both automated and manual evaluations, utilizing simulated clients to assess MI skill competency, client's state inference accuracy, topic exploration proficiency, and overall counseling success. Results show that CAMI not only outperforms several state-of-the-art methods but also shows more realistic counselor-like behavior. Additionally, our ablation study underscores the critical roles of state inference and topic exploration in achieving this performance.
Authors: Yibo Yan, Shen Wang, Jiahao Huo, Jingheng Ye, Zhendong Chu, Xuming Hu, Philip S. Yu, Carla Gomes, Bart Selman, Qingsong Wen
Abstract: Scientific reasoning, the process through which humans apply logic, evidence, and critical thinking to explore and interpret scientific phenomena, is essential in advancing knowledge reasoning across diverse fields. However, despite significant progress, current scientific reasoning models still struggle with generalization across domains and often fall short of multimodal perception. Multimodal Large Language Models (MLLMs), which integrate text, images, and other modalities, present an exciting opportunity to overcome these limitations and enhance scientific reasoning. Therefore, this position paper argues that MLLMs can significantly advance scientific reasoning across disciplines such as mathematics, physics, chemistry, and biology. First, we propose a four-stage research roadmap of scientific reasoning capabilities, and highlight the current state of MLLM applications in scientific reasoning, noting their ability to integrate and reason over diverse data types. Second, we summarize the key challenges that remain obstacles to achieving MLLM's full potential. To address these challenges, we propose actionable insights and suggestions for the future. Overall, our work offers a novel perspective on MLLM integration with scientific reasoning, providing the LLM community with a valuable vision for achieving Artificial General Intelligence (AGI).
Authors: Daniel Gahler, Dean Thomas, Slawomir Lach, Leroy Cronin
Abstract: The most fundamental abstraction underlying all modern computers is the Turing Machine, that is if any modern computer can simulate a Turing Machine, an equivalence which is called Turing completeness, it is theoretically possible to achieve any task that can be algorithmically described by executing a series of discrete unit operations. In chemistry, the ability to program chemical processes is demanding because it is hard to ensure that the process can be understood at a high level of abstraction, and then reduced to practice. Herein we exploit the concept of Turing completeness applied to robotic platforms for chemistry that can be used to synthesise complex molecules through unit operations that execute chemical processes using a chemically-aware programming language, XDL. We leverage the concept of computability by computers to synthesizability of chemical compounds by automated synthesis machines. The results of an interactive demonstration of Turing completeness using the colour gamut and conditional logic are presented and examples of chemical use-cases are discussed. Over 16.7 million combinations of Red, Green, Blue (RGB) colour space were binned into 5 discrete values and measured over 10 regions of interest (ROIs), affording 78 million possible states per step and served as a proxy for conceptual, chemical space exploration. This formal description establishes a formal framework in future chemical programming languages to ensure complex logic operations are expressed and executed correctly, with the possibility of error correction, in the automated and autonomous pursuit of increasingly complex molecules.
Authors: Yejian Zhang, Shingo Takada
Abstract: With the internet's evolution, consumers increasingly rely on online reviews for service or product choices, necessitating that businesses analyze extensive customer feedback to enhance their offerings. While machine learning-based sentiment classification shows promise in this realm, its technical complexity often bars small businesses and individuals from leveraging such advancements, which may end up making the competitive gap between small and large businesses even bigger in terms of improving customer satisfaction. This paper introduces an approach that integrates large language models (LLMs), specifically Generative Pre-trained Transformer (GPT) and Bidirectional Encoder Representations from Transformers (BERT)-based models, making it accessible to a wider audience. Our experiments across various datasets confirm that our approach retains high classification accuracy without the need for manual labeling, expert knowledge in tuning and data annotation, or substantial computational power. By significantly lowering the barriers to applying sentiment classification techniques, our methodology enhances competitiveness and paves the way for making machine learning technology accessible to a broader audience.
Authors: Bradley P. Allen, Paul T. Groth
Abstract: Evaluating large language models (LLMs) for tasks like fact extraction in support of knowledge graph construction frequently involves computing accuracy metrics using a ground truth benchmark based on a knowledge graph (KG). These evaluations assume that errors represent factual disagreements. However, human discourse frequently features metalinguistic disagreement, where agents differ not on facts but on the meaning of the language used to express them. Given the complexity of natural language processing and generation using LLMs, we ask: do metalinguistic disagreements occur between LLMs and KGs? Based on an investigation using the T-REx knowledge alignment dataset, we hypothesize that metalinguistic disagreement does in fact occur between LLMs and KGs, with potential relevance for the practice of knowledge graph engineering. We propose a benchmark for evaluating the detection of factual and metalinguistic disagreements between LLMs and KGs. An initial proof of concept of such a benchmark is available on Github.
Authors: Sahil Manchanda, Pannaga Shivaswamy
Abstract: Text-embedding models often exhibit biases arising from the data on which they are trained. In this paper, we examine a hitherto unexplored bias in text-embeddings: bias arising from the presence of $\textit{names}$ such as persons, locations, organizations etc. in the text. Our study shows how the presence of $\textit{name-bias}$ in text-embedding models can potentially lead to erroneous conclusions in assessment of thematic similarity.Text-embeddings can mistakenly indicate similarity between texts based on names in the text, even when their actual semantic content has no similarity or indicate dissimilarity simply because of the names in the text even when the texts match semantically. We first demonstrate the presence of name bias in different text-embedding models and then propose $\textit{text-anonymization}$ during inference which involves removing references to names, while preserving the core theme of the text. The efficacy of the anonymization approach is demonstrated on two downstream NLP tasks, achieving significant performance gains. Our simple and training-optimization-free approach offers a practical and easily implementable solution to mitigate name bias.
Authors: T. Chay-intr, Y. Chen, K. Viriyayudhakorn, T. Theeramunkong
Abstract: We present LLaVAC, a method for constructing a classifier for multimodal sentiment analysis. This method leverages fine-tuning of the Large Language and Vision Assistant (LLaVA) to predict sentiment labels across both image and text modalities. Our approach involves designing a structured prompt that incorporates both unimodal and multimodal labels to fine-tune LLaVA, enabling it to perform sentiment classification effectively. Experiments on the MVSA-Single dataset demonstrate that LLaVAC outperforms existing methods in multimodal sentiment analysis across three data processing procedures. The implementation of LLaVAC is publicly available at https://github.com/tchayintr/llavac.
Authors: Ziwei Wang, Jie Zhou, Qin Chen, Min Zhang, Bo Jiang, Aimin Zhou, Qinchun Bai, Liang He
Abstract: The knowledge tracing (KT) problem is an extremely important topic in personalized education, which aims to predict whether students can correctly answer the next question based on their past question-answer records. Prior work on this task mainly focused on learning the sequence of behaviors based on the IDs or textual information. However, these studies usually fail to capture students' sufficient behavioral patterns without reasoning with rich world knowledge about questions. In this paper, we propose a large language models (LLMs)-based framework for KT, named \texttt{\textbf{LLM-KT}}, to integrate the strengths of LLMs and traditional sequence interaction models. For task-level alignment, we design Plug-and-Play instruction to align LLMs with KT, leveraging LLMs' rich knowledge and powerful reasoning capacity. For modality-level alignment, we design the plug-in context and sequence to integrate multiple modalities learned by traditional methods. To capture the long context of history records, we present a plug-in context to flexibly insert the compressed context embedding into LLMs using question-specific and concept-specific tokens. Furthermore, we introduce a plug-in sequence to enhance LLMs with sequence interaction behavior representation learned by traditional sequence models using a sequence adapter. Extensive experiments show that \texttt{\textbf{LLM-KT}} obtains state-of-the-art performance on four typical datasets by comparing it with approximately 20 strong baselines.
Authors: Qinzhuo Wu, Wei Liu, Jian Luan, Bin Wang
Abstract: Recently, mobile AI agents have gained increasing attention. Given a task, mobile AI agents can interact with mobile devices in multiple steps and finally form a GUI flow that solves the task. However, existing agents tend to focus on most task-relevant elements at each step, leading to local optimal solutions and ignoring the overall GUI flow. To address this issue, we constructed a training dataset called MobileReach, which breaks the task into page reaching and operation subtasks. Furthermore, we propose ReachAgent, a two-stage framework that focuses on improving its task-completion abilities. It utilizes the page reaching and page operation subtasks, along with reward-based preference GUI flows, to further enhance the agent. Experimental results show that ReachAgent significantly improves the IoU Acc and Text Acc by 7.12% and 7.69% on the step-level and 4.72% and 4.63% on the task-level compared to the SOTA agent. Our data and code will be released upon acceptance.
Authors: Paul Youssef, Zhixue Zhao, Daniel Braun, J\"org Schl\"otterer, Christin Seifert
Abstract: Large Language Models (LLMs) contain large amounts of facts about the world. These facts can become outdated over time, which has led to the development of knowledge editing methods (KEs) that can change specific facts in LLMs with limited side effects. This position paper argues that editing LLMs poses serious safety risks that have been largely overlooked. First, we note the fact that KEs are widely available, computationally inexpensive, highly performant, and stealthy makes them an attractive tool for malicious actors. Second, we discuss malicious use cases of KEs, showing how KEs can be easily adapted for a variety of malicious purposes. Third, we highlight vulnerabilities in the AI ecosystem that allow unrestricted uploading and downloading of updated models without verification. Fourth, we argue that a lack of social and institutional awareness exacerbates this risk, and discuss the implications for different stakeholders. We call on the community to (i) research tamper-resistant models and countermeasures against malicious model editing, and (ii) actively engage in securing the AI ecosystem.
Authors: Renjun Hu, Yi Cheng, Libin Meng, Jiaxin Xia, Yi Zong, Xing Shi, Wei Lin
Abstract: The rapid advancement of large language models (LLMs) has opened new possibilities for their adoption as evaluative judges. This paper introduces Themis, a fine-tuned LLM judge that delivers sophisticated context-aware evaluations. We provide a comprehensive overview of the development pipeline for Themis, highlighting its scenario-dependent evaluation prompts and two novel methods for controlled instruction generation. These designs enable Themis to effectively distill evaluative skills from teacher models, while retaining flexibility for continuous development. We introduce two human-labeled benchmarks for meta-evaluation, demonstrating that Themis can achieve high alignment with human preferences in an economical manner. Additionally, we explore insights into the LLM-as-a-judge paradigm, revealing nuances in performance and the varied effects of reference answers. Notably, we observe that pure knowledge distillation from strong LLMs, though common, does not guarantee performance improvement through scaling. We propose a mitigation strategy based on instruction-following difficulty. Furthermore, we provide practical guidelines covering data balancing, prompt customization, multi-objective training, and metric aggregation. We aim for our method and findings, along with the fine-tuning data, benchmarks, and model checkpoints, to support future research and development in this area.
Authors: Seonok Kim
Abstract: Large Language Models (LLMs) have demonstrated impressive capabilities across natural language processing tasks. However, their application to specialized domains such as medicine and biology requires further optimization to ensure factual accuracy, reliability, and contextual depth. We introduce MedBioLM, a domain-adapted biomedical question-answering model designed to enhance both short-form and long-form queries. By integrating fine-tuning and retrieval-augmented generation (RAG), MedBioLM dynamically incorporates domain-specific knowledge, improving reasoning abilities and factual accuracy. To evaluate its effectiveness, we fine-tuned the model on diverse biomedical QA datasets, covering structured multiple-choice assessments and complex clinical reasoning tasks. Fine-tuning significantly improves accuracy on benchmark datasets, while RAG enhances factual consistency. These results highlight the potential of domain-optimized LLMs in advancing biomedical research, medical education, and clinical decision support.
Authors: Mohannad Takrouri, Nicol\'as M. Cuadrado, Martin Tak\'a\v{c}
Abstract: Machine learning (ML) is increasingly vital for smart-grid research, yet restricted access to realistic, diverse data - often due to privacy concerns - slows progress and fuels doubts within the energy sector about adopting ML-based strategies. We propose integrating Large Language Models (LLMs) in energy modeling to generate realistic, culturally sensitive, and behavior-specific data for household energy usage across diverse geographies. In this study, we employ and compare five different LLMs to systematically produce family structures, weather patterns, and daily consumption profiles for households in six distinct countries. A four-stage methodology synthesizes contextual daily data, including culturally nuanced activities, realistic weather ranges, HVAC operations, and distinct `energy signatures' that capture unique consumption footprints. Additionally, we explore an alternative strategy where external weather datasets can be directly integrated, bypassing intermediate weather modeling stages while ensuring physically consistent data inputs. The resulting dataset provides insights into how cultural, climatic, and behavioral factors converge to shape carbon emissions, offering a cost-effective avenue for scenario-based energy optimization. This approach underscores how prompt engineering, combined with knowledge distillation, can advance sustainable energy research and climate mitigation efforts. Source code is available at https://github.com/Singularity-AI-Lab/LLM-Energy-Knowledge-Distillation .
URLs: https://github.com/Singularity-AI-Lab/LLM-Energy-Knowledge-Distillation
Authors: Mariam Nakhl\'e, Marco Dinarelli, Raheel Qader, Emmanuelle Esperan\c{c}a-Rodier, Herv\'e Blanchon
Abstract: Despite the strong research interest in document-level Machine Translation (MT), the test sets dedicated to this task are still scarce. The existing test sets mainly cover topics from the general domain and fall short on specialised domains, such as legal and financial. Also, in spite of their document-level aspect, they still follow a sentence-level logic that does not allow for including certain linguistic phenomena such as information reorganisation. In this work, we aim to fill this gap by proposing a novel test set: DOLFIN. The dataset is built from specialised financial documents, and it makes a step towards true document-level MT by abandoning the paradigm of perfectly aligned sentences, presenting data in units of sections rather than sentences. The test set consists of an average of 1950 aligned sections for five language pairs. We present a detailed data collection pipeline that can serve as inspiration for aligning new document-level datasets. We demonstrate the usefulness and quality of this test set by evaluating a number of models. Our results show that the test set is able to discriminate between context-sensitive and context-agnostic models and shows the weaknesses when models fail to accurately translate financial texts. The test set is made public for the community.
Authors: Aissatou Diallo, Antonis Bikakis, Luke Dickens, Anthony Hunter, Rob Miller
Abstract: While Large Language Models (LLMs) demonstrate impressive reasoning capabilities, understanding and validating their knowledge utilization remains challenging. Chain-of-thought (CoT) prompting partially addresses this by revealing intermediate reasoning steps, but the knowledge flow and application remain implicit. We introduce IAO (Input-Action-Output) prompting, a structured template-based method that explicitly models how LLMs access and apply their knowledge during complex reasoning tasks. IAO decomposes problems into sequential steps, each clearly identifying the input knowledge being used, the action being performed, and the resulting output. This structured decomposition enables us to trace knowledge flow, verify factual consistency, and identify potential knowledge gaps or misapplications. Through experiments across diverse reasoning tasks, we demonstrate that IAO not only improves zero-shot performance but also provides transparency in how LLMs leverage their stored knowledge. Human evaluation confirms that this structured approach enhances our ability to verify knowledge utilization and detect potential hallucinations or reasoning errors. Our findings provide insights into both knowledge representation within LLMs and methods for more reliable knowledge application.
Authors: Jonathan Delena, Augustin Moreau, Dominic Ravensdale, Frederick Chatterton
Abstract: Memory retention mechanisms play a central role in determining the efficiency of computational architectures designed for processing extended sequences. Conventional methods for token management often impose fixed retention thresholds or rely on uniform attention weight distributions, leading to inefficient memory utilization and premature information loss in extended sequence modeling. Structured Token Retention (STR) introduces a probabilistic selection framework that dynamically adjusts token persistence based on contextual significance, ensuring that computational resources are allocated to semantically relevant elements. Computational Memory Paths (CMP) extend this framework through hierarchical memory allocation, refining retention efficiency through structured reallocation of token embeddings. Comparative assessments against baseline models demonstrate that STR and CMP improve token survival rates across long input sequences while reducing cumulative error propagation across processing layers. Experimental results further indicate reductions in computational overhead, improving inference speed without degrading contextual coherence. Token distribution analyses reveal that structured memory allocation prevents excessive redundancy in attention weight calculations, optimizing information retrieval efficiency in large-scale generative architectures. The integration of STR and CMP into an open-source model illustrates the adaptability of structured memory retention methodologies, highlighting their applicability in generative text processing, long-context comprehension, and scalable sequence modeling.
Authors: Felix Schneider (Zoom Communications), Marco Turchi (Zoom Communications), Alex Waibel (Karlsruhe Institute of Technology)
Abstract: With more and more meetings moving to a digital domain, meeting summarization has recently gained interest in both academic and commercial research. However, prior academic research focuses on meeting summarization as an offline task, performed after the meeting concludes. In this paper, we perform the first systematic study of online meeting summarization. For this purpose, we propose several policies for conducting online summarization. We discuss the unique challenges of this task compared to the offline setting and define novel metrics to evaluate latency and partial summary quality. The experiments on the AutoMin dataset show that 1) online models can produce strong summaries, 2) our metrics allow a detailed analysis of different systems' quality-latency trade-off, also taking into account intermediate outputs and 3) adaptive policies perform better than fixed scheduled ones. These findings provide a starting point for the wider research community to explore this important task.
Authors: Zhen Qian, Xiuzhen Zhang, Xiaofei Xu, Feng Xia
Abstract: Number-focused headline generation is a summarization task requiring both high textual quality and precise numerical accuracy, which poses a unique challenge for Large Language Models (LLMs). Existing studies in the literature focus only on either textual quality or numerical reasoning and thus are inadequate to address this challenge. In this paper, we propose a novel chain-of-thought framework for using rationales comprising key elements of the Topic, Entities, and Numerical reasoning (TEN) in news articles to enhance the capability for LLMs to generate topic-aligned high-quality texts with precise numerical accuracy. Specifically, a teacher LLM is employed to generate TEN rationales as supervision data, which are then used to teach and fine-tune a student LLM. Our approach teaches the student LLM automatic generation of rationales with enhanced capability for numerical reasoning and topic-aligned numerical headline generation. Experiments show that our approach achieves superior performance in both textual quality and numerical accuracy.
Authors: Xumeng Wen, Shun Zheng, Zhen Xu, Yiming Sun, Jiang Bian
Abstract: Recent studies have shown that large language models (LLMs), when customized with post-training on tabular data, can acquire general tabular in-context learning (TabICL) capabilities. These models are able to transfer effectively across diverse data schemas and different task domains. However, existing LLM-based TabICL approaches are constrained to few-shot scenarios due to the sequence length limitations of LLMs, as tabular instances represented in plain text consume substantial tokens. To address this limitation and enable scalable TabICL for any data size, we propose retrieval-augmented LLMs tailored to tabular data. Our approach incorporates a customized retrieval module, combined with retrieval-guided instruction-tuning for LLMs. This enables LLMs to effectively leverage larger datasets, achieving significantly improved performance across 69 widely recognized datasets and demonstrating promising scaling behavior. Extensive comparisons with state-of-the-art tabular models reveal that, while LLM-based TabICL still lags behind well-tuned numeric models in overall performance, it uncovers powerful algorithms under limited contexts, enhances ensemble diversity, and excels on specific datasets. These unique properties underscore the potential of language as a universal and accessible interface for scalable tabular data learning.
Authors: Maite Heredia, Jeremy Barnes, Aitor Soroa
Abstract: Code-switching (CS) remains a significant challenge in Natural Language Processing (NLP), mainly due a lack of relevant data. In the context of the contact between the Basque and Spanish languages in the north of the Iberian Peninsula, CS frequently occurs in both formal and informal spontaneous interactions. However, resources to analyse this phenomenon and support the development and evaluation of models capable of understanding and generating code-switched language for this language pair are almost non-existent. We introduce a first approach to develop a naturally sourced corpus for Basque-Spanish code-switching. Our methodology consists of identifying CS texts from previously available corpora using language identification models, which are then manually validated to obtain a reliable subset of CS instances. We present the properties of our corpus and make it available under the name Euska\~nolDS.
Authors: Jialiang Wu, Yi Shen, Sijia Liu, Yi Tang, Sen Song, Xiaoyi Wang, Longjun Cai
Abstract: Despite their impressive capacities, Large language models (LLMs) often struggle with the hallucination issue of generating inaccurate or fabricated content even when they possess correct knowledge. In this paper, we extend the exploration of the correlation between hidden-state prediction changes and output factuality into a deeper, token-wise level. Based on the insights , we propose cross-layer Entropy eNhanced Decoding (END), a decoding method that mitigates hallucinations without requiring extra training. END leverages inner probability changes across layers to individually quantify the factual knowledge required for each candidate token, and adjusts the final predicting distribution to prioritize tokens with higher factuality. Experiments on both hallucination and QA benchmarks demonstrate that END significantly enhances the truthfulness and informativeness of generated content while maintaining robust QA accuracy. Moreover, our work provides a deeper perspective on understanding the correlations between inherent knowledge and output factuality.
Authors: Julius Mayer, Mohamad Ballout, Serwan Jassim, Farbod Nosrat Nezami, Elia Bruni
Abstract: Vision-Language Models (VLMs) are known to struggle with spatial reasoning and visual alignment. To help overcome these limitations, we introduce iVISPAR, an interactive multi-modal benchmark designed to evaluate the spatial reasoning capabilities of VLMs acting as agents. iVISPAR is based on a variant of the sliding tile puzzle-a classic problem that demands logical planning, spatial awareness, and multi-step reasoning. The benchmark supports visual 2D, 3D, and text-based input modalities, enabling comprehensive assessments of VLMs' planning and reasoning skills. We evaluate a broad suite of state-of-the-art open-source and closed-source VLMs, comparing their performance while also providing optimal path solutions and a human baseline to assess the task's complexity and feasibility for humans. Results indicate that while some VLMs perform well on simple spatial tasks, they encounter difficulties with more complex configurations and problem properties. Notably, while VLMs generally perform better in 2D vision compared to 3D or text-based representations, they consistently fall short of human performance, illustrating the persistent challenge of visual alignment. This highlights critical gaps in current VLM capabilities, highlighting their limitations in achieving human-level cognition.
Authors: Napat Laosaengpha, Thanit Tativannarat, Attapol Rutherford, Ekapol Chuangsuwanich
Abstract: Understanding the textual components of resumes and job postings is critical for improving job-matching accuracy and optimizing job search systems in online recruitment platforms. However, existing works primarily focus on analyzing individual components within this information, requiring multiple specialized tools to analyze each aspect. Such disjointed methods could potentially hinder overall generalizability in recruitment-related text processing. Therefore, we propose a unified sentence encoder that utilized multi-task dual-encoder framework for jointly learning multiple component into the unified sentence encoder. The results show that our method outperforms other state-of-the-art models, despite its smaller model size. Moreover, we propose a novel metric, Language Bias Kullback-Leibler Divergence (LBKL), to evaluate language bias in the encoder, demonstrating significant bias reduction and superior cross-lingual performance.
Authors: Volker Emmrich
Abstract: Koch and Oesterreicher's model of "N\"ahe und Distanz" (N\"ahe = immediacy, conceptual orality; Distanz = distance, conceptual literacy) is constantly used in German linguistics. However, there is no statistical foundation for use in corpus linguistic analyzes, while it is increasingly moving into empirical corpus linguistics. Theoretically, it is stipulated, among other things, that written texts can be rated on a scale of conceptual orality and literacy by linguistic features. This article establishes such a scale based on PCA and combines it with automatic analysis. Two corpora of New High German serve as examples. When evaluating established features, a central finding is that features of conceptual orality and literacy must be distinguished in order to rank texts in a differentiated manner. The scale is also discussed with a view to its use in corpus compilation and as a guide for analyzes in larger corpora. With a theory-driven starting point and as a "tailored" dimension, the approach compared to Biber's Dimension 1 is particularly suitable for these supporting, controlling tasks.
Authors: Antonio Laverghetta Jr., Tuhin Chakrabarty, Tom Hope, Jimmy Pronchick, Krupa Bhawsar, Roger E. Beaty
Abstract: Creativity assessment in science and engineering is increasingly based on both human and AI judgment, but the cognitive processes and biases behind these evaluations remain poorly understood. We conducted two experiments examining how including example solutions with ratings impact creativity evaluation, using a finegrained annotation protocol where raters were tasked with explaining their originality scores and rating for the facets of remoteness (whether the response is "far" from everyday ideas), uncommonness (whether the response is rare), and cleverness. In Study 1, we analyzed creativity ratings from 72 experts with formal science or engineering training, comparing those who received example solutions with ratings (example) to those who did not (no example). Computational text analysis revealed that, compared to experts with examples, no-example experts used more comparative language (e.g., "better/worse") and emphasized solution uncommonness, suggesting they may have relied more on memory retrieval for comparisons. In Study 2, parallel analyses with state-of-the-art LLMs revealed that models prioritized uncommonness and remoteness of ideas when rating originality, suggesting an evaluative process rooted around the semantic similarity of ideas. In the example condition, while LLM accuracy in predicting the true originality scores improved, the correlations of remoteness, uncommonness, and cleverness with originality also increased substantially - to upwards of 0.99 - suggesting a homogenization in the LLMs evaluation of the individual facets. These findings highlight important implications for how humans and AI reason about creativity and suggest diverging preferences for what different populations prioritize when rating.
Authors: Thibaut Fabacher, Erik-Andr\'e Sauleau, Emmanuelle Arcay, Bineta Faye, Maxime Alter, Archia Chahard, Nathan Miraillet, Adrien Coulet, Aur\'elie N\'ev\'eol
Abstract: Objective: To evaluate the accuracy, computational cost and portability of a new Natural Language Processing (NLP) method for extracting medication information from clinical narratives. Materials and Methods: We propose an original transformer-based architecture for the extraction of entities and their relations pertaining to patients' medication regimen. First, we used this approach to train and evaluate a model on French clinical notes, using a newly annotated corpus from H\^opitaux Universitaires de Strasbourg. Second, the portability of the approach was assessed by conducting an evaluation on clinical documents in English from the 2018 n2c2 shared task. Information extraction accuracy and computational cost were assessed by comparison with an available method using transformers. Results: The proposed architecture achieves on the task of relation extraction itself performance that are competitive with the state-of-the-art on both French and English (F-measures 0.82 and 0.96 vs 0.81 and 0.95), but reduce the computational cost by 10. End-to-end (Named Entity recognition and Relation Extraction) F1 performance is 0.69 and 0.82 for French and English corpus. Discussion: While an existing system developed for English notes was deployed in a French hospital setting with reasonable effort, we found that an alternative architecture offered end-to-end drug information extraction with comparable extraction performance and lower computational impact for both French and English clinical text processing, respectively. Conclusion: The proposed architecture can be used to extract medication information from clinical text with high performance and low computational cost and consequently suits with usually limited hospital IT resources
Authors: DiJia Su, Hanlin Zhu, Yingchen Xu, Jiantao Jiao, Yuandong Tian, Qinqing Zheng
Abstract: Large Language Models (LLMs) excel at reasoning and planning when trained on chainof-thought (CoT) data, where the step-by-step thought process is explicitly outlined by text tokens. However, this results in lengthy inputs where many words support textual coherence rather than core reasoning information, and processing these inputs consumes substantial computation resources. In this work, we propose a hybrid representation of the reasoning process, where we partially abstract away the initial reasoning steps using latent discrete tokens generated by VQ-VAE, significantly reducing the length of reasoning traces. We explore the use of latent trace abstractions in two scenarios: 1) training the model from scratch for the Keys-Finding Maze problem, 2) fine-tuning LLMs on this hybrid data with an extended vocabulary including unseen latent tokens, for both logical and mathematical reasoning problems. To facilitate effective learning, we introduce a simple training procedure that randomly mixes latent and text tokens, which enables fast adaptation to new latent tokens. Our approach consistently outperforms the baselines methods in various benchmarks.
Authors: Aida Halitaj, Arkaitz Zubiaga
Abstract: Citation Worthiness Detection (CWD) consists in determining which sentences, within an article or collection, should be backed up with a citation to validate the information it provides. This study, introduces ALPET, a framework combining Active Learning (AL) and Pattern-Exploiting Training (PET), to enhance CWD for languages with limited data resources. Applied to Catalan, Basque, and Albanian Wikipedia datasets, ALPET outperforms the existing CCW baseline while reducing the amount of labeled data in some cases above 80\%. ALPET's performance plateaus after 300 labeled samples, showing it suitability for low-resource scenarios where large, labeled datasets are not common. While specific active learning query strategies, like those employing K-Means clustering, can offer advantages, their effectiveness is not universal and often yields marginal gains over random sampling, particularly with smaller datasets. This suggests that random sampling, despite its simplicity, remains a strong baseline for CWD in constraint resource environments. Overall, ALPET's ability to achieve high performance with fewer labeled samples makes it a promising tool for enhancing the verifiability of online content in low-resource language settings.
Authors: Amin Dada, Osman Alperen Koras, Marie Bauer, Amanda Butler, Kaleb E. Smith, Jens Kleesiek, Julian Friedrich
Abstract: While increasing patients' access to medical documents improves medical care, this benefit is limited by varying health literacy levels and complex medical terminology. Large language models (LLMs) offer solutions by simplifying medical information. However, evaluating LLMs for safe and patient-friendly text generation is difficult due to the lack of standardized evaluation resources. To fill this gap, we developed MeDiSumQA. MeDiSumQA is a dataset created from MIMIC-IV discharge summaries through an automated pipeline combining LLM-based question-answer generation with manual quality checks. We use this dataset to evaluate various LLMs on patient-oriented question-answering. Our findings reveal that general-purpose LLMs frequently surpass biomedical-adapted models, while automated metrics correlate with human judgment. By releasing MeDiSumQA on PhysioNet, we aim to advance the development of LLMs to enhance patient understanding and ultimately improve care outcomes.
Authors: Momin Abbas, Muneeza Azmat, Raya Horesh, Mikhail Yurochkin
Abstract: Distinguishing in- and out-of-distribution (OOD) inputs is crucial for reliable deployment of classification systems. However, OOD data is typically unavailable or difficult to collect, posing a significant challenge for accurate OOD detection. In this work, we present a method that harnesses the generative capabilities of Large Language Models (LLMs) to create high-quality synthetic OOD proxies, eliminating the dependency on any external OOD data source. We study the efficacy of our method on classical text classification tasks such as toxicity detection and sentiment classification as well as classification tasks arising in LLM development and deployment, such as training a reward model for RLHF and detecting misaligned generations. Extensive experiments on nine InD-OOD dataset pairs and various model sizes show that our approach dramatically lowers false positive rates (achieving a perfect zero in some cases) while maintaining high accuracy on in-distribution tasks, outperforming baseline methods by a significant margin.
Authors: Qiguang Chen, Libo Qin, Jinhao Liu, Dengyun Peng, Jiaqi Wang, Mengkang Hu, Zhi Chen, Wanxiang Che, Ting Liu
Abstract: Recent advancements in large language models (LLMs) have led to significant successes across various applications, where the most noticeable is to a series of emerging capabilities, particularly in the areas of In-Context Learning (ICL) and Chain-of-Thought (CoT). To better understand and control model performance, many studies have begun investigating the underlying causes of these phenomena and their impact on task outcomes. However, existing explanatory frameworks predominantly focus on isolating and explaining ICL and CoT independently, leading to an incomplete understanding of their combined influence on model performance. To address this gap, we propose the Electronic Circuit Model (ECM), which provides a foundation for developing scalable, learnable policies and improving the management of AI-generated content. Specifically, ECM conceptualizes model behavior as an electronic circuit: ICL is represented as semantic magnetic field to providing an additional voltage following Faraday's Law, while CoT is modeled as series resistors to constrain the model output performance following Ohm's Law. Experimental results demonstrate that the ECM effectively predicts and explains LLM performance across a variety of prompting strategies. Furthermore, we apply ECM to advanced reasoning strategy optimization on a series of tasks, such as the International Olympiad in Informatics (IOI) and the International Mathematical Olympiad (IMO), achieving competitive performance that surpasses nearly 80% of top human competitors.
Authors: Menglin Xia, Victor Ruehle, Saravan Rajmohan, Reza Shokri
Abstract: How effectively can LLM-based AI assistants utilize their memory (context) to perform various tasks? Traditional data benchmarks, which are often manually crafted, suffer from several limitations: they are static, susceptible to overfitting, difficult to interpret, and lack actionable insights--failing to pinpoint the specific capabilities a model lacks when it does not pass a test. In this paper, we present a framework for automatically generating a comprehensive set of tests to evaluate models' abilities to use their memory effectively. Our framework extends the range of capability tests beyond the commonly explored (passkey, key-value, needle in the haystack) search, a dominant focus in the literature. Specifically, we evaluate models on atomic tasks such as searching, recalling, editing, matching, comparing information in context memory, and performing basic operations when inputs are structured into distinct blocks, simulating real-world data. Additionally, we design composite tests to investigate the models' ability to maintain state while operating on memory. Our benchmark enables an interpretable, detailed assessment of memory capabilities of LLMs.
Authors: Edward Yeo, Yuxuan Tong, Morry Niu, Graham Neubig, Xiang Yue
Abstract: Scaling inference compute enhances reasoning in large language models (LLMs), with long chains-of-thought (CoTs) enabling strategies like backtracking and error correction. Reinforcement learning (RL) has emerged as a crucial method for developing these capabilities, yet the conditions under which long CoTs emerge remain unclear, and RL training requires careful design choices. In this study, we systematically investigate the mechanics of long CoT reasoning, identifying the key factors that enable models to generate long CoT trajectories. Through extensive supervised fine-tuning (SFT) and RL experiments, we present four main findings: (1) While SFT is not strictly necessary, it simplifies training and improves efficiency; (2) Reasoning capabilities tend to emerge with increased training compute, but their development is not guaranteed, making reward shaping crucial for stabilizing CoT length growth; (3) Scaling verifiable reward signals is critical for RL. We find that leveraging noisy, web-extracted solutions with filtering mechanisms shows strong potential, particularly for out-of-distribution (OOD) tasks such as STEM reasoning; and (4) Core abilities like error correction are inherently present in base models, but incentivizing these skills effectively for complex tasks via RL demands significant compute, and measuring their emergence requires a nuanced approach. These insights provide practical guidance for optimizing training strategies to enhance long CoT reasoning in LLMs. Our code is available at: https://github.com/eddycmu/demystify-long-cot.
Authors: Shiyi Tan, Constantin Or\u{a}san, Sabine Braun
Abstract: This paper reports on the results from a pilot study investigating the impact of automatic speech recognition (ASR) technology on interpreting quality in remote healthcare interpreting settings. Employing a within-subjects experiment design with four randomised conditions, this study utilises scripted medical consultations to simulate dialogue interpreting tasks. It involves four trainee interpreters with a language combination of Chinese and English. It also gathers participants' experience and perceptions of ASR support through cued retrospective reports and semi-structured interviews. Preliminary data suggest that the availability of ASR, specifically the access to full ASR transcripts and to ChatGPT-generated summaries based on ASR, effectively improved interpreting quality. Varying types of ASR output had different impacts on the distribution of interpreting error types. Participants reported similar interactive experiences with the technology, expressing their preference for full ASR transcripts. This pilot study shows encouraging results of applying ASR to dialogue-based healthcare interpreting and offers insights into the optimal ways to present ASR output to enhance interpreter experience and performance. However, it should be emphasised that the main purpose of this study was to validate the methodology and that further research with a larger sample size is necessary to confirm these findings.
Authors: Tom Labiausse, Laurent Mazar\'e, Edouard Grave, Patrick P\'erez, Alexandre D\'efossez, Neil Zeghidour
Abstract: We introduce Hibiki, a decoder-only model for simultaneous speech translation. Hibiki leverages a multistream language model to synchronously process source and target speech, and jointly produces text and audio tokens to perform speech-to-text and speech-to-speech translation. We furthermore address the fundamental challenge of simultaneous interpretation, which unlike its consecutive counterpart, where one waits for the end of the source utterance to start translating, adapts its flow to accumulate just enough context to produce a correct translation in real-time, chunk by chunk. To do so, we introduce a weakly-supervised method that leverages the perplexity of an off-the-shelf text translation system to identify optimal delays on a per-word basis and create aligned synthetic data. After supervised training, Hibiki performs adaptive, simultaneous speech translation with vanilla temperature sampling. On a French-English simultaneous speech translation task, Hibiki demonstrates state-of-the-art performance in translation quality, speaker fidelity and naturalness. Moreover, the simplicity of its inference process makes it compatible with batched translation and even real-time on-device deployment. We provide examples as well as models and inference code.
Authors: Yixin Ye, Zhen Huang, Yang Xiao, Ethan Chern, Shijie Xia, Pengfei Liu
Abstract: We present a fundamental discovery that challenges our understanding of how complex reasoning emerges in large language models. While conventional wisdom suggests that sophisticated reasoning tasks demand extensive training data (>100,000 examples), we demonstrate that complex mathematical reasoning abilities can be effectively elicited with surprisingly few examples. Through comprehensive experiments, our proposed model LIMO demonstrates unprecedented performance in mathematical reasoning. With merely 817 curated training samples, LIMO achieves 57.1% accuracy on AIME and 94.8% on MATH, improving from previous SFT-based models' 6.5% and 59.2% respectively, while only using 1% of the training data required by previous approaches. LIMO demonstrates exceptional out-of-distribution generalization, achieving 40.5% absolute improvement across 10 diverse benchmarks, outperforming models trained on 100x more data, challenging the notion that SFT leads to memorization rather than generalization. Based on these results, we propose the Less-Is-More Reasoning Hypothesis (LIMO Hypothesis): In foundation models where domain knowledge has been comprehensively encoded during pre-training, sophisticated reasoning capabilities can emerge through minimal but precisely orchestrated demonstrations of cognitive processes. This hypothesis posits that the elicitation threshold for complex reasoning is determined by two key factors: (1) the completeness of the model's encoded knowledge foundation during pre-training, and (2) the effectiveness of post-training examples as "cognitive templates" that show the model how to utilize its knowledge base to solve complex reasoning tasks. To facilitate reproducibility and future research in data-efficient reasoning, we release LIMO as a comprehensive open-source suite at https://github.com/GAIR-NLP/LIMO.
Authors: Hongli Zhan, Muneeza Azmat, Raya Horesh, Junyi Jessy Li, Mikhail Yurochkin
Abstract: Aligning Large Language Models to integrate and reflect human values, especially for tasks that demand intricate human oversight, is arduous since it is resource-intensive and time-consuming to depend on human expertise for context-specific guidance. Prior work has utilized predefined sets of rules or principles to steer the behavior of models (Bai et al., 2022; Sun et al., 2023). However, these principles tend to be generic, making it challenging to adapt them to each individual input query or context. In this work, we present Situated-PRInciples (SPRI), a framework requiring minimal or no human effort that is designed to automatically generate guiding principles in real-time for each input query and utilize them to align each response. We evaluate SPRI on three tasks, and show that 1) SPRI can derive principles in a complex domain-specific task that leads to on-par performance as expert-crafted ones; 2) SPRI-generated principles lead to instance-specific rubrics that outperform prior LLM-as-a-judge frameworks; 3) using SPRI to generate synthetic SFT data leads to substantial improvement on truthfulness. We release our code and model generations at https://github.com/honglizhan/SPRI-public.
Authors: Nikta Gohari Sadr, Sangmitra Madhusudan, Ali Emami
Abstract: Zero-shot prompting techniques have significantly improved the performance of Large Language Models (LLMs). However, we lack a clear understanding of why zero-shot prompts are so effective. For example, in the prompt "Let's think step-by-step," is "think" or "step-by-step" more crucial to its success? Existing interpretability methods, such as gradient-based and attention-based approaches, are computationally intensive and restricted to open-source models. We introduce the ZIP score (Zero-shot Importance of Perturbation score), a versatile metric applicable to both open and closed-source models, based on systematic input word perturbations. Our experiments across four recent LLMs, seven widely-used prompts, and several tasks, reveal interesting patterns in word importance. For instance, while both 'step-by-step' and 'think' show high ZIP scores, which one is more influential depends on the model and task. We validate our method using controlled experiments and compare our results with human judgments, finding that proprietary models align more closely with human intuition regarding word significance. These findings enhance our understanding of LLM behavior and contribute to developing more effective zero-shot prompts and improved model analysis.
Authors: Ming Liu, Hao Chen, Jindong Wang, Liwen Wang, Bhiksha Raj Ramakrishnan, Wensheng Zhang
Abstract: Unified multimodal large language models (U-MLLMs) have demonstrated impressive performance in visual understanding and generation in an end-to-end pipeline. Compared with generation-only models (e.g., Stable Diffusion), U-MLLMs may raise new questions about bias in their outputs, which can be affected by their unified capabilities. This gap is particularly concerning given the under-explored risk of propagating harmful stereotypes. In this paper, we benchmark the latest U-MLLMs and find that most exhibit significant demographic biases, such as gender and race bias. To better understand and mitigate this issue, we propose a locate-then-fix strategy, where we audit and show how the individual model component is affected by bias. Our analysis shows that bias originates primarily from the language model. More interestingly, we observe a "partial alignment" phenomenon in U-MLLMs, where understanding bias appears minimal, but generation bias remains substantial. Thus, we propose a novel balanced preference model to balance the demographic distribution with synthetic data. Experiments demonstrate that our approach reduces demographic bias while preserving semantic fidelity. We hope our findings underscore the need for more holistic interpretation and debiasing strategies of U-MLLMs in the future.
Authors: Chunyu Sun, Bingyu Liu, Zhichao Cui, Anbin Qi, Tian-hao Zhang, Dinghao Zhou, Lewei Lu
Abstract: Embedding-based retrieval models have made significant strides in retrieval-augmented generation (RAG) techniques for text and multimodal large language models (LLMs) applications. However, when it comes to speech larage language models (SLLMs), these methods are limited to a two-stage process, where automatic speech recognition (ASR) is combined with text-based retrieval. This sequential architecture suffers from high latency and error propagation. To address these limitations, we propose a unified embedding framework that eliminates the need for intermediate text representations. Specifically, the framework includes separate speech and text encoders, followed by a shared scaling layer that maps both modalities into a common embedding space. Our model reduces pipeline latency by 50\% while achieving higher retrieval accuracy compared to traditional two-stage methods. We also provide a theoretical analysis of the challenges inherent in end-to-end speech retrieval and introduce architectural principles for effective speech-to-document matching. Extensive experiments demonstrate the robustness of our approach across diverse acoustic conditions and speaker variations, paving the way for a new paradigm in multimodal SLLMs retrieval systems.
Authors: Zechun Liu, Changsheng Zhao, Hanxian Huang, Sijia Chen, Jing Zhang, Jiawei Zhao, Scott Roy, Lisa Jin, Yunyang Xiong, Yangyang Shi, Lin Xiao, Yuandong Tian, Bilge Soran, Raghuraman Krishnamoorthi, Tijmen Blankevoort, Vikas Chandra
Abstract: The optimal bit-width for achieving the best trade-off between quantized model size and accuracy has been a subject of ongoing debate. While some advocate for 4-bit quantization, others propose that 1.58-bit offers superior results. However, the lack of a cohesive framework for different bits has left such conclusions relatively tenuous. We present ParetoQ, the first unified framework that facilitates rigorous comparisons across 1-bit, 1.58-bit, 2-bit, 3-bit, and 4-bit quantization settings. Our findings reveal a notable learning transition between 2 and 3 bits: For 3-bits and above, the fine-tuned models stay close to their original pre-trained distributions, whereas for learning 2-bit networks or below, the representations change drastically. By optimizing training schemes and refining quantization functions, ParetoQ surpasses all previous methods tailored to specific bit widths. Remarkably, our ParetoQ ternary 600M-parameter model even outperforms the previous SoTA ternary 3B-parameter model in accuracy, using only one-fifth of the parameters. Extensive experimentation shows that ternary, 2-bit, and 3-bit quantization maintains comparable performance in the size-accuracy trade-off and generally exceeds 4-bit and binary quantization. Considering hardware constraints, 2-bit quantization offers promising potential for memory reduction and speedup.
Authors: Daniil Tiapkin, Daniele Calandriello, Johan Ferret, Sarah Perrin, Nino Vieillard, Alexandre Ram\'e, Mathieu Blondel
Abstract: Post-training of language models (LMs) increasingly relies on the following two stages: (i) knowledge distillation, where the LM is trained to imitate a larger teacher LM, and (ii) reinforcement learning from human feedback (RLHF), where the LM is aligned by optimizing a reward model. In the second RLHF stage, a well-known challenge is reward hacking, where the LM over-optimizes the reward model. Such phenomenon is in line with Goodhart's law and can lead to degraded performance on the true objective. In this paper, we investigate whether a similar phenomenon, that we call teacher hacking, can occur during knowledge distillation. This could arise because the teacher LM is itself an imperfect approximation of the true distribution. To study this, we propose a controlled experimental setup involving: (i) an oracle LM representing the ground-truth distribution, (ii) a teacher LM distilled from the oracle, and (iii) a student LM distilled from the teacher. Our experiments reveal the following insights. When using a fixed offline dataset for distillation, teacher hacking occurs; moreover, we can detect it by observing when the optimization process deviates from polynomial convergence laws. In contrast, employing online data generation techniques effectively mitigates teacher hacking. More precisely, we identify data diversity as the key factor in preventing hacking. Overall, our findings provide a deeper understanding of the benefits and limitations of distillation for building robust and efficient LMs.
Authors: Peidong Wang, Naoyuki Kanda, Jian Xue, Jinyu Li, Xiaofei Wang, Aswin Shanmugam Subramanian, Junkun Chen, Sunit Sivasankaran, Xiong Xiao, Yong Zhao
Abstract: Streaming multi-talker speech translation is a task that involves not only generating accurate and fluent translations with low latency but also recognizing when a speaker change occurs and what the speaker's gender is. Speaker change information can be used to create audio prompts for a zero-shot text-to-speech system, and gender can help to select speaker profiles in a conventional text-to-speech model. We propose to tackle streaming speaker change detection and gender classification by incorporating speaker embeddings into a transducer-based streaming end-to-end speech translation model. Our experiments demonstrate that the proposed methods can achieve high accuracy for both speaker change detection and gender classification.
Authors: Shawn Im, Yixuan Li
Abstract: Steering methods provide a practical approach to controlling large language models by applying steering vectors to intermediate activations, guiding outputs toward desired behaviors while avoiding retraining. Despite their growing importance, the field lacks a unified understanding and consistent evaluation across tasks and datasets, hindering progress. This paper introduces a unified framework for analyzing and evaluating steering methods, formalizing their core principles and offering theoretical insights into their effectiveness. Through comprehensive empirical evaluations on multiple-choice and open-ended text generation tasks, we validate these insights, identifying key factors that influence performance and demonstrating the superiority of certain methods. Our work bridges theoretical and practical perspectives, offering actionable guidance for advancing the design, optimization, and deployment of steering methods in LLMs.
Authors: Jeonghoon Kim, Byeongchan Lee, Cheonbok Park, Yeontaek Oh, Beomjun Kim, Taehwan Yoo, Seongjin Shin, Dongyoon Han, Jinwoo Shin, Kang Min Yoo
Abstract: Designing Transformer architectures with the optimal layer normalization (LN) strategy that ensures large-scale training stability and expedite convergence has remained elusive, even in this era of large language models (LLMs). To this end, we present a comprehensive analytical foundation for understanding how different LN strategies influence training dynamics in large-scale Transformer training. Until recently, Pre-LN and Post-LN have long dominated standard practices despite their limitations in large-scale training. However, several open-source large-scale models have recently begun silently adopting a third strategy without much explanation. This strategy places layer normalization (LN) peripherally around sublayers, a design we term Peri-LN. While Peri-LN has demonstrated promising empirical performance, its precise mechanisms and benefits remain almost unexplored. Our in-depth analysis shows that Peri-LN strikes an ideal balance in variance growth -- unlike Pre-LN and Post-LN, which are prone to vanishing gradients and ``massive activations.'' To validate our theoretical insight, we conduct large-scale experiments on Transformers up to 3.2B parameters, showing that Peri-LN consistently achieves more balanced variance growth, steadier gradient flow, and convergence stability. Our results suggest that Peri-LN warrants broader consideration for large-scale Transformer architectures, providing renewed insights into the optimal placement and application of LN.
Authors: Chaofan Lin, Jiaming Tang, Shuo Yang, Hanshuo Wang, Tian Tang, Boyu Tian, Ion Stoica, Song Han, Mingyu Gao
Abstract: Leveraging attention sparsity to accelerate long-context large language models (LLMs) has been a hot research topic. However, current algorithms such as sparse attention or key-value (KV) cache compression tend to use a fixed budget, which presents a significant challenge during deployment because it fails to account for the dynamic nature of real-world scenarios, where the optimal balance between accuracy and efficiency can vary greatly. In this paper, we find that borrowing top-$p$ sampling (nucleus sampling) to sparse attention can surprisingly achieve adaptive budgeting. Based on this, we propose Twilight, a framework to bring adaptive sparsity to any existing sparse attention algorithm without sacrificing their accuracy. Empirical results show that Twilight can adaptively prune at most 98% of redundant tokens, leading to $15.4\times$ acceleration in self-attention operations and $3.9\times$ acceleration in end-to-end per token latency in long context LLM decoding.
Authors: Ram\'on Calvo Gonz\'alez, Daniele Paliotta, Matteo Pagliardini, Martin Jaggi, Fran\c{c}ois Fleuret
Abstract: Large Language Models demonstrate remarkable capabilities at the cost of high compute requirements. While recent research has shown that intermediate layers can be removed or have their order shuffled without impacting performance significantly, these findings have not been employed to reduce the computational cost of inference. We investigate several potential ways to reduce the depth of pre-trained LLMs without significantly affecting performance. Leveraging our insights, we present a novel approach that exploits this decoupling between layers by grouping some of them into pairs that can be evaluated in parallel. This modification of the computational graph -- through better parallelism -- results in an average improvement of around 1.20x on the number of tokens generated per second, without re-training nor fine-tuning, while retaining 95%-99% of the original accuracy. Empirical evaluation demonstrates that this approach significantly improves serving efficiency while maintaining model performance, offering a practical improvement for large-scale LLM deployment.
Authors: Linghe Wang, Minhwa Lee, Ross Volkov, Luan Tuyen Chau, Dongyeop Kang
Abstract: Writing is a cognitively demanding task involving continuous decision-making, heavy use of working memory, and frequent switching between multiple activities. Scholarly writing is particularly complex as it requires authors to coordinate many pieces of multiform knowledge. To fully understand writers' cognitive thought process, one should fully decode the end-to-end writing data (from individual ideas to final manuscript) and understand their complex cognitive mechanisms in scholarly writing. We introduce ScholaWrite dataset, the first-of-its-kind keystroke logs of an end-to-end scholarly writing process for complete manuscripts, with thorough annotations of cognitive writing intentions behind each keystroke. Our dataset includes LaTeX-based keystroke data from five preprints with nearly 62K total text changes and annotations across 4 months of paper writing. ScholaWrite shows promising usability and applications (e.g., iterative self-writing) for the future development of AI writing assistants for academic research, which necessitate complex methods beyond LLM prompting. Our experiments clearly demonstrated the importance of collection of end-to-end writing data, rather than the final manuscript, for the development of future writing assistants to support the cognitive thinking process of scientists. Our de-identified dataset, demo, and code repository are available on our project page.
Authors: Dinithi Jayasuriya (Intel Labs, Oregon), Sina Tayebati (Intel Labs, Oregon), Davide Ettori (Intel Labs, Oregon), Ranganath Krishnan (Intel Labs, Oregon), Amit Ranjan Trivedi (Intel Labs, Oregon)
Abstract: We propose SPARC, a lightweight continual learning framework for large language models (LLMs) that enables efficient task adaptation through prompt tuning in a lower-dimensional space. By leveraging principal component analysis (PCA), we identify a compact subspace of the training data. Optimizing prompts in this lower-dimensional space enhances training efficiency, as it focuses updates on the most relevant features while reducing computational overhead. Furthermore, since the model's internal structure remains unaltered, the extensive knowledge gained from pretraining is fully preserved, ensuring that previously learned information is not compromised during adaptation. Our method achieves high knowledge retention in both task-incremental and domain-incremental continual learning setups while fine-tuning only 0.04% of the model's parameters. Additionally, by integrating LoRA, we enhance adaptability to computational constraints, allowing for a tradeoff between accuracy and training cost. Experiments on the SuperGLUE benchmark demonstrate that our PCA-based prompt tuning combined with LoRA maintains full knowledge retention while improving accuracy, utilizing only 1% of the model's parameters. These results establish our approach as a scalable and resource-efficient solution for continual learning in LLMs.
Authors: Seng Pei Liew, Takuya Kato, Sho Takase
Abstract: Pretraining large language models (LLMs) is resource-intensive, often requiring months of training time even with high-end GPU clusters. There are two approaches of mitigating such computational demands: reusing smaller models to train larger ones (upcycling), and training computationally efficient models like mixture-of-experts (MoE). In this paper, we study the upcycling of LLMs to MoE models, of which the scaling behavior remains underexplored. Through extensive experiments, we identify empirical scaling laws that describe how performance depends on dataset size and model configuration. Particularly, we show that, while scaling these factors improves performance, there is a novel interaction term between the dense and upcycled training dataset that limits the efficiency of upcycling at large computational budgets. Based on these findings, we provide guidance to scale upcycling, and establish conditions under which upcycling outperforms from-scratch trainings within budget constraints.
Authors: Daniil Laptev, Nikita Balagansky, Yaroslav Aksenov, Daniil Gavrilov
Abstract: We introduce a new approach to systematically map features discovered by sparse autoencoder across consecutive layers of large language models, extending earlier work that examined inter-layer feature links. By using a data-free cosine similarity technique, we trace how specific features persist, transform, or first appear at each stage. This method yields granular flow graphs of feature evolution, enabling fine-grained interpretability and mechanistic insights into model computations. Crucially, we demonstrate how these cross-layer feature maps facilitate direct steering of model behavior by amplifying or suppressing chosen features, achieving targeted thematic control in text generation. Together, our findings highlight the utility of a causal, cross-layer interpretability framework that not only clarifies how features develop through forward passes but also provides new means for transparent manipulation of large language models.
Authors: Ben Liu, Jihai Zhang, Fangquan Lin, Cheng Yang, Min Peng, Wotao Yin
Abstract: Recent advancements have highlighted that Large Language Models (LLMs) are prone to hallucinations when solving complex reasoning problems, leading to erroneous results. To tackle this issue, researchers incorporate Knowledge Graphs (KGs) to improve the reasoning ability of LLMs. However, existing methods face two limitations: 1) they typically assume that all answers to the questions are contained in KGs, neglecting the incompleteness issue of KGs, and 2) they treat the KG as a static repository and overlook the implicit logical reasoning structures inherent in KGs. In this paper, we introduce SymAgent, an innovative neural-symbolic agent framework that achieves collaborative augmentation between KGs and LLMs. We conceptualize KGs as dynamic environments and transform complex reasoning tasks into a multi-step interactive process, enabling KGs to participate deeply in the reasoning process. SymAgent consists of two modules: Agent-Planner and Agent-Executor. The Agent-Planner leverages LLM's inductive reasoning capability to extract symbolic rules from KGs, guiding efficient question decomposition. The Agent-Executor autonomously invokes predefined action tools to integrate information from KGs and external documents, addressing the issues of KG incompleteness. Furthermore, we design a self-learning framework comprising online exploration and offline iterative policy updating phases, enabling the agent to automatically synthesize reasoning trajectories and improve performance. Experimental results demonstrate that SymAgent with weak LLM backbones (i.e., 7B series) yields better or comparable performance compared to various strong baselines. Further analysis reveals that our agent can identify missing triples, facilitating automatic KG updates.
Authors: Qitao Tan, Jun Liu, Zheng Zhan, Caiwei Ding, Yanzhi Wang, Jin Lu, Geng Yuan
Abstract: Large language models (LLMs) excel across various tasks, but standard first-order (FO) fine-tuning demands considerable memory, significantly limiting real-world deployment. Recently, zeroth-order (ZO) optimization stood out as a promising memory-efficient training paradigm, avoiding backward passes and relying solely on forward passes for gradient estimation, making it attractive for resource-constrained scenarios. However, ZO method lags far behind FO method in both convergence speed and accuracy. To bridge the gap, we introduce a novel layer-wise divergence analysis that uncovers the distinct update pattern of FO and ZO optimization. Aiming to resemble the learning capacity of FO method from the findings, we propose \textbf{Di}vergence-driven \textbf{Z}eroth-\textbf{O}rder (\textbf{DiZO}) optimization. DiZO conducts divergence-driven layer adaptation by incorporating projections to ZO updates, generating diverse-magnitude updates precisely scaled to layer-wise individual optimization needs. Our results demonstrate that DiZO significantly reduces the needed iterations for convergence without sacrificing throughput, cutting training GPU hours by up to 48\% on various datasets. Moreover, DiZO consistently outperforms the representative ZO baselines in fine-tuning RoBERTa-large, OPT-series, and Llama-series on downstream tasks and, in some cases, even surpasses memory-intensive FO fine-tuning.
Authors: Boyao Wang, Rui Pan, Shizhe Diao, Xingyuan Pan, Jipeng Zhang, Renjie Pi, Tong Zhang
Abstract: Small language models (SLMs) have attracted considerable attention from both academia and industry due to their broad range of applications in edge devices. To obtain SLMs with strong performance, conventional approaches either pre-train the models from scratch, which incurs substantial computational costs, or compress/prune existing large language models (LLMs), which results in performance drops and falls short in comparison to pre-training. In this paper, we investigate the family of acceleration methods that involve both structured pruning and model training. We found 1) layer-wise adaptive pruning (Adapt-Pruner) is extremely effective in LLMs and yields significant improvements over existing pruning techniques, 2) adaptive pruning equipped with further training leads to models comparable to those pre-training from scratch, 3) incremental pruning brings non-trivial performance gain by interleaving pruning with training and only removing a small portion of neurons ($\sim$5%) at a time. Experimental results on LLaMA-3.1-8B demonstrate that Adapt-Pruner outperforms conventional pruning methods, such as LLM-Pruner, FLAP, and SliceGPT, by an average of 1%-7% in accuracy on commonsense benchmarks. Additionally, Adapt-Pruner restores the performance of MobileLLM-125M to 600M on the MMLU benchmark with 200$\times$ fewer tokens via pruning from its larger counterparts, and discovers a new 1B model that surpasses LLaMA-3.2-1B in multiple benchmarks.
Authors: Joshua Vendrow, Edward Vendrow, Sara Beery, Aleksander Madry
Abstract: When deploying large language models (LLMs), it is important to ensure that these models are not only capable, but also reliable. Many benchmarks have been created to track LLMs' growing capabilities, however there has been no similar focus on measuring their reliability. To understand the potential ramifications of this gap, we investigate how well current benchmarks quantify model reliability. We find that pervasive label errors can compromise these evaluations, obscuring lingering model failures and hiding unreliable behavior. Motivated by this gap in the evaluation of reliability, we then propose the concept of so-called platinum benchmarks, i.e., benchmarks carefully curated to minimize label errors and ambiguity. As a first attempt at constructing such benchmarks, we revise examples from fifteen existing popular benchmarks. We evaluate a wide range of models on these platinum benchmarks and find that, indeed, frontier LLMs still exhibit failures on simple tasks such as elementary-level math word problems. Analyzing these failures further reveals previously unidentified patterns of problems on which frontier models consistently struggle. We provide code at https://github.com/MadryLab/platinum-benchmarks
Authors: Linkai Peng, Baorian Nuchged, Yingming Gao
Abstract: People have long hoped for a conversational system that can assist in real-life situations, and recent progress on large language models (LLMs) is bringing this idea closer to reality. While LLMs are often impressive in performance, their efficacy in real-world scenarios that demand expert knowledge remains unclear. LLMs are believed to hold the most potential and value in education, especially in the development of Artificial intelligence (AI) based virtual teachers capable of facilitating language learning. Our focus is centered on evaluating the efficacy of LLMs in the realm of education, specifically in the areas of spoken language learning which encompass phonetics, phonology, and second language acquisition. We introduce a new multiple-choice question dataset to evaluate the effectiveness of LLMs in the aforementioned scenarios, including understanding and application of spoken language knowledge. In addition, we investigate the influence of various prompting techniques such as zero- and few-shot method (prepending the question with question-answer exemplars), chain-of-thought (CoT, think step-by-step), in-domain exampler and external tools (Google, Wikipedia). We conducted large-scale evaluation on popular LLMs (20 distinct models) using these methods. We achieved significant performance improvements compared to the zero-shot baseline in the practical questions reasoning (GPT-3.5, 49.1% -> 63.1%; LLaMA2-70B-Chat, 42.2% -> 48.6%). We found that models of different sizes have good understanding of concepts in phonetics, phonology, and second language acquisition, but show limitations in reasoning for real-world problems. Additionally, we also explore preliminary findings on conversational communication.
Authors: Aounon Kumar, Chirag Agarwal, Suraj Srinivas, Aaron Jiaxun Li, Soheil Feizi, Himabindu Lakkaraju
Abstract: Large language models (LLMs) are vulnerable to adversarial attacks that add malicious tokens to an input prompt to bypass the safety guardrails of an LLM and cause it to produce harmful content. In this work, we introduce erase-and-check, the first framework for defending against adversarial prompts with certifiable safety guarantees. Given a prompt, our procedure erases tokens individually and inspects the resulting subsequences using a safety filter. Our safety certificate guarantees that harmful prompts are not mislabeled as safe due to an adversarial attack up to a certain size. We implement the safety filter in two ways, using Llama 2 and DistilBERT, and compare the performance of erase-and-check for the two cases. We defend against three attack modes: i) adversarial suffix, where an adversarial sequence is appended at the end of a harmful prompt; ii) adversarial insertion, where the adversarial sequence is inserted anywhere in the middle of the prompt; and iii) adversarial infusion, where adversarial tokens are inserted at arbitrary positions in the prompt, not necessarily as a contiguous block. Our experimental results demonstrate that this procedure can obtain strong certified safety guarantees on harmful prompts while maintaining good empirical performance on safe prompts. Additionally, we propose three efficient empirical defenses: i) RandEC, a randomized subsampling version of erase-and-check; ii) GreedyEC, which greedily erases tokens that maximize the softmax score of the harmful class; and iii) GradEC, which uses gradient information to optimize tokens to erase. We demonstrate their effectiveness against adversarial prompts generated by the Greedy Coordinate Gradient (GCG) attack algorithm. The code for our experiments is available at https://github.com/aounon/certified-llm-safety.
Authors: Jinu Lee, Wonseok Hwang
Abstract: To improve the performance and explainability of LLM-based natural language reasoning, structured reasoning can be applied to generate explicitly structured proofs. Among different methods for structured reasoning, we specifically focus on backward chaining, where the proof goal is recursively decomposed to subgoals by searching and applying rules. We argue that current LLM-based backward chaining systems (e.g. Least-to-most prompting and LAMBADA) are incomplete, as they omit crucial algorithmic components identified from the classic backward chaining algorithm in computational logic (SLD Resolution). To this end, we propose a novel backward chaining system, SymBa (Symbolic Backward Chaining), which integrates a symbolic solver and an LLM. In SymBa, the solver controls the proof process, and the LLM is only called when the solver requires new information to complete the proof. Empowered by completeness, SymBa achieves a significant improvement in seven deductive, relational, and arithmetic reasoning benchmarks compared to the baselines.
Authors: Yunyi Zhang, Ruozhen Yang, Xueqiang Xu, Rui Li, Jinfeng Xiao, Jiaming Shen, Jiawei Han
Abstract: Hierarchical text classification aims to categorize each document into a set of classes in a label taxonomy, which is a fundamental web text mining task with broad applications such as web content analysis and semantic indexing. Most earlier works focus on fully or semi-supervised methods that require a large amount of human annotated data which is costly and time-consuming to acquire. To alleviate human efforts, in this paper, we work on hierarchical text classification with a minimal amount of supervision: using the sole class name of each node as the only supervision. Recently, large language models (LLM) have shown competitive performance on various tasks through zero-shot prompting, but this method performs poorly in the hierarchical setting because it is ineffective to include the large and structured label space in a prompt. On the other hand, previous weakly-supervised hierarchical text classification methods only utilize the raw taxonomy skeleton and ignore the rich information hidden in the text corpus that can serve as additional class-indicative features. To tackle the above challenges, we propose TELEClass, Taxonomy Enrichment and LLM-Enhanced weakly-supervised hierarchical text Classification, which combines the general knowledge of LLMs and task-specific features mined from an unlabeled corpus. TELEClass automatically enriches the raw taxonomy with class-indicative features for better label space understanding and utilizes novel LLM-based data annotation and generation methods specifically tailored for the hierarchical setting. Experiments show that TELEClass can significantly outperform previous baselines while achieving comparable performance to zero-shot prompting of LLMs with drastically less inference cost.
Authors: Yiyang Luo, Ke Lin, Chao Gu, Jiahui Hou, Lijie Wen, Ping Luo
Abstract: The proliferation of large language models (LLMs) in generating content raises concerns about text copyright. Watermarking methods, particularly logit-based approaches, embed imperceptible identifiers into text to address these challenges. However, the widespread usage of watermarking across diverse LLMs has led to an inevitable issue known as watermark collision during common tasks, such as paraphrasing or translation. In this paper, we introduce watermark collision as a novel and general philosophy for watermark attacks, aimed at enhancing attack performance on top of any other attacking methods. We also provide a comprehensive demonstration that watermark collision poses a threat to all logit-based watermark algorithms, impacting not only specific attack scenarios but also downstream applications.
Authors: Daryna Dementieva, Valeriia Khylenko, Georg Groh
Abstract: Despite the extensive amount of labeled datasets in the NLP text classification field, the persistent imbalance in data availability across various languages remains evident. To support further fair development of NLP models, exploring the possibilities of effective knowledge transfer to new languages is crucial. Ukrainian, in particular, stands as a language that still can benefit from the continued refinement of cross-lingual methodologies. Due to our knowledge, there is a tremendous lack of Ukrainian corpora for typical text classification tasks, i.e., different types of style, or harmful speech, or texts relationships. However, the amount of resources required for such corpora collection from scratch is understandable. In this work, we leverage the state-of-the-art advances in NLP, exploring cross-lingual knowledge transfer methods avoiding manual data curation: large multilingual encoders and translation systems, LLMs, and language adapters. We test the approaches on three text classification tasks -- toxicity classification, formality classification, and natural language inference (NLI) -- providing the ``recipe'' for the optimal setups for each task.
Authors: Bowen Xu, Shaoyu Wu, Kai Liu, Lulu Hu
Abstract: With the proliferation of large language models (LLMs), the comprehensive alignment of such models across multiple tasks has emerged as a critical area of research. Existing alignment methodologies primarily address single task, such as multi-turn dialogue, coding, mathematical problem-solving, and tool usage. Although there is a large amount of high-quality data available for those tasks, most of them provide only questions and answers without including the system prompt. Though a detailed analysis of the Qwen language model, we found that the system prompt has a significant impact on both training and inference processes of LLM. We attributes this phenomenon to overfitting to the system prompt. In address this issue, we introduce a novel technique termed Mixture-of-Instructions (MoI), which employs a strategy of instruction packing combined with diverse system prompts to boost the alignment efficiency of language models. We have also compiled a diverse set of seven benchmark datasets to rigorously evaluate the alignment efficacy of the MoI-enhanced language model. Our methodology was applied to the open-source Qwen-7B-chat model, culminating in the development of Qwen-SFT-MoI. This enhanced model demonstrates significant advancements in generative capabilities across coding, mathematics, and tool use tasks.
Authors: Jerry Zhi-Yang He, Sashrika Pandey, Mariah L. Schrum, Anca Dragan
Abstract: When querying a large language model (LLM), the context, i.e. personal, demographic, and cultural information specific to an end-user, can significantly shape the response of the LLM. For example, asking the model to explain Newton's second law with the context "I am a toddler" yields a different answer compared to the context "I am a physics professor." Proper usage of the context enables the LLM to generate personalized responses, whereas inappropriate contextual influence can lead to stereotypical and potentially harmful generations (e.g. associating "female" with "housekeeper"). In practice, striking the right balance when leveraging context is a nuanced and challenging problem that is often situation-dependent. One common approach to address this challenge is to fine-tune LLMs on contextually appropriate responses. However, this approach is expensive, time-consuming, and not controllable for end-users in different situations. In this work, we propose Context Steering (CoS) - a simple training-free method that can be easily applied to autoregressive LLMs at inference time. By measuring the contextual influence in terms of token prediction likelihood and modulating it, our method enables practitioners to determine the appropriate level of contextual influence based on their specific use case and end-user base. We showcase a variety of applications of CoS including amplifying the contextual influence to achieve better personalization and mitigating unwanted influence for reducing model bias. In addition, we show that we can combine CoS with Bayesian Inference to quantify the extent of hate speech on the internet. We demonstrate the effectiveness of CoS on state-of-the-art LLMs and benchmarks.
Authors: Bo Zhang, Hui Ma, Jian Ding, Jian Wang, Bo Xu, Hongfei Lin
Abstract: Integrating multimodal knowledge into large language models (LLMs) represents a significant advancement in dialogue generation capabilities. However, the effective incorporation of such knowledge in zero-resource scenarios remains a substantial challenge due to the scarcity of diverse, high-quality dialogue datasets. To address this, we propose the Visual Implicit Knowledge Distillation Framework (VIKDF), an innovative approach aimed at enhancing LLMs for enriched dialogue generation in zero-resource contexts by leveraging implicit multimodal knowledge. VIKDF comprises two main stages: knowledge distillation, using an Implicit Query Transformer to extract and encode visual implicit knowledge from image-text pairs into knowledge vectors; and knowledge integration, employing a novel Bidirectional Variational Information Fusion technique to seamlessly integrate these distilled vectors into LLMs. This enables the LLMs to generate dialogues that are not only coherent and engaging but also exhibit a deep understanding of the context through implicit multimodal cues, effectively overcoming the limitations of zero-resource scenarios. Our extensive experimentation across two dialogue datasets shows that VIKDF outperforms existing state-of-the-art models in generating high-quality dialogues. The code is available at https://github.com/zhangbo-nlp/VIKDF.
Authors: Leonidas Gee, Milan Gritta, Gerasimos Lampouras, Ignacio Iacobacci
Abstract: Code Language Models have been trained to generate accurate solutions, typically with no regard for runtime. On the other hand, previous works that explored execution optimisation have observed corresponding drops in functional correctness. To that end, we introduce Code-Optimise, a framework that incorporates both correctness (passed, failed) and runtime (quick, slow) as learning signals via self-generated preference data. Our framework is both lightweight and robust as it dynamically selects solutions to reduce overfitting while avoiding a reliance on larger models for learning signals. Code-Optimise achieves significant improvements in pass@k while decreasing the competitive baseline runtimes by an additional 6% for in-domain data and up to 3% for out-of-domain data. As a by-product, the average length of the generated solutions is reduced by up to 48% on MBPP and 23% on HumanEval, resulting in faster and cheaper inference. The generated data and codebase is open-sourced at https://github.com/huawei-noah/HEBO/tree/Code_Optimise.
URLs: https://github.com/huawei-noah/HEBO/tree/Code_Optimise.
Authors: Hakaze Cho, Yoshihiro Sakai, Mariko Kato, Kenshiro Tanaka, Akira Ishii, Naoya Inoue
Abstract: In-Context Learning (ICL) typically utilizes classification criteria from output probabilities of manually selected label tokens. However, we argue that such token-based classification criteria lead to suboptimal decision boundaries, despite delicate calibrations through translation and constrained rotation applied. To address this problem, we propose Hidden Calibration, which renounces token probabilities and uses the nearest centroid classifier on the LM's last hidden states. In detail, we assign the label of the nearest centroid previously estimated from a calibration set to the test sample as the predicted label. Our experiments on 6 models and 10 classification datasets indicate that Hidden Calibration consistently outperforms current token-based baselines by about 20%~50%, achieving a strong state-of-the-art in ICL. Our further analysis demonstrates that Hidden Calibration finds better classification criteria with less inter-class overlap, and LMs provide linearly separable intra-class clusters with the help of demonstrations, which supports Hidden Calibration and gives new insights into the principle of ICL. Our official code implementation can be found at https://github.com/hc495/Hidden_Calibration.
Authors: Pedro Ferreira, Ivan Titov, Wilker Aziz
Abstract: Explanation regularisation (ER) has been introduced as a way to guide text classifiers to form their predictions relying on input tokens that humans consider plausible. This is achieved by introducing an auxiliary explanation loss that measures how well the output of an input attribution technique for the model agrees with human-annotated rationales. The guidance appears to benefit performance in out-of-domain (OOD) settings, presumably due to an increased reliance on "plausible" tokens. However, previous work has under-explored the impact of guidance on that reliance, particularly when reliance is measured using attribution techniques different from those used to guide the model. In this work, we seek to close this gap, and also explore the relationship between reliance on plausible features and OOD performance. We find that the connection between ER and the ability of a classifier to rely on plausible features has been overstated and that a stronger reliance on plausible tokens does not seem to be the cause for OOD improvements.
Authors: Yeongbin Seo, Dongha Lee, Jinyoung Yeo
Abstract: Previous studies on continual knowledge learning (CKL) in large language models (LLMs) have predominantly focused on approaches such as regularization, architectural modifications, and rehearsal techniques to mitigate catastrophic forgetting. However, these methods naively inherit the inefficiencies of standard training procedures, indiscriminately applying uniform weight across all tokens, which can lead to unnecessary parameter updates and increased forgetting. To address these shortcomings, we propose a novel CKL approach termed Train-Attention-Augmented Language Model (TAALM), which enhances learning efficiency by dynamically predicting and applying weights to tokens based on their usefulness. This method employs a meta-learning framework that optimizes token importance predictions, facilitating targeted knowledge updates and minimizing forgetting. Also, we observe that existing benchmarks do not clearly exhibit the trade-off between learning and retaining, therefore we propose a new benchmark, \textsc{LAMA-ckl}, to address this issue. Through experiments conducted on both newly introduced and established CKL benchmarks, TAALM proves the state-of-the-art performance upon the baselines, and also shows synergistic compatibility when integrated with previous CKL approaches.
Authors: Quim Motger, Alessio Miaschi, Felice Dell'Orletta, Xavier Franch, Jordi Marco
Abstract: Mobile app review analysis presents unique challenges due to the low quality, subjective bias, and noisy content of user-generated documents. Extracting features from these reviews is essential for tasks such as feature prioritization and sentiment analysis, but it remains a challenging task. Meanwhile, encoder-only models based on the Transformer architecture have shown promising results for classification and information extraction tasks for multiple software engineering processes. This study explores the hypothesis that encoder-only large language models can enhance feature extraction from mobile app reviews. By leveraging crowdsourced annotations from an industrial context, we redefine feature extraction as a supervised token classification task. Our approach includes extending the pre-training of these models with a large corpus of user reviews to improve contextual understanding and employing instance selection techniques to optimize model fine-tuning. Empirical evaluations demonstrate that this method improves the precision and recall of extracted features and enhances performance efficiency. Key contributions include a novel approach to feature extraction, annotated datasets, extended pre-trained models, and an instance selection mechanism for cost-effective fine-tuning. This research provides practical methods and empirical evidence in applying large language models to natural language processing tasks within mobile app reviews, offering improved performance in feature extraction.
Authors: Yuxin Jiang, Bo Huang, Yufei Wang, Xingshan Zeng, Liangyou Li, Yasheng Wang, Xin Jiang, Lifeng Shang, Ruiming Tang, Wei Wang
Abstract: Direct preference optimization (DPO), a widely adopted offline preference optimization algorithm, aims to align large language models (LLMs) with human-desired behaviors using pairwise preference data. However, the generation of the winning response and the losing response within pairwise data are typically isolated, leading to weak correlations between them as well as suboptimal alignment performance. To address this issue, we propose an effective framework for Bridging and Modeling Correlations in pairwise data, named BMC. Firstly, we increase the consistency and informativeness of the pairwise preference signals through targeted modifications, synthesizing a pseudo-winning response by improving the losing response with the winning response as a reference. Secondly, we identify that DPO alone is insufficient to model these correlations and capture nuanced variations. Therefore, we propose learning token-level correlations by dynamically leveraging the policy model's confidence during training. Comprehensive experiments on QA, math, and instruction-following tasks demonstrate the effectiveness of our approach, significantly surpassing competitive baselines, including DPO. Additionally, our in-depth quantitative analysis reveals the reasons behind our method's superior performance over DPO and showcases its versatility to other DPO variants. We release our repository at https://github.com/YJiangcm/BMC.
Authors: Wei Chen, Zhen Huang, Liang Xie, Binbin Lin, Houqiang Li, Le Lu, Xinmei Tian, Deng Cai, Yonggang Zhang, Wenxiao Wang, Xu Shen, Jieping Ye
Abstract: Large Language Models (LLMs) tend to prioritize adherence to user prompts over providing veracious responses, leading to the sycophancy issue. When challenged by users, LLMs tend to admit mistakes and provide inaccurate responses even if they initially provided the correct answer. Recent works propose to employ supervised fine-tuning (SFT) to mitigate the sycophancy issue, while it typically leads to the degeneration of LLMs' general capability. To address the challenge, we propose a novel supervised pinpoint tuning (SPT), where the region-of-interest modules are tuned for a given objective. Specifically, SPT first reveals and verifies a small percentage (<5%) of the basic modules, which significantly affect a particular behavior of LLMs. i.e., sycophancy. Subsequently, SPT merely fine-tunes these identified modules while freezing the rest. To verify the effectiveness of the proposed SPT, we conduct comprehensive experiments, demonstrating that SPT significantly mitigates the sycophancy issue of LLMs (even better than SFT). Moreover, SPT introduces limited or even no side effects on the general capability of LLMs. Our results shed light on how to precisely, effectively, and efficiently explain and improve the targeted ability of LLMs. Code and data are available at https://github.com/yellowtownhz/sycophancy-interpretability.
URLs: https://github.com/yellowtownhz/sycophancy-interpretability.
Authors: Shikhar Murty, Hao Zhu, Dzmitry Bahdanau, Christopher D. Manning
Abstract: We introduce NNetNav, a method for unsupervised interaction with websites that generates synthetic demonstrations for training browser agents. Given any website, NNetNav produces these demonstrations by retroactively labeling action sequences from an exploration policy. Most work on training browser agents has relied on expensive human supervision, and the limited prior work on such interaction-based techniques has failed to provide effective search through the exponentially large space of exploration. In contrast, NNetNav exploits the hierarchical structure of language instructions to make this search more tractable: Complex instructions are typically decomposable into simpler sub-tasks, allowing NNetNav to automatically prune interaction episodes when an intermediate trajectory cannot be annotated with a meaningful sub-task. \texttt{LLama-3.1-8b} finetuned on 10k NNetNav self-generated demonstrations obtains over 16\% success rate on WebArena, and 35\% on WebVoyager, an improvement of 15pts and 31pts respectively over zero-shot \texttt{LLama-3.1-8b}, outperforming zero-shot GPT-4 and reaching the state-of-the-art among unsupervised methods, for both benchmarks.
Authors: Rhea Sanjay Sukthanker, Benedikt Staffler, Frank Hutter, Aaron Klein
Abstract: Large Language Models (LLMs) demonstrate exceptional reasoning abilities, enabling strong generalization across diverse tasks such as commonsense reasoning and instruction following. However, as LLMs scale, inference costs become increasingly prohibitive, accumulating significantly over their life cycle. In this paper we consider model compression for LLMs to reduce model size while improving downstream task performance. We phrase this as a neural architecture search problem that automatically prunes structural components, such as attention heads, neurons, and layers by searching for the Pareto-optimal set of sub-networks balancing between performance and on-device latency. Compared to state-of-the-art structural pruning approaches and fine-tuned smaller sub-networks extracted from the pre-trained model, our method achieves upto 9.85% improvement on average on 11 diverse downstream tasks, while achieving up to 22% improvement of on-device latency.
Authors: Cheng Li, May Fung, Qingyun Wang, Chi Han, Manling Li, Jindong Wang, Heng Ji
Abstract: Mental health disorders are one of the most serious diseases in the world. Most people with such a disease lack access to adequate care, which highlights the importance of training models for the diagnosis and treatment of mental health disorders. However, in the mental health domain, privacy concerns limit the accessibility of personalized treatment data, making it challenging to build powerful models. In this paper, we introduce MentalArena, a self-play framework to train language models by generating domain-specific personalized data, where we obtain a better model capable of making a personalized diagnosis and treatment (as a therapist) and providing information (as a patient). To accurately model human-like mental health patients, we devise Symptom Encoder, which simulates a real patient from both cognition and behavior perspectives. To address intent bias during patient-therapist interactions, we propose Symptom Decoder to compare diagnosed symptoms with encoded symptoms, and dynamically manage the dialogue between patient and therapist according to the identified deviations. We evaluated MentalArena against 6 benchmarks, including biomedicalQA and mental health tasks, compared to 6 advanced models. Our models, fine-tuned on both GPT-3.5 and Llama-3-8b, significantly outperform their counterparts, including GPT-4o. We hope that our work can inspire future research on personalized care. Code is available in https://github.com/Scarelette/MentalArena/tree/main
Authors: Ling Yang, Zhaochen Yu, Tianjun Zhang, Minkai Xu, Joseph E. Gonzalez, Bin Cui, Shuicheng Yan
Abstract: Large language models (LLMs) like GPT-4, PaLM, and LLaMA have shown significant improvements in various reasoning tasks. However, smaller models such as Llama-3-8B and DeepSeekMath-Base still struggle with complex mathematical reasoning because they fail to effectively identify and correct reasoning errors. Recent reflection-based methods aim to address these issues by enabling self-reflection and self-correction, but they still face challenges in independently detecting errors in their reasoning steps. To overcome these limitations, we propose SuperCorrect, a novel two-stage framework that uses a large teacher model to supervise and correct both the reasoning and reflection processes of a smaller student model. In the first stage, we extract hierarchical high-level and detailed thought templates from the teacher model to guide the student model in eliciting more fine-grained reasoning thoughts. In the second stage, we introduce cross-model collaborative direct preference optimization (DPO) to enhance the self-correction abilities of the student model by following the teacher's correction traces during training. This cross-model DPO approach teaches the student model to effectively locate and resolve erroneous thoughts with error-driven insights from the teacher model, breaking the bottleneck of its thoughts and acquiring new skills and knowledge to tackle challenging problems. Extensive experiments consistently demonstrate our superiority over previous methods. Notably, our SuperCorrect-7B model significantly surpasses powerful DeepSeekMath-7B by 7.8%/5.3% and Qwen2.5-Math-7B by 15.1%/6.3% on MATH/GSM8K benchmarks, achieving new SOTA performance among all 7B models. Code: https://github.com/YangLing0818/SuperCorrect-llm
Authors: Yu-Neng Chuang, Helen Zhou, Prathusha Kameswara Sarma, Parikshit Gopalan, John Boccio, Sara Bolouki, Xia Hu
Abstract: Large language models (LLMs) have demonstrated impressive performance on several tasks and are increasingly deployed in real-world applications. However, especially in high-stakes settings, it becomes vital to know when the output of an LLM may be unreliable. Depending on whether an answer is trustworthy, a system can then choose to route the question to another expert, or otherwise fall back on a safe default behavior. In this work, we study the extent to which LLMs can reliably indicate confidence in their answers, and how this notion of confidence can translate into downstream accuracy gains. We propose Self-REF, a lightweight training strategy to teach LLMs to express confidence in whether their answers are correct in a reliable manner. Self-REF introduces confidence tokens into the LLM, from which a confidence score can be extracted. Compared to conventional approaches such as verbalizing confidence and examining token probabilities, we demonstrate empirically that confidence tokens show significant improvements in downstream routing and rejection learning tasks.
Authors: Georgios Chochlakis, Alexandros Potamianos, Kristina Lerman, Shrikanth Narayanan
Abstract: In-context Learning (ICL) has become the primary method for performing natural language tasks with Large Language Models (LLMs). The knowledge acquired during pre-training is crucial for this few-shot capability, providing the model with task priors. However, recent studies have shown that ICL predominantly relies on retrieving task priors rather than "learning" to perform tasks. This limitation is particularly evident in complex subjective domains such as emotion and morality, where priors significantly influence posterior predictions. In this work, we examine whether this is the result of the aggregation used in corresponding datasets, where trying to combine low-agreement, disparate annotations might lead to annotation artifacts that create detrimental noise in the prompt. Moreover, we evaluate the posterior bias towards certain annotators by grounding our study in appropriate, quantitative measures of LLM priors. Our results indicate that aggregation is a confounding factor in the modeling of subjective tasks, and advocate focusing on modeling individuals instead. However, aggregation does not explain the entire gap between ICL and the state of the art, meaning other factors in such tasks also account for the observed phenomena. Finally, by rigorously studying annotator-level labels, we find that it is possible for minority annotators to both better align with LLMs and have their perspectives further amplified.
Authors: SeongYeub Chu, JongWoo Kim, Bryan Wong, MunYong Yi
Abstract: Existing automated essay scoring (AES) has solely relied on essay text without using explanatory rationales for the scores, thereby forgoing an opportunity to capture the specific aspects evaluated by rubric indicators in a fine-grained manner. This paper introduces Rationale-based Multiple Trait Scoring (RMTS), a novel approach for multi-trait essay scoring that integrates prompt-engineering-based large language models (LLMs) with a fine-tuning-based essay scoring model using a smaller large language model (S-LLM). RMTS uses an LLM-based trait-wise rationale generation system where a separate LLM agent generates trait-specific rationales based on rubric guidelines, which the scoring model uses to accurately predict multi-trait scores. Extensive experiments on benchmark datasets, including ASAP, ASAP++, and Feedback Prize, show that RMTS significantly outperforms state-of-the-art models and vanilla S-LLMs in trait-specific scoring. By assisting quantitative assessment with fine-grained qualitative rationales, RMTS enhances the trait-wise reliability, providing partial explanations about essays. The code is available at https://github.com/BBeeChu/RMTS.git.
Authors: You Wu, Haoyi Wu, Kewei Tu
Abstract: Recently, sharing key-value (KV) cache across layers has been found effective in efficient inference of large language models (LLMs). To systematically investigate different techniques of cross-layer KV sharing, we propose a unified framework that covers several recent methods and their novel variants. We conduct comprehensive experiments on all the configurations of the framework, evaluating their generation throughput and performance in language modeling and downstream tasks. We find that when reducing the size of the KV cache by 2$\times$, most configurations can achieve higher throughput than standard transformers while maintaining competitive performance. When further reducing the size of the KV cache, however, pairing queries of all layers with KVs of upper layers performs better, at the expense of additional training cost and prefilling latency. We hope that this work will help users make more informed choices of cross-layer KV sharing approaches and facilitate future research on efficient LLM inference.
Authors: Koji Inoue, Divesh Lala, Gabriel Skantze, Tatsuya Kawahara
Abstract: In human conversations, short backchannel utterances such as "yeah" and "oh" play a crucial role in facilitating smooth and engaging dialogue. These backchannels signal attentiveness and understanding without interrupting the speaker, making their accurate prediction essential for creating more natural conversational agents. This paper proposes a novel method for real-time, continuous backchannel prediction using a fine-tuned Voice Activity Projection (VAP) model. While existing approaches have relied on turn-based or artificially balanced datasets, our approach predicts both the timing and type of backchannels in a continuous and frame-wise manner on unbalanced, real-world datasets. We first pre-train the VAP model on a general dialogue corpus to capture conversational dynamics and then fine-tune it on a specialized dataset focused on backchannel behavior. Experimental results demonstrate that our model outperforms baseline methods in both timing and type prediction tasks, achieving robust performance in real-time environments. This research offers a promising step toward more responsive and human-like dialogue systems, with implications for interactive spoken dialogue applications such as virtual assistants and robots.
Authors: Mufei Li, Siqi Miao, Pan Li
Abstract: Large Language Models (LLMs) demonstrate strong reasoning abilities but face limitations such as hallucinations and outdated knowledge. Knowledge Graph (KG)-based Retrieval-Augmented Generation (RAG) addresses these issues by grounding LLM outputs in structured external knowledge from KGs. However, current KG-based RAG frameworks still struggle to optimize the trade-off between retrieval effectiveness and efficiency in identifying a suitable amount of relevant graph information for the LLM to digest. We introduce SubgraphRAG, extending the KG-based RAG framework that retrieves subgraphs and leverages LLMs for reasoning and answer prediction. Our approach innovatively integrates a lightweight multilayer perceptron with a parallel triple-scoring mechanism for efficient and flexible subgraph retrieval while encoding directional structural distances to enhance retrieval effectiveness. The size of retrieved subgraphs can be flexibly adjusted to match the query's need and the downstream LLM's capabilities. This design strikes a balance between model complexity and reasoning power, enabling scalable and generalizable retrieval processes. Notably, based on our retrieved subgraphs, smaller LLMs like Llama3.1-8B-Instruct deliver competitive results with explainable reasoning, while larger models like GPT-4o achieve state-of-the-art accuracy compared with previous baselines -- all without fine-tuning. Extensive evaluations on the WebQSP and CWQ benchmarks highlight SubgraphRAG's strengths in efficiency, accuracy, and reliability by reducing hallucinations and improving response grounding.
Authors: Cau\~a Ferreira Barros, Bruna Borges Azevedo, Valdemar Vicente Graciano Neto, Mohamad Kassab, Marcos Kalinowski, Hugo Alexandre D. do Nascimento, Michelle C. G. S. P. Bandeira
Abstract: The exponential growth of text-based data in domains such as healthcare, education, and social sciences has outpaced the capacity of traditional qualitative analysis methods, which are time-intensive and prone to subjectivity. Large Language Models (LLMs), powered by advanced generative AI, have emerged as transformative tools capable of automating and enhancing qualitative analysis. This study systematically maps the literature on the use of LLMs for qualitative research, exploring their application contexts, configurations, methodologies, and evaluation metrics. Findings reveal that LLMs are utilized across diverse fields, demonstrating the potential to automate processes traditionally requiring extensive human input. However, challenges such as reliance on prompt engineering, occasional inaccuracies, and contextual limitations remain significant barriers. This research highlights opportunities for integrating LLMs with human expertise, improving model robustness, and refining evaluation methodologies. By synthesizing trends and identifying research gaps, this study aims to guide future innovations in the application of LLMs for qualitative analysis.
Authors: Xiangyu Peng, Prafulla Kumar Choubey, Caiming Xiong, Chien-Sheng Wu
Abstract: Existing evaluation frameworks for retrieval-augmented generation (RAG) systems focus on answerable queries, but they overlook the importance of appropriately rejecting unanswerable requests. In this paper, we introduce UAEval4RAG, a framework designed to evaluate whether RAG systems can handle unanswerable queries effectively. We define a taxonomy with six unanswerable categories, and UAEval4RAG automatically synthesizes diverse and challenging queries for any given knowledge base with unanswered ratio and acceptable ratio metrics. We conduct experiments with various RAG components, including retrieval models, rewriting methods, rerankers, language models, and prompting strategies, and reveal hidden trade-offs in performance of RAG systems. Our findings highlight the critical role of component selection and prompt design in optimizing RAG systems to balance the accuracy of answerable queries with high rejection rates of unanswerable ones. UAEval4RAG provides valuable insights and tools for developing more robust and reliable RAG systems.
Authors: Weijia Shi, Xiaochuang Han, Chunting Zhou, Weixin Liang, Xi Victoria Lin, Luke Zettlemoyer, Lili Yu
Abstract: We present LMFusion, a framework for empowering pretrained text-only large language models (LLMs) with multimodal generative capabilities, enabling them to understand and generate both text and images in arbitrary sequences. LMFusion leverages existing Llama-3's weights for processing texts autoregressively while introducing additional and parallel transformer modules for processing images with diffusion. During training, the data from each modality is routed to its dedicated modules: modality-specific feedforward layers, query-key-value projections, and normalization layers process each modality independently, while the shared self-attention layers allow interactions across text and image features. By freezing the text-specific modules and only training the image-specific modules, LMFusion preserves the language capabilities of text-only LLMs while developing strong visual understanding and generation abilities. Compared to methods that pretrain multimodal generative models from scratch, our experiments demonstrate that, LMFusion improves image understanding by 20% and image generation by 3.6% using only 50% of the FLOPs while maintaining Llama-3's language capabilities. We also demonstrate that this framework can adapt existing vision-language models with multimodal generation ability. Overall, this framework not only leverages existing computational investments in text-only LLMs but also enables the parallel development of language and vision capabilities, presenting a promising direction for efficient multimodal model development.
Authors: Tsz Kin Lam, Marco Gaido, Sara Papi, Luisa Bentivogli, Barry Haddow
Abstract: Following the remarkable success of Large Language Models (LLMs) in NLP tasks, there is increasing interest in extending their capabilities to speech -- the most common form of communication. The most widespread approach to integrating speech into LLMs is dense feature prepending (DFP), which prepends the projected speech representations to the textual representations, allowing end-to-end training with a speech encoder. This raises questions about the need for a sophisticated speech encoder for DFP and how its performance compares with a standard encoder-decoder (i.e., cross-attention) architecture. We compare DFP and cross-attention under a variety of configurations, such as CTC compression, sequence-level knowledge distillation, on monolingual, bilingual, and multilingual models. To perform a controlled architectural comparison, we train all models from scratch rather than using large pretrained models and use comparable data and parameter settings, testing speech-to-text recognition (ASR) and translation (ST) on MuST-C v1.0 and CoVoST2 datasets. Despite the wide adoption of DFP, our results do not indicate a clear advantage of DFP over cross-attention.
Authors: Mohamed Bayan Kmainasi, Ali Ezzat Shahroor, Amani Al-Ghraibah
Abstract: Large Language Models (LLMs) have shown exceptional capabilities in Natural Language Processing (NLP) across diverse domains. However, their application in specialized tasks such as Legal Judgment Prediction (LJP) for low-resource languages like Arabic remains underexplored. In this work, we address this gap by developing an Arabic LJP dataset, collected and preprocessed from Saudi commercial court judgments. We benchmark state-of-the-art open-source LLMs, including LLaMA-3.2-3B and LLaMA-3.1-8B, under varying configurations such as zero-shot, one-shot, and fine-tuning using QLoRA. Additionally, we used a comprehensive evaluation framework combining quantitative metrics (BLEU and ROUGE) and qualitative assessments (Coherence, legal language, clarity). Our results demonstrate that fine-tuned smaller models achieve comparable performance to larger models in task-specific contexts while offering significant resource efficiency. Furthermore, we investigate the effects of prompt engineering and fine-tuning on model outputs, providing insights into performance variability and instruction sensitivity. By making the dataset, implementation code, and models publicly available, we establish a robust foundation for future research in Arabic legal NLP.
Authors: Nitay Calderon, Roi Reichart, Rotem Dror
Abstract: The "LLM-as-a-judge" paradigm employs Large Language Models (LLMs) as annotators and evaluators in tasks traditionally performed by humans. LLM annotations are widely used, not only in NLP research but also in fields like medicine, psychology, and social science. Despite their role in shaping study results and insights, there is no standard or rigorous procedure to determine whether LLMs can replace human annotators. In this paper, we propose a novel statistical procedure -- the Alternative Annotator Test (alt-test) -- that requires only a modest subset of annotated examples to justify using LLM annotations. Additionally, we introduce a versatile and interpretable measure for comparing LLM judges. To demonstrate our procedure, we curated a diverse collection of ten datasets, consisting of language and vision-language tasks, and conducted experiments with six LLMs and four prompting techniques. Our results show that LLMs can sometimes replace humans with closed-source LLMs (such as GPT-4o), outperforming open-source LLMs, and that prompting techniques yield judges of varying quality. We hope this study encourages more rigorous and reliable practices.
Authors: Giorgio Robino
Abstract: This study introduces Conversation Routines (CR), a structured prompt engineering framework for developing task-oriented dialog systems using Large Language Models (LLMs). While LLMs demonstrate remarkable natural language understanding capabilities, engineering them to reliably execute complex business workflows remains challenging. The proposed CR framework enables the development of Conversation Agentic Systems (CAS) through natural language specifications, embedding task-oriented logic within LLM prompts. This approach provides a systematic methodology for designing and implementing complex conversational workflows while maintaining behavioral consistency. We demonstrate the framework's effectiveness through two proof-of-concept implementations: a Train Ticket Booking System and an Interactive Troubleshooting Copilot. These case studies validate CR's capability to encode sophisticated behavioral patterns and decision logic while preserving natural conversational flexibility. Results show that CR enables domain experts to design conversational workflows in natural language while leveraging custom functions (tools) developed by software engineers, creating an efficient division of responsibilities where developers focus on core API implementation and domain experts handle conversation design. While the framework shows promise in accessibility and adaptability, we identify key challenges including computational overhead, non-deterministic behavior, and domain-specific logic optimization. Future research directions include CR evaluation methods based on prompt engineering frameworks driven by goal-oriented grading criteria, improving scalability for complex multi-agent interactions, and enhancing system robustness to address the identified limitations across diverse business applications.
Authors: Weizhi Fei, Xueyan Niu, Guoqing Xie, Yingqing Liu, Bo Bai, Wei Han
Abstract: Although applications involving long-context inputs are crucial for the effective utilization of large language models (LLMs), they also result in increased computational costs and reduced performance. To address this challenge, we propose an efficient, training-free prompt compression method that retains key information within compressed prompts. We identify specific attention heads in transformer-based LLMs, which we designate as evaluator heads, that are capable of selecting tokens in long inputs that are most significant for inference. Building on this discovery, we develop EHPC, an Evaluator Head-based Prompt Compression method, which enables LLMs to rapidly "skim through" input prompts by leveraging only the first few layers with evaluator heads during the pre-filling stage, subsequently passing only the important tokens to the model for inference. EHPC achieves state-of-the-art results across two mainstream benchmarks: prompt compression and long-context inference acceleration. Consequently, it effectively reduces the complexity and costs associated with commercial API calls. We further demonstrate that EHPC attains competitive results compared to key-value cache-based acceleration methods, thereby highlighting its potential to enhance the efficiency of LLMs for long-context tasks.
Authors: Kangjie Zheng, Junwei Yang, Siyue Liang, Bin Feng, Zequn Liu, Wei Ju, Zhiping Xiao, Ming Zhang
Abstract: Masked Language Models (MLMs) have achieved remarkable success in many self-supervised representation learning tasks. MLMs are trained by randomly masking portions of the input sequences with [MASK] tokens and learning to reconstruct the original content based on the remaining context. This paper explores the impact of [MASK] tokens on MLMs. Analytical studies show that masking tokens can introduce the corrupted semantics problem, wherein the corrupted context may convey multiple, ambiguous meanings. This problem is also a key factor affecting the performance of MLMs on downstream tasks. Based on these findings, we propose a novel enhanced-context MLM, ExLM. Our approach expands [MASK] tokens in the input context and models the dependencies between these expanded states. This enhancement increases context capacity and enables the model to capture richer semantic information, effectively mitigating the corrupted semantics problem during pre-training. Experimental results demonstrate that ExLM achieves significant performance improvements in both text modeling and SMILES modeling tasks. Further analysis confirms that ExLM enriches semantic representations through context enhancement, and effectively reduces the semantic multimodality commonly observed in MLMs.
Authors: Guangzhi Sun, Xiao Zhan, Shutong Feng, Philip C. Woodland, Jose Such
Abstract: Aligning large language models (LLMs) with human values is essential for their safe deployment and widespread adoption. Current LLM safety benchmarks often focus solely on the refusal of individual problematic queries, which overlooks the importance of the context where the query occurs and may cause undesired refusal of queries under safe contexts that diminish user experience. Addressing this gap, we introduce CASE-Bench, a Context-Aware SafEty Benchmark that integrates context into safety assessments of LLMs. CASE-Bench assigns distinct, formally described contexts to categorized queries based on Contextual Integrity theory. Additionally, in contrast to previous studies which mainly rely on majority voting from just a few annotators, we recruited a sufficient number of annotators necessary to ensure the detection of statistically significant differences among the experimental conditions based on power analysis. Our extensive analysis using CASE-Bench on various open-source and commercial LLMs reveals a substantial and significant influence of context on human judgments (p<0.0001 from a z-test), underscoring the necessity of context in safety evaluations. We also identify notable mismatches between human judgments and LLM responses, particularly in commercial models within safe contexts.
Authors: Yubo Wang, Xiang Yue, Wenhu Chen
Abstract: Supervised Fine-Tuning (SFT) is commonly used to train language models to imitate annotated responses for given instructions. In this paper, we challenge this paradigm and propose Critique Fine-Tuning (CFT), a strategy where models learn to critique noisy responses rather than simply imitate correct ones. Inspired by human learning processes that emphasize critical thinking, CFT encourages deeper analysis and nuanced understanding-traits often overlooked by standard SFT. To validate the effectiveness of CFT, we construct a 50K-sample dataset from WebInstruct, using GPT-4o as the teacher to generate critiques in the form of ([query; noisy response], critique). CFT on this dataset yields a consistent 4-10% improvement over SFT on six math benchmarks with different base models like Qwen2.5, Qwen2.5-Math and DeepSeek-Math. We further expand to MetaMath and NuminaMath datasets and observe similar gains over SFT. Notably, our model Qwen2.5-Math-CFT only requires 1 hour training on 8xH100 over the 50K examples. It can match or outperform strong competitors like Qwen2.5-Math-Instruct on most benchmarks, which use over 2M samples. Moreover, it can match the performance of SimpleRL, which is a deepseek-r1 replication trained with 140x more compute. Ablation studies show that CFT is robust to the source of noisy response and teacher critique model. Through these findings, we argue that CFT offers a more effective alternative to advance the reasoning of language models.
Authors: Lang Cao
Abstract: Tables serve as a fundamental format for representing structured relational data. While current language models (LMs) excel at many text-based tasks, they still face challenges in table understanding due to the complex characteristics of tabular data, such as their structured nature. In this paper, we aim to enhance LMs for improved table understanding. We identify four key challenges: 1) difficulty in locating target data, 2) deficiency in table semantics, 3) numerical inaccuracies in textual reasoning, and 4) semantic inflexibility in symbolic reasoning. To address these issues, we propose TableMaster, a recipe and comprehensive framework that integrates multiple solutions to overcome these obstacles. TableMaster first extracts relevant table content and verbalizes it with enriched semantic context. Additionally, we introduce adaptive reasoning, a flexible approach that dynamically adjusts between textual and symbolic reasoning, tailoring the reasoning process to each query. Extensive analyses and experiments demonstrate our findings and the effectiveness of TableMaster. On the WikiTQ dataset, TableMaster achieves an accuracy of 78.13% using GPT-4o-mini, surpassing existing baselines.
Authors: Xin Xu, Qiyun Xu, Tong Xiao, Tianhao Chen, Yuchen Yan, Jiaxin Zhang, Shizhe Diao, Can Yang, Yang Wang
Abstract: Large language models (LLMs) have demonstrated remarkable capabilities in solving complex reasoning tasks, particularly in mathematics. However, the domain of physics reasoning presents unique challenges that have received significantly less attention. Existing benchmarks often fall short in evaluating LLMs' abilities on the breadth and depth of undergraduate-level physics, underscoring the need for a comprehensive evaluation. To fill this gap, we introduce UGPhysics, a large-scale and comprehensive benchmark specifically designed to evaluate UnderGraduate-level Physics (UGPhysics) reasoning with LLMs. UGPhysics includes 5,520 undergraduate-level physics problems in both English and Chinese, covering 13 subjects with seven different answer types and four distinct physics reasoning skills, all rigorously screened for data leakage. Additionally, we develop a Model-Assistant Rule-based Judgment (MARJ) pipeline specifically tailored for assessing answer correctness of physics problems, ensuring accurate evaluation. Our evaluation of 31 leading LLMs shows that the highest overall accuracy, 49.8% (achieved by OpenAI-o1-mini), emphasizes the necessity for models with stronger physics reasoning skills, beyond math abilities. We hope UGPhysics, along with MARJ, will drive future advancements in AI for physics reasoning. Codes and data are available at https://github.com/YangLabHKUST/UGPhysics .
Authors: Donglei Yu, Yang Zhao, Jie Zhu, Yangyifan Xu, Yu Zhou, Chengqing Zong
Abstract: Simultaneous Machine Translation (SiMT) generates translations while receiving streaming source inputs. This requires the SiMT model to learn a read/write policy, deciding when to translate and when to wait for more source input. Numerous linguistic studies indicate that audiences in SiMT scenarios have distinct preferences, such as accurate translations, simpler syntax, and no unnecessary latency. Aligning SiMT models with these human preferences is crucial to improve their performances. However, this issue still remains unexplored. Additionally, preference optimization for SiMT task is also challenging. Existing methods focus solely on optimizing the generated responses, ignoring human preferences related to latency and the optimization of read/write policy during the preference optimization phase. To address these challenges, we propose Simultaneous Preference Learning (SimulPL), a preference learning framework tailored for the SiMT task. In the SimulPL framework, we categorize SiMT human preferences into five aspects: \textbf{translation quality preference}, \textbf{monotonicity preference}, \textbf{key point preference}, \textbf{simplicity preference}, and \textbf{latency preference}. By leveraging the first four preferences, we construct human preference prompts to efficiently guide GPT-4/4o in generating preference data for the SiMT task. In the preference optimization phase, SimulPL integrates \textbf{latency preference} into the optimization objective and enables SiMT models to improve the read/write policy, thereby aligning with human preferences more effectively. Experimental results indicate that SimulPL exhibits better alignment with human preferences across all latency levels in Zh$\rightarrow$En, De$\rightarrow$En and En$\rightarrow$Zh SiMT tasks. Our data and code will be available at https://github.com/EurekaForNLP/SimulPL.
Authors: Alan Zhu, Parth Asawa, Jared Quincy Davis, Lingjiao Chen, Boris Hanin, Ion Stoica, Joseph E. Gonzalez, Matei Zaharia
Abstract: As the demand for high-quality data in model training grows, researchers and developers are increasingly generating synthetic data to tune and train LLMs. A common assumption about synthetic data is that sampling from instruct-tuned models is sufficient; however, these models struggle to produce diverse outputs-a key requirement for generalization. Despite various prompting methods, in this work we show that achieving meaningful diversity from instruct-tuned models remains challenging. In contrast, we find base models without post-training exhibit greater diversity, but are less capable at instruction following and hence of lower quality. Leveraging this insight, we propose Base-Refine (BARE), a synthetic data generation method that combines the diversity of base models with the quality of instruct-tuned models through a two-stage process. With minimal few-shot examples and curation, BARE generates diverse and high-quality datasets, improving downstream task performance. We show that fine-tuning with as few as 1,000 BARE-generated samples can reach performance comparable to the best similarly sized models on LiveCodeBench tasks. Furthermore, fine-tuning with BARE-generated data achieves a 101% improvement over instruct-only data on GSM8K and a 18.4% improvement over SOTA methods on RAFT.
Authors: Alexei Figueroa, Justus Westerhoff, Golzar Atefi, Dennis Fast, Benjamin Winter, Felix Alexader Gers, Alexander L\"oser, Wolfang Nejdl
Abstract: Biologically inspired neural networks offer alternative avenues to model data distributions. FlyVec is a recent example that draws inspiration from the fruit fly's olfactory circuit to tackle the task of learning word embeddings. Surprisingly, this model performs competitively even against deep learning approaches specifically designed to encode text, and it does so with the highest degree of computational efficiency. We pose the question of whether this performance can be improved further. For this, we introduce Comply. By incorporating positional information through complex weights, we enable a single-layer neural network to learn sequence representations. Our experiments show that Comply not only supersedes FlyVec but also performs on par with significantly larger state-of-the-art models. We achieve this without additional parameters. Comply yields sparse contextual representations of sentences that can be interpreted explicitly from the neuron weights.
Authors: Wenhao Zheng, Yixiao Chen, Weitong Zhang, Souvik Kundu, Yun Li, Zhengzhong Liu, Eric P. Xing, Hongyi Wang, Huaxiu Yao
Abstract: Large language models have achieved remarkable success in various tasks but suffer from high computational costs during inference, limiting their deployment in resource-constrained applications. To address this issue, we propose a novel CITER (\textbf{C}ollaborative \textbf{I}nference with \textbf{T}oken-l\textbf{E}vel \textbf{R}outing) framework that enables efficient collaboration between small and large language models (SLMs & LLMs) through a token-level routing strategy. Specifically, CITER routes non-critical tokens to an SLM for efficiency and routes critical tokens to an LLM for generalization quality. We formulate router training as a policy optimization, where the router receives rewards based on both the quality of predictions and the inference costs of generation. This allows the router to learn to predict token-level routing scores and make routing decisions based on both the current token and the future impact of its decisions. To further accelerate the reward evaluation process, we introduce a shortcut which significantly reduces the costs of the reward estimation and improving the practicality of our approach. Extensive experiments on five benchmark datasets demonstrate that CITER reduces the inference costs while preserving high-quality generation, offering a promising solution for real-time and resource-constrained applications.
Authors: Tunazzina Islam, Dan Goldwasser
Abstract: Nowadays, social media is pivotal in shaping public discourse, especially on polarizing issues like vaccination, where diverse moral perspectives influence individual opinions. In NLP, data scarcity and complexity of psycholinguistic tasks, such as identifying morality frames, make relying solely on human annotators costly, time-consuming, and prone to inconsistency due to cognitive load. To address these issues, we leverage large language models (LLMs), which are adept at adapting new tasks through few-shot learning, utilizing a handful of in-context examples coupled with explanations that connect examples to task principles. Our research explores LLMs' potential to assist human annotators in identifying morality frames within vaccination debates on social media. We employ a two-step process: generating concepts and explanations with LLMs, followed by human evaluation using a "think-aloud" tool. Our study shows that integrating LLMs into the annotation process enhances accuracy, reduces task difficulty, lowers cognitive load, suggesting a promising avenue for human-AI collaboration in complex psycholinguistic tasks.
Authors: Calvin Yixiang Cheng, Scott A Hale
Abstract: This study explores computational approaches for measuring moral foundations (MFs) in non-English corpora. Since most resources are developed primarily for English, cross-linguistic applications of moral foundation theory remain limited. Using Chinese as a case study, this paper evaluates the effectiveness of applying English resources to machine translated text, local language lexicons, multilingual language models, and large language models (LLMs) in measuring MFs in non-English texts. The results indicate that machine translation and local lexicon approaches are insufficient for complex moral assessments, frequently resulting in a substantial loss of cultural information. In contrast, multilingual models and LLMs demonstrate reliable cross-language performance with transfer learning, with LLMs excelling in terms of data efficiency. Importantly, this study also underscores the need for human-in-the-loop validation of automated MF assessment, as the most advanced models may overlook cultural nuances in cross-language measurements. The findings highlight the potential of LLMs for cross-language MF measurements and other complex multilingual deductive coding tasks.
Authors: Menglong Cui, Pengzhi Gao, Wei Liu, Jian Luan, Bin Wang
Abstract: Large language models (LLMs) have shown continuously improving multilingual capabilities, and even small-scale open-source models have demonstrated rapid performance enhancement. In this paper, we systematically explore the abilities of open LLMs with less than ten billion parameters to handle multilingual machine translation (MT) tasks. We conduct comprehensive evaluations on six popular LLMs and find that models like Gemma2-9B exhibit impressive multilingual translation capabilities. We then introduce the Parallel-First Monolingual-Second (PFMS) data mixing strategy in the continual pretraining stage to further enhance the MT performance and present GemmaX2-28, a 9B model achieving top-tier multilingual translation performance across 28 languages. Specifically, GemmaX2-28 consistently outperforms the state-of-the-art (SOTA) models such as TowerInstruct and XALMA and achieves competitive performance with Google Translate and GPT-4-turbo.
Authors: Yu Zhang, Yanzhen Shen, SeongKu Kang, Xiusi Chen, Bowen Jin, Jiawei Han
Abstract: With the rapid increase in paper submissions to academic conferences, the need for automated and accurate paper-reviewer matching is more critical than ever. Previous efforts in this area have considered various factors to assess the relevance of a reviewer's expertise to a paper, such as the semantic similarity, shared topics, and citation connections between the paper and the reviewer's previous works. However, most of these studies focus on only one factor, resulting in an incomplete evaluation of the paper-reviewer relevance. To address this issue, we propose a unified model for paper-reviewer matching that jointly considers semantic, topic, and citation factors. To be specific, during training, we instruction-tune a contextualized language model shared across all factors to capture their commonalities and characteristics; during inference, we chain the three factors to enable step-by-step, coarse-to-fine search for qualified reviewers given a submission. Experiments on four datasets (one of which is newly contributed by us) spanning various fields such as machine learning, computer vision, information retrieval, and data mining consistently demonstrate the effectiveness of our proposed Chain-of-Factors model in comparison with state-of-the-art paper-reviewer matching methods and scientific pre-trained language models.
Authors: Yifan Du, Hangyu Guo, Kun Zhou, Wayne Xin Zhao, Jinpeng Wang, Chuyuan Wang, Mingchen Cai, Ruihua Song, Ji-Rong Wen
Abstract: Visual instruction tuning is crucial for enhancing the zero-shot generalization capability of Multi-modal Large Language Models (MLLMs). In this paper, we aim to investigate a fundamental question: ''what makes for good visual instructions''. Through a comprehensive empirical study, we find that instructions focusing on complex visual reasoning tasks are particularly effective in improving the performance of MLLMs, with results correlating to instruction complexity. Based on this insight, we develop a systematic approach to automatically create high-quality complex visual reasoning instructions. Our approach employs a synthesize-complicate-reformulate paradigm, leveraging multiple stages to gradually increase the complexity of the instructions while guaranteeing quality. Based on this approach, we create the ComVint dataset with 32K examples, and fine-tune four MLLMs on it. Experimental results consistently demonstrate the enhanced performance of all compared MLLMs, such as a 27.86% and 27.60% improvement for LLaVA on MME-Perception and MME-Cognition, respectively. Our code and data are publicly available at the link: https://github.com/RUCAIBox/ComVint.
Authors: Minghao Yan, Saurabh Agarwal, Shivaram Venkataraman
Abstract: Speculative Decoding is a widely used technique to speed up inference for Large Language Models (LLMs) without sacrificing quality. When performing inference, speculative decoding uses a smaller draft model to generate speculative tokens and then uses the target LLM to verify those draft tokens. The speedup provided by speculative decoding heavily depends on the choice of the draft model. In this work, we perform a detailed study comprising over 350 experiments with LLaMA-65B and OPT-66B using speculative decoding and delineate the factors that affect the performance gain provided by speculative decoding. Our experiments indicate that the performance of speculative decoding depends heavily on the latency of the draft model, and the draft model's capability in language modeling does not correlate strongly with its performance in speculative decoding. Based on these insights we explore a new design space for draft models and design hardware-efficient draft models for speculative decoding. Our newly designed draft model can provide 111% higher throughput than existing draft models and our approach generalizes further to all LLaMA models (1/2/3.1) and supervised fine-tuned models.
Authors: Franz Louis Cesista
Abstract: Multimodal Foundation Models (MMFMs) have demonstrated strong performance in both computer vision and natural language processing tasks. However, their performance diminishes in tasks that require a high degree of integration between these modalities, such as document understanding. Moreover, finetuning these models and deploying them requires significantly more compute and more engineering effort than unimodal models. In this work, we present Multimodal Structured Generation, a framework that forces (frozen) MMFMs to produce outputs in a strictly structured format by applying hard constraints directly to the output logits. This approach not only ensures that the model generates parseable outputs that downstream APIs can easily ingest but also allows us to force the model to reason before answering, which significantly boosts performance without the need for expensive fine-tuning. We demonstrate the effectiveness of our method through competitive results in the CVPR 2nd MMFM Challenge, highlighting that carefully designed lightweight engineering can outperform expensive and complicated modeling approaches. All of our scripts, deployment steps, and evaluation results can be accessed in https://github.com/leloykun/MMFM-Challenge
Authors: Chen Henry Wu, Rishi Shah, Jing Yu Koh, Ruslan Salakhutdinov, Daniel Fried, Aditi Raghunathan
Abstract: As language models (LMs) are used to build autonomous agents in real environments, ensuring their adversarial robustness becomes a critical challenge. Unlike chatbots, agents are compound systems with multiple components taking actions, which existing LMs safety evaluations do not adequately address. To bridge this gap, we manually create 200 targeted adversarial tasks and evaluation scripts in a realistic threat model on top of VisualWebArena, a real environment for web agents. To systematically examine the robustness of agents, we propose the Agent Robustness Evaluation (ARE) framework. ARE views the agent as a graph showing the flow of intermediate outputs between components and decomposes robustness as the flow of adversarial information on the graph. We find that we can successfully break latest agents that use black-box frontier LMs, including those that perform reflection and tree search. With imperceptible perturbations to a single image (less than 5% of total web page pixels), an attacker can hijack these agents to execute targeted adversarial goals with success rates up to 67%. We also use ARE to rigorously evaluate how the robustness changes as new components are added. We find that inference-time compute that typically improves benign performance can open up new vulnerabilities and harm robustness. An attacker can compromise the evaluator used by the reflexion agent and the value function of the tree search agent, which increases the attack success relatively by 15% and 20%. Our data and code for attacks, defenses, and evaluation are at https://github.com/ChenWu98/agent-attack
Authors: Salem Lahlou, Abdalgader Abubaker, Hakim Hacid
Abstract: Preference optimization methods have been successfully applied to improve not only the alignment of large language models (LLMs) with human values, but also specific natural language tasks such as summarization and stylistic continuations. This paper proposes using preference optimization methods on Chain-of-Thought steps in order to improve the mathematical reasoning performances of language models. While the chosen answers are obtained from datasets that include reasoning traces, we propose two complementary schemes for generating rejected answers: weak LLM prompting, and digit corruption. Our approach leads to increased accuracy on the GSM8K and AQuA-RAT mathematical reasoning benchmarks for Falcon2-11B and Mistral-7B. Additionally, the improved abilities transfer to non-mathematical tasks, including the ARC benchmark and symbolic reasoning challenges. For example, our method can lead to up to relative 8.47% and 18.73% increases in accuracy on the GSM8K and AQuA benchmarks respectively, without any extra annotations. This work suggests that the path towards better language reasoning abilities goes through spending resources on creating high-quality datasets of reasoning traces.
Authors: Manuel Faysse, Hugues Sibille, Tony Wu, Bilel Omrani, Gautier Viaud, C\'eline Hudelot, Pierre Colombo
Abstract: Documents are visually rich structures that convey information through text, but also figures, page layouts, tables, or even fonts. Since modern retrieval systems mainly rely on the textual information they extract from document pages to index documents -often through lengthy and brittle processes-, they struggle to exploit key visual cues efficiently. This limits their capabilities in many practical document retrieval applications such as Retrieval Augmented Generation (RAG). To benchmark current systems on visually rich document retrieval, we introduce the Visual Document Retrieval Benchmark ViDoRe, composed of various page-level retrieval tasks spanning multiple domains, languages, and practical settings. The inherent complexity and performance shortcomings of modern systems motivate a new concept; doing document retrieval by directly embedding the images of the document pages. We release ColPali, a Vision Language Model trained to produce high-quality multi-vector embeddings from images of document pages. Combined with a late interaction matching mechanism, ColPali largely outperforms modern document retrieval pipelines while being drastically simpler, faster and end-to-end trainable. We release models, data, code and benchmarks under open licenses at https://huggingface.co/vidore.
Authors: Inhwa Song, SoHyun Park, Sachin R. Pendse, Jessica Lee Schleider, Munmun De Choudhury, Young-Ho Kim
Abstract: Expressing stressful experiences in words is proven to improve mental and physical health, but individuals often disengage with writing interventions as they struggle to organize their thoughts and emotions. Reflective prompts have been used to provide direction, and large language models (LLMs) have demonstrated the potential to provide tailored guidance. However, current systems often limit users' flexibility to direct their reflections. We thus present ExploreSelf, an LLM-driven application designed to empower users to control their reflective journey, providing adaptive support through dynamically generated questions. Through an exploratory study with 19 participants, we examine how participants explore and reflect on personal challenges using ExploreSelf. Our findings demonstrate that participants valued the flexible navigation of adaptive guidance to control their reflective journey, leading to deeper engagement and insight. Building on our findings, we discuss the implications of designing LLM-driven tools that facilitate user-driven and effective reflection of personal challenges.
Authors: Andy T. Liu, Yi-Cheng Lin, Haibin Wu, Stefan Winkler, Hung-yi Lee
Abstract: Despite their impressive success, training foundation models remains computationally costly. This paper investigates how to efficiently train speech foundation models with self-supervised learning (SSL) under a limited compute budget. We examine critical factors in SSL that impact the budget, including model architecture, model size, and data size. Our goal is to make analytical steps toward understanding the training dynamics of speech foundation models. We benchmark SSL objectives in an entirely comparable setting and find that other factors contribute more significantly to the success of SSL. Our results show that slimmer model architectures outperform common small architectures under the same compute and parameter budget. We demonstrate that the size of the pre-training data remains crucial, even with data augmentation during SSL training, as performance suffers when iterating over limited data. Finally, we identify a trade-off between model size and data size, highlighting an optimal model size for a given compute budget.
Authors: Guobin Shen, Dongcheng Zhao, Yiting Dong, Xiang He, Yi Zeng
Abstract: As large language models (LLMs) become integral to various applications, ensuring both their safety and utility is paramount. Jailbreak attacks, which manipulate LLMs into generating harmful content, pose significant challenges to this balance. Existing defenses, such as prompt engineering and safety fine-tuning, often introduce computational overhead, increase inference latency, and lack runtime flexibility. Moreover, overly restrictive safety measures can degrade model utility by causing refusals of benign queries. In this paper, we introduce Jailbreak Antidote, a method that enables real-time adjustment of LLM safety preferences by manipulating a sparse subset of the model's internal states during inference. By shifting the model's hidden representations along a safety direction with varying strengths, we achieve flexible control over the safety-utility balance without additional token overhead or inference delays. Our analysis reveals that safety-related information in LLMs is sparsely distributed; adjusting approximately 5% of the internal state is as effective as modifying the entire state. Extensive experiments on nine LLMs (ranging from 2 billion to 72 billion parameters), evaluated against ten jailbreak attack methods and compared with six defense strategies, validate the effectiveness and efficiency of our approach. By directly manipulating internal states during reasoning, Jailbreak Antidote offers a lightweight, scalable solution that enhances LLM safety while preserving utility, opening new possibilities for real-time safety mechanisms in widely-deployed AI systems.
Authors: Zhongwu Chen, Chengjin Xu, Dingmin Wang, Zhen Huang, Yong Dou, Jian Guo
Abstract: Retrieval-augmented generation (RAG) has shown promising potential in knowledge intensive question answering (QA). However, existing approaches only consider the query itself, neither specifying the retrieval preferences for the retrievers nor informing the generators of how to refer to the retrieved documents for the answers, which poses a significant challenge to the QA performance. To address these issues, we propose Rule-guided Retrieval-Augmented Generation with LMs, which explicitly introduces rules for in-context learning (RuleRAG-ICL) to guide retrievers to recall related documents in the directions of rules and uniformly guide generators to reason attributed by the same rules. Moreover, most existing RAG datasets were constructed without considering rules and Knowledge Graphs (KGs) are recognized as providing high-quality rules. Therefore, we construct five rule-aware RAG benchmarks for QA, RuleQA, based on KGs to stress the significance of retrieval and reasoning with rules. Experiments on RuleQA demonstrate RuleRAG-ICL improves the retrieval quality of +89.2% in Recall@10 and answer accuracy of +103.1% in Exact Match, and RuleRAG-FT yields more enhancement. In addition, experiments on four existing RAG datasets show RuleRAG is also effective by offering rules in RuleQA to them, further proving the generalization of rule guidance in RuleRAG.
Authors: Mohammed Saqr, Sonsoles L\'opez-Pernas, Tiina T\"orm\"anen, Rogers Kaliisa, Kamila Misiejuk, Santtu Tikka
Abstract: This paper presents a novel learning analytics method: Transition Network Analysis (TNA), a method that integrates Stochastic Process Mining and probabilistic graph representation to model, visualize, and identify transition patterns in the learning process data. Combining the relational and temporal aspects into a single lens offers capabilities beyond either framework, including centralities to capture important learning events, community detection to identify behavior patterns, and clustering to reveal temporal patterns. Furthermore, TNA introduces several significance tests that go beyond either method and add rigor to the analysis. Here, we introduce the theoretical and mathematical foundations of TNA and we demonstrate the functionalities of TNA with a case study where students (n=191) engaged in small-group collaboration to map patterns of group dynamics using the theories of co-regulation and socially-shared regulated learning. The analysis revealed that TNA can map the regulatory processes as well as identify important events, patterns, and clusters. Bootstrap validation established the significant transitions and eliminated spurious transitions. As such, TNA can capture learning dynamics and provide a robust framework for investigating the temporal evolution of learning processes. Future directions include -- inter alia -- expanding estimation methods, reliability assessment, and building longitudinal TNA.
Authors: Xuehai He, Shuohang Wang, Jianwei Yang, Xiaoxia Wu, Yiping Wang, Kuan Wang, Zheng Zhan, Olatunji Ruwase, Yelong Shen, Xin Eric Wang
Abstract: Recent advancements in diffusion models have shown great promise in producing high-quality video content. However, efficiently training video diffusion models capable of integrating directional guidance and controllable motion intensity remains a challenging and under-explored area. To tackle these challenges, this paper introduces Mojito, a diffusion model that incorporates both motion trajectory and intensity control for text-to-video generation. Specifically, Mojito features a Directional Motion Control (DMC) module that leverages cross-attention to efficiently direct the generated object's motion without training, alongside a Motion Intensity Modulator (MIM) that uses optical flow maps generated from videos to guide varying levels of motion intensity. Extensive experiments demonstrate Mojito's effectiveness in achieving precise trajectory and intensity control with high computational efficiency, generating motion patterns that closely match specified directions and intensities, providing realistic dynamics that align well with natural motion in real-world scenarios.
Authors: Ahmed Heakl, Sara Ghaboura, Omkar Thawkar, Fahad Shahbaz Khan, Hisham Cholakkal, Rao Muhammad Anwer, Salman Khan
Abstract: Amid the swift progress of large language models (LLMs) and their evolution into large multimodal models (LMMs), significant strides have been made in high-resource languages such as English and Chinese. While Arabic LLMs have seen notable progress, Arabic LMMs remain largely unexplored, often narrowly focusing on a few specific aspects of the language and visual understanding. To bridge this gap, we introduce AIN-the Arabic Inclusive Multimodal Model-designed to excel across diverse domains. AIN is an English-Arabic bilingual LMM designed to excel in English and Arabic, leveraging carefully constructed 3.6 million high-quality Arabic-English multimodal data samples. AIN demonstrates state-of-the-art Arabic performance, while also possessing strong English-language visual capabilities. On the recent CAMEL-Bench benchmark comprising 38 sub-domains including, multi-image understanding, complex visual perception, handwritten document understanding, video understanding, medical imaging, plant diseases, and remote sensing-based land use understanding, our AIN demonstrates strong performance with the 7B model outperforming GPT-4o by an absolute gain of 3.4% averaged over eight domains and 38 sub-domains. AIN's superior capabilities position it as a significant step toward empowering Arabic speakers with advanced multimodal generative AI tools across diverse applications.
Authors: Xiaotong Ji, Shyam Sundhar Ramesh, Matthieu Zimmer, Ilija Bogunovic, Jun Wang, Haitham Bou Ammar
Abstract: Even highly capable large language models (LLMs) can produce biased or unsafe responses, and alignment techniques, such as RLHF, aimed at mitigating this issue, are expensive and prone to overfitting as they retrain the LLM. This paper introduces a novel inference-time alignment approach that ensures LLMs generate safe responses almost surely, i.e., with a probability approaching one. We achieve this by framing the safe generation of inference-time responses as a constrained Markov decision process within the LLM's latent space. Crucially, we augment a safety state that tracks the evolution of safety constraints and enables us to demonstrate formal safety guarantees upon solving the MDP in the latent space. Building on this foundation, we propose InferenceGuard, a practical implementation that safely aligns LLMs without modifying the model weights. Empirically, we demonstrate InferenceGuard effectively balances safety and task performance, outperforming existing inference-time alignment methods in generating safe and aligned responses.
Authors: Abdelrahman Abdallah, Bhawna Piryani, Jamshid Mozafari, Mohammed Ali, Adam Jatowt
Abstract: Retrieval, re-ranking, and retrieval-augmented generation (RAG) are critical components of modern natural language processing (NLP) applications in information retrieval, question answering, and knowledge-based text generation. However, existing solutions are often fragmented, lacking a unified framework that easily integrates these essential processes. The absence of a standardized implementation, coupled with the complexity of retrieval and re-ranking workflows, makes it challenging for researchers to compare and evaluate different approaches in a consistent environment. While existing toolkits such as Rerankers and RankLLM provide general-purpose reranking pipelines, they often lack the flexibility required for fine-grained experimentation and benchmarking. In response to these challenges, we introduce \textbf{Rankify}, a powerful and modular open-source toolkit designed to unify retrieval, re-ranking, and RAG within a cohesive framework. Rankify supports a wide range of retrieval techniques, including dense and sparse retrievers, while incorporating state-of-the-art re-ranking models to enhance retrieval quality. Additionally, Rankify includes a collection of pre-retrieved datasets to facilitate benchmarking, available at Huggingface (https://huggingface.co/datasets/abdoelsayed/reranking-datasets). To encourage adoption and ease of integration, we provide comprehensive documentation (http://rankify.readthedocs.io/), an open-source implementation on GitHub(https://github.com/DataScienceUIBK/rankify), and a PyPI package for effortless installation(https://pypi.org/project/rankify/). By providing a unified and lightweight framework, Rankify allows researchers and practitioners to advance retrieval and re-ranking methodologies while ensuring consistency, scalability, and ease of use.
URLs: https://huggingface.co/datasets/abdoelsayed/reranking-datasets)., http://rankify.readthedocs.io/),, https://github.com/DataScienceUIBK/rankify),, https://pypi.org/project/rankify/).