llm - 2024_12
Navigation
Papers
| Paper | Date | Comment |
|---|---|---|
| TurboAttention: Efficient Attention Approximation For High Throughputs LLMs Large language model (LLM) inference demands significant amount of computation and memory, especially in the key attention mechanism. While techniques, such as quantization and acceleration algorithms, like FlashAttention, have improved efficiency of the overall inference, they address different aspects of the problem: quantization focuses on weight-activation operations, while FlashAttention improves execution but requires high-precision formats. Recent Key-value (KV) cache quantization reduces memory bandwidth but still needs floating-point dequantization for attention operation. We present TurboAttention, a comprehensive approach to enable quantized execution of attention that simultaneously addresses both memory and computational efficiency. Our solution introduces two key innovations: FlashQ, a headwise attention quantization technique that enables both compression of KV cache and quantized execution of activation-activation multiplication, and Sparsity-based Softmax Approximation (SAS), which eliminates the need for dequantization to FP32 during exponentiation operation in attention. Experimental results demonstrate that TurboAttention achieves 1.2-1.8x speedup in attention, reduces the KV cache size by over 4.4x, and enables up to 2.37x maximum throughput over the FP16 baseline while outperforming state-of-the-art quantization and compression techniques across various datasets and models. |
2024-12-17 | |
| LLMCL-GEC: Advancing Grammatical Error Correction with LLM-Driven Curriculum Learning While large-scale language models (LLMs) have demonstrated remarkable capabilities in specific natural language processing (NLP) tasks, they may still lack proficiency compared to specialized models in certain domains, such as grammatical error correction (GEC). Drawing inspiration from the concept of curriculum learning, we have delved into refining LLMs into proficient GEC experts by devising effective curriculum learning (CL) strategies. In this paper, we introduce a novel approach, termed LLM-based curriculum learning, which capitalizes on the robust semantic comprehension and discriminative prowess inherent in LLMs to gauge the complexity of GEC training data. Unlike traditional curriculum learning techniques, our method closely mirrors human expert-designed curriculums. Leveraging the proposed LLM-based CL method, we sequentially select varying levels of curriculums ranging from easy to hard, and iteratively train and refine using the pretrianed T5 and LLaMA series models. Through rigorous testing and analysis across diverse benchmark assessments in English GEC, including the CoNLL14 test, BEA19 test, and BEA19 development sets, our approach showcases a significant performance boost over baseline models and conventional curriculum learning methodologies. |
2024-12-17 | Derek F. Wong is the corresponding author. The preprint version consists of 15 Pages, 5 Figures, 5 Tables, and 3 Appendices |
| Recent advancements in LLM Red-Teaming: Techniques, Defenses, and Ethical Considerations Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language processing tasks, but their vulnerability to jailbreak attacks poses significant security risks. This survey paper presents a comprehensive analysis of recent advancements in attack strategies and defense mechanisms within the field of Large Language Model (LLM) red-teaming. We analyze various attack methods, including gradient-based optimization, reinforcement learning, and prompt engineering approaches. We discuss the implications of these attacks on LLM safety and the need for improved defense mechanisms. This work aims to provide a thorough understanding of the current landscape of red-teaming attacks and defenses on LLMs, enabling the development of more secure and reliable language models. |
2024-12-17 | 16 pages, 2 figures |
| On the Structural Memory of LLM Agents Memory plays a pivotal role in enabling large language model~(LLM)-based agents to engage in complex and long-term interactions, such as question answering (QA) and dialogue systems. While various memory modules have been proposed for these tasks, the impact of different memory structures across tasks remains insufficiently explored. This paper investigates how memory structures and memory retrieval methods affect the performance of LLM-based agents. Specifically, we evaluate four types of memory structures, including chunks, knowledge triples, atomic facts, and summaries, along with mixed memory that combines these components. In addition, we evaluate three widely used memory retrieval methods: single-step retrieval, reranking, and iterative retrieval. Extensive experiments conducted across four tasks and six datasets yield the following key insights: (1) Different memory structures offer distinct advantages, enabling them to be tailored to specific tasks; (2) Mixed memory structures demonstrate remarkable resilience in noisy environments; (3) Iterative retrieval consistently outperforms other methods across various scenarios. Our investigation aims to inspire further research into the design of memory systems for LLM-based agents. |
2024-12-17 | |
| RL-LLM-DT: An Automatic Decision Tree Generation Method Based on RL Evaluation and LLM Enhancement Traditionally, AI development for two-player zero-sum games has relied on two primary techniques: decision trees and reinforcement learning (RL). A common approach involves using a fixed decision tree as one player’s strategy while training an RL agent as the opponent to identify vulnerabilities in the decision tree, thereby improving its strategic strength iteratively. However, this process often requires significant human intervention to refine the decision tree after identifying its weaknesses, resulting in inefficiencies and hindering full automation of the strategy enhancement process. Fortunately, the advent of Large Language Models (LLMs) offers a transformative opportunity to automate the process. We propose RL-LLM-DT, an automatic decision tree generation method based on RL Evaluation and LLM Enhancement. Given an initial decision tree, the method involves two important iterative steps. Response Policy Search: RL is used to discover counter-strategies targeting the decision tree. Policy Improvement: LLMs analyze failure scenarios and generate improved decision tree code. In our method, RL focuses on finding the decision tree’s flaws while LLM is prompted to generate an improved version of the decision tree. The iterative refinement process terminates when RL can’t find any flaw of the tree or LLM fails to improve the tree. To evaluate the effectiveness of this integrated approach, we conducted experiments in a curling game. After iterative refinements, our curling AI based on the decision tree ranks first on the Jidi platform among 34 curling AIs in total, which demonstrates that LLMs can significantly enhance the robustness and adaptability of decision trees, representing a substantial advancement in the field of Game AI. Our code is available at https://github.com/Linjunjie99/RL-LLM-DT. |
2024-12-17 | Length:10 pages. Figures:10 figures. Additional Notes:In this paper, we have introduced a novel hybrid approach which leverages the strengths of both RL and LLMs to itera- tively refine decision tree tactics, enhancing their performance and adaptability |
| Can You Trust LLM Judgments? Reliability of LLM-as-a-Judge Large Language Models (LLMs) have become increasingly powerful and ubiquitous, but their stochastic nature poses challenges to the reliability of their outputs. While deterministic settings can improve consistency, they do not guarantee reliability, as a single sample from the model’s probability distribution can still be misleading. Building upon the concept of LLM-as-a-judge, we introduce a novel framework for rigorously evaluating the reliability of LLM judgments, leveraging McDonald’s omega. We evaluate the reliability of LLMs when judging the outputs of other LLMs on standard single-turn and multi-turn benchmarks, simultaneously investigating the impact of temperature on reliability. By analyzing these results, we demonstrate the limitations of fixed randomness and the importance of considering multiple samples, which we show has significant implications for downstream applications. Our findings highlight the need for a nuanced understanding of LLM reliability and the potential risks associated with over-reliance on single-shot evaluations. This work provides a crucial step towards building more trustworthy and reliable LLM-based systems and applications. |
2024-12-17 | |
| Boosting LLM-based Relevance Modeling with Distribution-Aware Robust Learning With the rapid advancement of pre-trained large language models (LLMs), recent endeavors have leveraged the capabilities of LLMs in relevance modeling, resulting in enhanced performance. This is usually done through the process of fine-tuning LLMs on specifically annotated datasets to determine the relevance between queries and items. However, there are two limitations when LLMs are naively employed for relevance modeling through fine-tuning and inference. First, it is not inherently efficient for performing nuanced tasks beyond simple yes or no answers, such as assessing search relevance. It may therefore tend to be overconfident and struggle to distinguish fine-grained degrees of relevance (e.g., strong relevance, weak relevance, irrelevance) used in search engines. Second, it exhibits significant performance degradation when confronted with data distribution shift in real-world scenarios. In this paper, we propose a novel Distribution-Aware Robust Learning framework (DaRL) for relevance modeling in Alipay Search. Specifically, we design an effective loss function to enhance the discriminability of LLM-based relevance modeling across various fine-grained degrees of query-item relevance. To improve the generalizability of LLM-based relevance modeling, we first propose the Distribution-Aware Sample Augmentation (DASA) module. This module utilizes out-of-distribution (OOD) detection techniques to actively select appropriate samples that are not well covered by the original training set for model fine-tuning. Furthermore, we adopt a multi-stage fine-tuning strategy to simultaneously improve in-distribution (ID) and OOD performance, bridging the performance gap between them. DaRL has been deployed online to serve the Alipay’s insurance product search… |
2024-12-17 | 8 pages |
| A System for Microserving of LLMs The recent advances in LLMs bring a strong demand for efficient system support to improve overall serving efficiency. As LLM inference scales towards multiple GPUs and even multiple compute nodes, various coordination patterns, such as prefill-decode disaggregation and context migration, arise in serving systems. Most inference services today expose a coarse-grained request-level API with a pre-configured coordination strategy, limiting the ability to customize and dynamically reconfigure the coordination. In this paper, we propose LLM microserving, a multi-level architecture for structuring and programming LLM inference services. We introduces simple yet effective microserving APIs to support fine-grained sub-request level actions. A programmable router transforms user requests into sub-request calls, enabling the dynamic reconfiguration of serving patterns. To support diverse execution patterns, we develop a unified KV cache interface that handles various KV compute, transfer, and reuse scenarios. Our evaluation shows that LLM microserving can be reconfigured to support multiple disaggregation orchestration strategies in a few lines of Python code while maintaining state-of-the-art performance for LLM inference tasks. Additionally, it allows us to explore new strategy variants that reduce up to 47% of job completion time compared to the existing strategies. |
2024-12-17 | |
| LLM is Knowledge Graph Reasoner: LLM’s Intuition-aware Knowledge Graph Reasoning for Cold-start Sequential Recommendation Knowledge Graphs (KGs) represent relationships between entities in a graph structure and have been widely studied as promising tools for realizing recommendations that consider the accurate content information of items. However, traditional KG-based recommendation methods face fundamental challenges: insufficient consideration of temporal information and poor performance in cold-start scenarios. On the other hand, Large Language Models (LLMs) can be considered databases with a wealth of knowledge learned from the web data, and they have recently gained attention due to their potential application as recommendation systems. Although approaches that treat LLMs as recommendation systems can leverage LLMs’ high recommendation literacy, their input token limitations make it impractical to consider the entire recommendation domain dataset and result in scalability issues. To address these challenges, we propose a LLM’s Intuition-aware Knowledge graph Reasoning model (LIKR). Our main idea is to treat LLMs as reasoners that output intuitive exploration strategies for KGs. To integrate the knowledge of LLMs and KGs, we trained a recommendation agent through reinforcement learning using a reward function that integrates different recommendation strategies, including LLM’s intuition and KG embeddings. By incorporating temporal awareness through prompt engineering and generating textual representations of user preferences from limited interactions, LIKR can improve recommendation performance in cold-start scenarios. Furthermore, LIKR can avoid scalability issues by using KGs to represent recommendation domain datasets and limiting the LLM’s output to KG exploration strategies. Experiments on real-world datasets demonstrate that our model outperforms state-of-the-art recommendation methods in cold-start sequential recommendation scenarios. |
2024-12-17 | Accepted to the 47th European Conference on Information Retrieval (ECIR2025) |
| LITA: An Efficient LLM-assisted Iterative Topic Augmentation Framework Topic modeling is widely used for uncovering thematic structures within text corpora, yet traditional models often struggle with specificity and coherence in domain-focused applications. Guided approaches, such as SeededLDA and CorEx, incorporate user-provided seed words to improve relevance but remain labor-intensive and static. Large language models (LLMs) offer potential for dynamic topic refinement and discovery, yet their application often incurs high API costs. To address these challenges, we propose the LLM-assisted Iterative Topic Augmentation framework (LITA), an LLM-assisted approach that integrates user-provided seeds with embedding-based clustering and iterative refinement. LITA identifies a small number of ambiguous documents and employs an LLM to reassign them to existing or new topics, minimizing API costs while enhancing topic quality. Experiments on two datasets across topic quality and clustering performance metrics demonstrate that LITA outperforms five baseline models, including LDA, SeededLDA, CorEx, BERTopic, and PromptTopic. Our work offers an efficient and adaptable framework for advancing topic modeling and text clustering. |
2024-12-17 | Under Review |
| Graph Learning in the Era of LLMs: A Survey from the Perspective of Data, Models, and Tasks With the increasing prevalence of cross-domain Text-Attributed Graph (TAG) Data (e.g., citation networks, recommendation systems, social networks, and ai4science), the integration of Graph Neural Networks (GNNs) and Large Language Models (LLMs) into a unified Model architecture (e.g., LLM as enhancer, LLM as collaborators, LLM as predictor) has emerged as a promising technological paradigm. The core of this new graph learning paradigm lies in the synergistic combination of GNNs’ ability to capture complex structural relationships and LLMs’ proficiency in understanding informative contexts from the rich textual descriptions of graphs. Therefore, we can leverage graph description texts with rich semantic context to fundamentally enhance Data quality, thereby improving the representational capacity of model-centric approaches in line with data-centric machine learning principles. By leveraging the strengths of these distinct neural network architectures, this integrated approach addresses a wide range of TAG-based Task (e.g., graph learning, graph reasoning, and graph question answering), particularly in complex industrial scenarios (e.g., supervised, few-shot, and zero-shot settings). In other words, we can treat text as a medium to enable cross-domain generalization of graph learning Model, allowing a single graph model to effectively handle the diversity of downstream graph-based Task across different data domains. This work serves as a foundational reference for researchers and practitioners looking to advance graph learning methodologies in the rapidly evolving landscape of LLM. We consistently maintain the related open-source materials at \url{https://github.com/xkLi-Allen/Awesome-GNN-in-LLMs-Papers}. |
2024-12-17 | In progress |
| Bridging the Gap: Enhancing LLM Performance for Low-Resource African Languages with New Benchmarks, Fine-Tuning, and Cultural Adjustments Large Language Models (LLMs) have shown remarkable performance across various tasks, yet significant disparities remain for non-English languages, and especially native African languages. This paper addresses these disparities by creating approximately 1 million human-translated words of new benchmark data in 8 low-resource African languages, covering a population of over 160 million speakers of: Amharic, Bambara, Igbo, Sepedi (Northern Sotho), Shona, Sesotho (Southern Sotho), Setswana, and Tsonga. Our benchmarks are translations of Winogrande and three sections of MMLU: college medicine, clinical knowledge, and virology. Using the translated benchmarks, we report previously unknown performance gaps between state-of-the-art (SOTA) LLMs in English and African languages. Finally, using results from over 400 fine-tuned models, we explore several methods to reduce the LLM performance gap, including high-quality dataset fine-tuning (using an LLM-as-an-Annotator), cross-lingual transfer, and cultural appropriateness adjustments. Key findings include average mono-lingual improvements of 5.6% with fine-tuning (with 5.4% average mono-lingual improvements when using high-quality data over low-quality data), 2.9% average gains from cross-lingual transfer, and a 3.0% out-of-the-box performance boost on culturally appropriate questions. The publicly available benchmarks, translations, and code from this study support further research and development aimed at creating more inclusive and effective language technologies. |
2024-12-16 | Accepted to AAAI 2025. Main content is 9 pages, 3 figures. Includes supplementary materials |
| Automated Generation of Massive Reasonable Empirical Theorems by Forward Reasoning Based on Strong Relevant Logics – A Solution to the Problem of LLM Pre-training Data Exhaustion Recently, it is often said that the data used for the pre-training of large language models (LLMs) have been exhausted. This paper proposes a solution to the problem: Automated generation of massive reasonable empirical theorems by forward reasoning based on strong relevant logics. In fact, this can be regarded as a part of our approach to the problems of ATF (Automated Theorem Finding) and AKA (Automated Knowledge Appreciation). |
2024-12-16 | 11 pages, 7 figures |
| Interpretable LLM-based Table Question Answering Interpretability for Table Question Answering (Table QA) is critical, particularly in high-stakes industries like finance or healthcare. Although recent approaches using Large Language Models (LLMs) have significantly improved Table QA performance, their explanations for how the answers are generated are ambiguous. To fill this gap, we introduce Plan-of-SQLs ( or POS), an interpretable, effective, and efficient approach to Table QA that answers an input query solely with SQL executions. Through qualitative and quantitative evaluations with human and LLM judges, we show that POS is most preferred among explanation methods, helps human users understand model decision boundaries, and facilitates model success and error identification. Furthermore, when evaluated in standard benchmarks (TabFact, WikiTQ, and FetaQA), POS achieves competitive or superior accuracy compared to existing methods, while maintaining greater efficiency by requiring significantly fewer LLM calls and database queries. |
2024-12-16 | |
| Advanced ingestion process powered by LLM parsing for RAG system Retrieval Augmented Generation (RAG) systems struggle with processing multimodal documents of varying structural complexity. This paper introduces a novel multi-strategy parsing approach using LLM-powered OCR to extract content from diverse document types, including presentations and high text density files both scanned or not. The methodology employs a node-based extraction technique that creates relationships between different information types and generates context-aware metadata. By implementing a Multimodal Assembler Agent and a flexible embedding strategy, the system enhances document comprehension and retrieval capabilities. Experimental evaluations across multiple knowledge bases demonstrate the approach’s effectiveness, showing improvements in answer relevancy and information faithfulness. |
2024-12-16 | 12 pages, 3 figures |
| Can Machines Think Like Humans? A Behavioral Evaluation of LLM-Agents in Dictator Games As Large Language Model (LLM)-based agents increasingly undertake real-world tasks and engage with human society, how well do we understand their behaviors? We (1) investigate how LLM agents’ prosocial behaviors – a fundamental social norm – can be induced by different personas and benchmarked against human behaviors; and (2) introduce a behavioral and social science approach to evaluate LLM agents’ decision-making. We explored how different personas and experimental framings affect these AI agents’ altruistic behavior in dictator games and compared their behaviors within the same LLM family, across various families, and with human behaviors. The findings reveal substantial variations and inconsistencies among LLMs and notable differences compared to human behaviors. Merely assigning a human-like identity to LLMs does not produce human-like behaviors. Despite being trained on extensive human-generated data, these AI agents are unable to capture the internal processes of human decision-making. Their alignment with human is highly variable and dependent on specific model architectures and prompt formulations; even worse, such dependence does not follow a clear pattern. LLMs can be useful task-specific tools but are not yet intelligent human-like agents. |
2024-12-16 | |
| Second Language (Arabic) Acquisition of LLMs via Progressive Vocabulary Expansion This paper addresses the critical need for democratizing large language models (LLM) in the Arab world, a region that has seen slower progress in developing models comparable to state-of-the-art offerings like GPT-4 or ChatGPT 3.5, due to a predominant focus on mainstream languages (e.g., English and Chinese). One practical objective for an Arabic LLM is to utilize an Arabic-specific vocabulary for the tokenizer that could speed up decoding. However, using a different vocabulary often leads to a degradation of learned knowledge since many words are initially out-of-vocabulary (OOV) when training starts. Inspired by the vocabulary learning during Second Language (Arabic) Acquisition for humans, the released AraLLaMA employs progressive vocabulary expansion, which is implemented by a modified BPE algorithm that progressively extends the Arabic subwords in its dynamic vocabulary during training, thereby balancing the OOV ratio at every stage. The ablation study demonstrated the effectiveness of Progressive Vocabulary Expansion. Moreover, AraLLaMA achieves decent performance comparable to the best Arabic LLMs across a variety of Arabic benchmarks. Models, training data, benchmarks, and codes will be all open-sourced. |
2024-12-16 | |
| LLMs for Cold-Start Cutting Plane Separator Configuration Mixed integer linear programming (MILP) solvers ship with a staggering number of parameters that are challenging to select a priori for all but expert optimization users, but can have an outsized impact on the performance of the MILP solver. Existing machine learning (ML) approaches to configure solvers require training ML models by solving thousands of related MILP instances, generalize poorly to new problem sizes, and often require implementing complex ML pipelines and custom solver interfaces that can be difficult to integrate into existing optimization workflows. In this paper, we introduce a new LLM-based framework to configure which cutting plane separators to use for a given MILP problem with little to no training data based on characteristics of the instance, such as a natural language description of the problem and the associated LaTeX formulation. We augment these LLMs with descriptions of cutting plane separators available in a given solver, grounded by summarizing the existing research literature on separators. While individual solver configurations have a large variance in performance, we present a novel ensembling strategy that clusters and aggregates configurations to create a small portfolio of high-performing configurations. Our LLM-based methodology requires no custom solver interface, can find a high-performing configuration by solving only a small number of MILPs, and can generate the configuration with simple API calls that run in under a second. Numerical results show our approach is competitive with existing configuration approaches on a suite of classic combinatorial optimization problems and real-world datasets with only a fraction of the training data and computation time. |
2024-12-16 | |
| RTP-LX: Can LLMs Evaluate Toxicity in Multilingual Scenarios? Large language models (LLMs) and small language models (SLMs) are being adopted at remarkable speed, although their safety still remains a serious concern. With the advent of multilingual S/LLMs, the question now becomes a matter of scale: can we expand multilingual safety evaluations of these models with the same velocity at which they are deployed? To this end, we introduce RTP-LX, a human-transcreated and human-annotated corpus of toxic prompts and outputs in 28 languages. RTP-LX follows participatory design practices, and a portion of the corpus is especially designed to detect culturally-specific toxic language. We evaluate 10 S/LLMs on their ability to detect toxic content in a culturally-sensitive, multilingual scenario. We find that, although they typically score acceptably in terms of accuracy, they have low agreement with human judges when scoring holistically the toxicity of a prompt; and have difficulty discerning harm in context-dependent scenarios, particularly with subtle-yet-harmful content (e.g. microaggressions, bias). We release this dataset to contribute to further reduce harmful uses of these models and improve their safe deployment. |
2024-12-16 | AAAI 2025–camera ready + extended abstract |
| The Open Source Advantage in Large Language Models (LLMs) Large language models (LLMs) mark a key shift in natural language processing (NLP), having advanced text generation, translation, and domain-specific reasoning. Closed-source models like GPT-4, powered by proprietary datasets and extensive computational resources, lead with state-of-the-art performance today. However, they face criticism for their “black box” nature and for limiting accessibility in a manner that hinders reproducibility and equitable AI development. By contrast, open-source initiatives like LLaMA and BLOOM prioritize democratization through community-driven development and computational efficiency. These models have significantly reduced performance gaps, particularly in linguistic diversity and domain-specific applications, while providing accessible tools for global researchers and developers. Notably, both paradigms rely on foundational architectural innovations, such as the Transformer framework by Vaswani et al. (2017). Closed-source models excel by scaling effectively, while open-source models adapt to real-world applications in underrepresented languages and domains. Techniques like Low-Rank Adaptation (LoRA) and instruction-tuning datasets enable open-source models to achieve competitive results despite limited resources. To be sure, the tension between closed-source and open-source approaches underscores a broader debate on transparency versus proprietary control in AI. Ethical considerations further highlight this divide. Closed-source systems restrict external scrutiny, while open-source models promote reproducibility and collaboration but lack standardized auditing documentation frameworks to mitigate biases. Hybrid approaches that leverage the strengths of both paradigms are likely to shape the future of LLM innovation, ensuring accessibility, competitive technical performance, and ethical deployment. |
2024-12-16 | 7 pages, 0 figures |
| SciFaultyQA: Benchmarking LLMs on Faulty Science Question Detection with a GAN-Inspired Approach to Synthetic Dataset Generation Consider the problem: ``If one man and one woman can produce one child in one year, how many children will be produced by one woman and three men in 0.5 years?” Current large language models (LLMs) such as GPT-4o, GPT-o1-preview, and Gemini Flash frequently answer “0.5,” which does not make sense. While these models sometimes acknowledge the unrealistic nature of the question, in many cases (8 out of 10 trials), they provide the nonsensical answer of “0.5 child.” Additionally, temporal variation has been observed: if an LLM answers correctly once (by recognizing the faulty nature of the question), subsequent responses are more likely to also reflect this understanding. However, this is inconsistent. These types of questions have motivated us to develop a dataset of science questions, SciFaultyQA, where the questions themselves are intentionally faulty. We observed that LLMs often proceed to answer these flawed questions without recognizing their inherent issues, producing results that are logically or scientifically invalid. By analyzing such patterns, we developed a novel method for generating synthetic datasets to evaluate and benchmark the performance of various LLMs in identifying these flawed questions. We have also developed novel approaches to reduce the errors. |
2024-12-16 | |
| Cost-Effective Label-free Node Classification with LLMs Graph neural networks (GNNs) have emerged as go-to models for node classification in graph data due to their powerful abilities in fusing graph structures and attributes. However, such models strongly rely on adequate high-quality labeled data for training, which are expensive to acquire in practice. With the advent of large language models (LLMs), a promising way is to leverage their superb zero-shot capabilities and massive knowledge for node labeling. Despite promising results reported, this methodology either demands considerable queries to LLMs, or suffers from compromised performance caused by noisy labels produced by LLMs. To remedy these issues, this work presents Cella, an active self-training framework that integrates LLMs into GNNs in a cost-effective manner. The design recipe of Cella is to iteratively identify small sets of “critical” samples using GNNs and extract informative pseudo-labels for them with both LLMs and GNNs as additional supervision signals to enhance model training. Particularly, Cella includes three major components: (i) an effective active node selection strategy for initial annotations; (ii) a judicious sample selection scheme to sift out the “critical” nodes based on label disharmonicity and entropy; and (iii) a label refinement module combining LLMs and GNNs with rewired topology. Our extensive experiments over five benchmark text-attributed graph datasets demonstrate that Cella significantly outperforms the state of the arts under the same query budget to LLMs in terms of label-free node classification. In particular, on the DBLP dataset with 14.3k nodes, Cella is able to achieve an 8.08% conspicuous improvement in accuracy over the state-of-the-art at a cost of less than one cent. |
2024-12-16 | 15 pages, 5 figures |
| When Backdoors Speak: Understanding LLM Backdoor Attacks Through Model-Generated Explanations Large Language Models (LLMs) are known to be vulnerable to backdoor attacks, where triggers embedded in poisoned samples can maliciously alter LLMs’ behaviors. In this paper, we move beyond attacking LLMs and instead examine backdoor attacks through the novel lens of natural language explanations. Specifically, we leverage LLMs’ generative capabilities to produce human-readable explanations for their decisions, enabling direct comparisons between explanations for clean and poisoned samples. Our results show that backdoored models produce coherent explanations for clean inputs but diverse and logically flawed explanations for poisoned data, a pattern consistent across classification and generation tasks for different backdoor attacks. Further analysis reveals key insights into the explanation generation process. At the token level, explanation tokens associated with poisoned samples only appear in the final few transformer layers. At the sentence level, attention dynamics indicate that poisoned inputs shift attention away from the original input context during explanation generation. These findings enhance our understanding of backdoor mechanisms in LLMs and present a promising framework for detecting vulnerabilities through explainability. |
2024-12-16 | |
| Uncovering LLM-Generated Code: A Zero-Shot Synthetic Code Detector via Code Rewriting Large Language Models (LLMs) have demonstrated remarkable proficiency in generating code. However, the misuse of LLM-generated (synthetic) code has raised concerns in both educational and industrial contexts, underscoring the urgent need for synthetic code detectors. Existing methods for detecting synthetic content are primarily designed for general text and struggle with code due to the unique grammatical structure of programming languages and the presence of numerous ‘‘low-entropy’’ tokens. Building on this, our work proposes a novel zero-shot synthetic code detector based on the similarity between the original code and its LLM-rewritten variants. Our method is based on the observation that differences between LLM-rewritten and original code tend to be smaller when the original code is synthetic. We utilize self-supervised contrastive learning to train a code similarity model and evaluate our approach on two synthetic code detection benchmarks. Our results demonstrate a significant improvement over existing SOTA synthetic content detectors, with AUROC scores increasing by 20.5% on the APPS benchmark and 29.1% on the MBPP benchmark. |
2024-12-16 | Accepted by AAAI 2025; previously submitted to EMNLP 2023 |
| Comprehensive Assessment of Jailbreak Attacks Against LLMs Jailbreak attacks aim to bypass the safeguards of LLMs. While researchers have studied different jailbreak attacks in depth, they have done so in isolation – either with unaligned experiment settings or comparing a limited range of methods. To fill this gap, we present the first large-scale measurement of various jailbreak attack methods. We collect 17 cutting-edge jailbreak methods, summarize their features, and establish a novel jailbreak attack taxonomy. Based on eight popular censored LLMs and 160 questions from 16 violation categories, we conduct a unified and impartial assessment of attack effectiveness as well as a comprehensive ablation study. Our extensive experimental results demonstrate that all the jailbreak attacks have a powerful effect on the LLMs. This indicates that all LLMs fail to cover all the violation categories, and they are susceptible to significant jailbreak risks, with even the well-aligned Llama3 facing a maximum attack success rate of 0.88. Additionally, we test jailbreak attacks under eight advanced external defenses and find none of the defenses could mitigate the jailbreak attacks entirely. Our study offers valuable insights for future research on jailbreak attacks and defenses and serves as a benchmark tool for researchers and practitioners to evaluate them effectively. |
2024-12-16 | 22 pages, 11 figures |
| Analyzing Images of Legal Documents: Toward Multi-Modal LLMs for Access to Justice Interacting with the legal system and the government requires the assembly and analysis of various pieces of information that can be spread across different (paper) documents, such as forms, certificates and contracts (e.g. leases). This information is required in order to understand one’s legal rights, as well as to fill out forms to file claims in court or obtain government benefits. However, finding the right information, locating the correct forms and filling them out can be challenging for laypeople. Large language models (LLMs) have emerged as a powerful technology that has the potential to address this gap, but still rely on the user to provide the correct information, which may be challenging and error-prone if the information is only available in complex paper documents. We present an investigation into utilizing multi-modal LLMs to analyze images of handwritten paper forms, in order to automatically extract relevant information in a structured format. Our initial results are promising, but reveal some limitations (e.g., when the image quality is low). Our work demonstrates the potential of integrating multi-modal LLMs to support laypeople and self-represented litigants in finding and assembling relevant information. |
2024-12-16 | Accepted at AI for Access to Justice Workshop at Jurix 2024, Brno, Czechia. Code and Data available at: https://github.com/hwestermann/AI4A2J_analyzing_images_of_legal_documents |
| GPTKB: Comprehensively Materializing Factual LLM Knowledge LLMs have majorly advanced NLP and AI, and next to their ability to perform a wide range of procedural tasks, a major success factor is their internalized factual knowledge. Since (Petroni et al., 2019), analyzing this knowledge has gained attention. However, most approaches investigate one question at a time via modest-sized pre-defined samples, introducing an availability bias (Tversky and Kahnemann, 1973) that prevents the discovery of knowledge (or beliefs) of LLMs beyond the experimenter’s predisposition. To address this challenge, we propose a novel methodology to comprehensively materializing an LLM’s factual knowledge through recursive querying and result consolidation. As a prototype, we employ GPT-4o-mini to construct GPTKB, a large-scale knowledge base (KB) comprising 105 million triples for over 2.9 million entities - achieved at 1% of the cost of previous KB projects. This work marks a milestone in two areas: For LLM research, for the first time, it provides constructive insights into the scope and structure of LLMs’ knowledge (or beliefs). For KB construction, it pioneers new pathways for the long-standing challenge of general-domain KB construction. GPTKB is accessible at https://gptkb.org. |
2024-12-16 | 13 pages, 4 tables, 10 figures |
| QUENCH: Measuring the gap between Indic and Non-Indic Contextual General Reasoning in LLMs The rise of large language models (LLMs) has created a need for advanced benchmarking systems beyond traditional setups. To this end, we introduce QUENCH, a novel text-based English Quizzing Benchmark manually curated and transcribed from YouTube quiz videos. QUENCH possesses masked entities and rationales for the LLMs to predict via generation. At the intersection of geographical context and common sense reasoning, QUENCH helps assess world knowledge and deduction capabilities of LLMs via a zero-shot, open-domain quizzing setup. We perform an extensive evaluation on 7 LLMs and 4 metrics, investigating the influence of model size, prompting style, geographical context, and gold-labeled rationale generation. The benchmarking concludes with an error analysis to which the LLMs are prone. |
2024-12-16 | 17 Pages, 6 Figures, 8 Tables, COLING 2025 |
| Harnessing Language for Coordination: A Framework and Benchmark for LLM-Driven Multi-Agent Control Large Language Models (LLMs) have demonstrated remarkable performance across various tasks. A promising but largely under-explored area is their potential to facilitate human coordination with many agents. Such capabilities would be useful in domains including disaster response, urban planning, and real-time strategy scenarios. In this work, we introduce (1) a real-time strategy game benchmark designed to evaluate these abilities and (2) a novel framework we term HIVE. HIVE empowers a single human to coordinate swarms of up to 2,000 agents using natural language dialog with an LLM. We present promising results on this multi-agent benchmark, with our hybrid approach solving tasks such as coordinating agent movements, exploiting unit weaknesses, leveraging human annotations, and understanding terrain and strategic points. However, our findings also highlight critical limitations of current models, including difficulties in processing spatial visual information and challenges in formulating long-term strategic plans. This work sheds light on the potential and limitations of LLMs in human-swarm coordination, paving the way for future research in this area. The HIVE project page, which includes videos of the system in action, can be found here: hive.syrkis.com. |
2024-12-16 | |
| Personalized LLM for Generating Customized Responses to the Same Query from Different Users Existing work on large language model (LLM) personalization assigned different responding roles to LLM, but overlooked the diversity of questioners. In this work, we propose a new form of questioner-aware LLM personalization, generating different responses even for the same query from different questioners. We design a dual-tower model architecture with a cross-questioner general encoder and a questioner-specific encoder. We further apply contrastive learning with multi-view augmentation, pulling close the dialogue representations of the same questioner, while pulling apart those of different questioners. To mitigate the impact of question diversity on questioner-contrastive learning, we cluster the dialogues based on question similarity and restrict the scope of contrastive learning within each cluster. We also build a multi-questioner dataset from English and Chinese scripts and WeChat records, called MQDialog, containing 173 questioners and 12 responders. Extensive evaluation with different metrics shows a significant improvement in the quality of personalized response generation. |
2024-12-16 | 9 pages |
| LLMs Can Simulate Standardized Patients via Agent Coevolution Training medical personnel using standardized patients (SPs) remains a complex challenge, requiring extensive domain expertise and role-specific practice. Most research on Large Language Model (LLM)-based simulated patients focuses on improving data retrieval accuracy or adjusting prompts through human feedback. However, this focus has overlooked the critical need for patient agents to learn a standardized presentation pattern that transforms data into human-like patient responses through unsupervised simulations. To address this gap, we propose EvoPatient, a novel simulated patient framework in which a patient agent and doctor agents simulate the diagnostic process through multi-turn dialogues, simultaneously gathering experience to improve the quality of both questions and answers, ultimately enabling human doctor training. Extensive experiments on various cases demonstrate that, by providing only overall SP requirements, our framework improves over existing reasoning methods by more than 10% in requirement alignment and better human preference, while achieving an optimal balance of resource consumption after evolving over 200 cases for 10 hours, with excellent generalizability. The code will be available at https://github.com/ZJUMAI/EvoPatient. |
2024-12-16 | Work in Progress |
| CoinMath: Harnessing the Power of Coding Instruction for Math LLMs Large Language Models (LLMs) have shown strong performance in solving mathematical problems, with code-based solutions proving particularly effective. However, the best practice to leverage coding instruction data to enhance mathematical reasoning remains underexplored. This study investigates three key questions: (1) How do different coding styles of mathematical code-based rationales impact LLMs’ learning performance? (2) Can general-domain coding instructions improve performance? (3) How does integrating textual rationales with code-based ones during training enhance mathematical reasoning abilities? Our findings reveal that code-based rationales with concise comments, descriptive naming, and hardcoded solutions are beneficial, while improvements from general-domain coding instructions and textual rationales are relatively minor. Based on these insights, we propose CoinMath, a learning strategy designed to enhance mathematical reasoning by diversifying the coding styles of code-based rationales. CoinMath generates a variety of code-based rationales incorporating concise comments, descriptive naming conventions, and hardcoded solutions. Experimental results demonstrate that CoinMath significantly outperforms its baseline model, MAmmoTH, one of the SOTA math LLMs. |
2024-12-16 | |
| Multimodal LLM for Intelligent Transportation Systems In the evolving landscape of transportation systems, integrating Large Language Models (LLMs) offers a promising frontier for advancing intelligent decision-making across various applications. This paper introduces a novel 3-dimensional framework that encapsulates the intersection of applications, machine learning methodologies, and hardware devices, particularly emphasizing the role of LLMs. Instead of using multiple machine learning algorithms, our framework uses a single, data-centric LLM architecture that can analyze time series, images, and videos. We explore how LLMs can enhance data interpretation and decision-making in transportation. We apply this LLM framework to different sensor datasets, including time-series data and visual data from sources like Oxford Radar RobotCar, D-Behavior (D-Set), nuScenes by Motional, and Comma2k19. The goal is to streamline data processing workflows, reduce the complexity of deploying multiple models, and make intelligent transportation systems more efficient and accurate. The study was conducted using state-of-the-art hardware, leveraging the computational power of AMD RTX 3060 GPUs and Intel i9-12900 processors. The experimental results demonstrate that our framework achieves an average accuracy of 91.33\% across these datasets, with the highest accuracy observed in time-series data (92.7\%), showcasing the model’s proficiency in handling sequential information essential for tasks such as motion planning and predictive maintenance. Through our exploration, we demonstrate the versatility and efficacy of LLMs in handling multimodal data within the transportation sector, ultimately providing insights into their application in real-world scenarios. Our findings align with the broader conference themes, highlighting the transformative potential of LLMs in advancing transportation technologies. |
2024-12-16 | Accepted at IEEE Symposium Series on Computational Intelligence (SSCI) 2025 |
| Prompto: An open source library for asynchronous querying of LLM endpoints Recent surge in Large Language Model (LLM) availability has opened exciting avenues for research. However, efficiently interacting with these models presents a significant hurdle since LLMs often reside on proprietary or self-hosted API endpoints, each requiring custom code for interaction. Conducting comparative studies between different models can therefore be time-consuming and necessitate significant engineering effort, hindering research efficiency and reproducibility. To address these challenges, we present prompto, an open source Python library which facilitates asynchronous querying of LLM endpoints enabling researchers to interact with multiple LLMs concurrently, while maximising efficiency and utilising individual rate limits. Our library empowers researchers and developers to interact with LLMs more effectively and allowing faster experimentation, data generation and evaluation. prompto is released with an introductory video (https://youtu.be/lWN9hXBOLyQ) under MIT License and is available via GitHub (https://github.com/alan-turing-institute/prompto). |
2024-12-16 | |
| LLM-DaaS: LLM-driven Drone-as-a-Service Operations from Text User Requests We propose LLM-DaaS, a novel Drone-as-a-Service (DaaS) framework that leverages Large Language Models (LLMs) to transform free-text user requests into structured, actionable DaaS operation tasks. Our approach addresses the key challenge of interpreting and structuring natural language input to automate drone service operations under uncertain conditions. The system is composed of three main components: free-text request processing, structured request generation, and dynamic DaaS selection and composition. First, we fine-tune different LLM models such as Phi-3.5, LLaMA-3.2 7b and Gemma 2b on a dataset of text user requests mapped to structured DaaS requests. Users interact with our model in a free conversational style, discussing package delivery requests, while the fine-tuned LLM extracts DaaS metadata such as delivery time, source and destination locations, and package weight. The DaaS service selection model is designed to select the best available drone capable of delivering the requested package from the delivery point to the nearest optimal destination. Additionally, the DaaS composition model composes a service from a set of the best available drones to deliver the package from the source to the final destination. Second, the system integrates real-time weather data to optimize drone route planning and scheduling, ensuring safe and efficient operations. Simulations demonstrate the system’s ability to significantly improve task accuracy, operational efficiency, and establish LLM-DaaS as a robust solution for DaaS operations in uncertain environments. |
2024-12-16 | |
| How Reliable are LLMs as Knowledge Bases? Re-thinking Facutality and Consistency Large Language Models (LLMs) are increasingly explored as knowledge bases (KBs), yet current evaluation methods focus too narrowly on knowledge retention, overlooking other crucial criteria for reliable performance. In this work, we rethink the requirements for evaluating reliable LLM-as-KB usage and highlight two essential factors: factuality, ensuring accurate responses to seen and unseen knowledge, and consistency, maintaining stable answers to questions about the same knowledge. We introduce UnseenQA, a dataset designed to assess LLM performance on unseen knowledge, and propose new criteria and metrics to quantify factuality and consistency, leading to a final reliability score. Our experiments on 26 LLMs reveal several challenges regarding their use as KBs, underscoring the need for more principled and comprehensive evaluation. |
2024-12-16 | |
| Private Yet Social: How LLM Chatbots Support and Challenge Eating Disorder Recovery Eating disorders (ED) are complex mental health conditions that require long-term management and support. Recent advancements in large language model (LLM)-based chatbots offer the potential to assist individuals in receiving immediate support. Yet, concerns remain about their reliability and safety in sensitive contexts such as ED. We explore the opportunities and potential harms of using LLM-based chatbots for ED recovery. We observe the interactions between 26 participants with ED and an LLM-based chatbot, WellnessBot, designed to support ED recovery, over 10 days. We discovered that our participants have felt empowered in recovery by discussing ED-related stories with the chatbot, which served as a personal yet social avenue. However, we also identified harmful chatbot responses, especially concerning individuals with ED, that went unnoticed partly due to participants’ unquestioning trust in the chatbot’s reliability. Based on these findings, we provide design implications for safe and effective LLM-based interventions in ED management. |
2024-12-16 | |
| Fool Me, Fool Me: User Attitudes Toward LLM Falsehoods While Large Language Models (LLMs) have become central tools in various fields, they often provide inaccurate or false information. This study examines user preferences regarding falsehood responses from LLMs. Specifically, we evaluate preferences for LLM responses where false statements are explicitly marked versus unmarked responses and preferences for confident falsehoods compared to LLM disclaimers acknowledging a lack of knowledge. Additionally, we investigate how requiring users to assess the truthfulness of statements influences these preferences. Surprisingly, 61\% of users prefer unmarked falsehood responses over marked ones, and 69\% prefer confident falsehoods over LLMs admitting lack of knowledge. In all our experiments, a total of 300 users participated, contributing valuable data to our analysis and conclusions. When users are required to evaluate the truthfulness of statements, preferences for unmarked and falsehood responses decrease slightly but remain high. These findings suggest that user preferences, which influence LLM training via feedback mechanisms, may inadvertently encourage the generation of falsehoods. Future research should address the ethical and practical implications of aligning LLM behavior with such preferences. |
2024-12-16 | 11 pages, 5 figures, 5 tables |
| EvoLlama: Enhancing LLMs’ Understanding of Proteins via Multimodal Structure and Sequence Representations Current Large Language Models (LLMs) for understanding proteins primarily treats amino acid sequences as a text modality. Meanwhile, Protein Language Models (PLMs), such as ESM-2, have learned massive sequential evolutionary knowledge from the universe of natural protein sequences. Furthermore, structure-based encoders like ProteinMPNN learn the structural information of proteins through Graph Neural Networks. However, whether the incorporation of protein encoders can enhance the protein understanding of LLMs has not been explored. To bridge this gap, we propose EvoLlama, a multimodal framework that connects a structure-based encoder, a sequence-based protein encoder and an LLM for protein understanding. EvoLlama consists of a ProteinMPNN structure encoder, an ESM-2 protein sequence encoder, a multimodal projector to align protein and text representations and a Llama-3 text decoder. To train EvoLlama, we fine-tune it on protein-oriented instructions and protein property prediction datasets verbalized via natural language instruction templates. Our experiments show that EvoLlama’s protein understanding capabilities have been significantly enhanced, outperforming other fine-tuned protein-oriented LLMs in zero-shot settings by an average of 1%-8% and surpassing the state-of-the-art baseline with supervised fine-tuning by an average of 6%. On protein property prediction datasets, our approach achieves promising results that are competitive with state-of-the-art task-specific baselines. We will release our code in a future version. |
2024-12-16 | |
| Intention Analysis Makes LLMs A Good Jailbreak Defender Aligning large language models (LLMs) with human values, particularly when facing complex and stealthy jailbreak attacks, presents a formidable challenge. Unfortunately, existing methods often overlook this intrinsic nature of jailbreaks, which limits their effectiveness in such complex scenarios. In this study, we present a simple yet highly effective defense strategy, i.e., Intention Analysis ($\mathbb{IA}$). $\mathbb{IA}$ works by triggering LLMs’ inherent self-correct and improve ability through a two-stage process: 1) analyzing the essential intention of the user input, and 2) providing final policy-aligned responses based on the first round conversation. Notably, $\mathbb{IA}$ is an inference-only method, thus could enhance LLM safety without compromising their helpfulness. Extensive experiments on varying jailbreak benchmarks across a wide range of LLMs show that $\mathbb{IA}$ could consistently and significantly reduce the harmfulness in responses (averagely -48.2% attack success rate). Encouragingly, with our $\mathbb{IA}$, Vicuna-7B even outperforms GPT-3.5 regarding attack success rate. We empirically demonstrate that, to some extent, $\mathbb{IA}$ is robust to errors in generated intentions. Further analyses reveal the underlying principle of $\mathbb{IA}$: suppressing LLM’s tendency to follow jailbreak prompts, thereby enhancing safety. |
2024-12-16 | COLING 2025 |
| Token Prepending: A Training-Free Approach for Eliciting Better Sentence Embeddings from LLMs Extracting sentence embeddings from large language models (LLMs) is a promising direction, as LLMs have demonstrated stronger semantic understanding capabilities. Previous studies typically focus on prompt engineering to elicit sentence embeddings from LLMs by prompting the model to encode sentence information into the embedding of the last token. However, LLMs are mostly decoder-only models with causal attention and the earlier tokens in the sentence cannot attend to the latter tokens, resulting in biased encoding of sentence information and cascading effects on the final decoded token. To this end, we propose a novel Token Prepending (TP) technique that prepends each layer’s decoded sentence embedding to the beginning of the sentence in the next layer’s input, allowing earlier tokens to attend to the complete sentence information under the causal attention mechanism. The proposed TP technique is a plug-and-play and training-free technique, which means it can be seamlessly integrated with various prompt-based sentence embedding methods and autoregressive LLMs. Extensive experiments on various Semantic Textual Similarity (STS) tasks and downstream classification tasks demonstrate that our proposed TP technique can significantly improve the performance of existing prompt-based sentence embedding methods across different LLMs, while incurring negligible additional inference cost. |
2024-12-16 | 14 pages, 5 figures |
| RankAdaptor: Hierarchical Rank Allocation for Efficient Fine-Tuning Pruned LLMs via Performance Model The efficient compression of large language models (LLMs) has become increasingly popular. However, recovering the performance of compressed LLMs remains a major challenge. The current practice in LLM compression entails the implementation of structural pruning, complemented by a recovery phase that leverages the Low-Rank Adaptation (LoRA) algorithm. Structural pruning’s uneven modification of model architecture, coupled with standard LoRA’s fixed configuration allocation across layers in an online pipeline, leads to suboptimal performance in various downstream tasks for pruned models. To address this challenge, we introduce RankAdaptor, a hierarchical rank allocation method that enables efficient fine-tuning of pruned LLMs according to layerwise specific recovery requirements. We employ a performance model that conducts offline meta-learning and online incremental learning to explore optimal rank values for each layer. Comprehensive experiments on popular benchmarks show that RankAdaptor consistently outperforms state-of-the-art methods across a variety of pruning settings and LLM architectures, with improvements ranging from 0.7\% to 5.5\%. |
2024-12-16 | |
| Specifications: The missing link to making the development of LLM systems an engineering discipline Despite the significant strides made by generative AI in just a few short years, its future progress is constrained by the challenge of building modular and robust systems. This capability has been a cornerstone of past technological revolutions, which relied on combining components to create increasingly sophisticated and reliable systems. Cars, airplanes, computers, and software consist of components-such as engines, wheels, CPUs, and libraries-that can be assembled, debugged, and replaced. A key tool for building such reliable and modular systems is specification: the precise description of the expected behavior, inputs, and outputs of each component. However, the generality of LLMs and the inherent ambiguity of natural language make defining specifications for LLM-based components (e.g., agents) both a challenging and urgent problem. In this paper, we discuss the progress the field has made so far-through advances like structured outputs, process supervision, and test-time compute-and outline several future directions for research to enable the development of modular and reliable LLM-based systems through improved specifications. |
2024-12-16 | |
| Let your LLM generate a few tokens and you will reduce the need for retrieval In this paper, we investigate how efficiently large language models (LLM) can be trained to check whether an answer is already stored in their parametric memory. We distill an LLM-as-a-judge to compute the IK (I Know) score. We found that this method is particularly beneficial in the context of retrieval-assisted augmented generation (RAG), with a respectable accuracy of 80%. It enables a significant reduction (more than 50%) in the number of search and reranking steps required for certain data sets. We have also introduced the IK score, which serves as a useful tool for characterising datasets by facilitating the classification task. Interestingly, through the inclusion of response tokens as input, our results suggest that only about 20,000 training samples are required to achieve good performance. The central element of this work is the use of a teacher model - the LLM as a judge - to generate training data. We also assess the robustness of the IK classifier by evaluating it with various types of teachers, including both string-based methods and LLMs, with the latter providing better results. |
2024-12-16 | |
| MQM-APE: Toward High-Quality Error Annotation Predictors with Automatic Post-Editing in LLM Translation Evaluators Large Language Models (LLMs) have shown significant potential as judges for Machine Translation (MT) quality assessment, providing both scores and fine-grained feedback. Although approaches such as GEMBA-MQM have shown state-of-the-art performance on reference-free evaluation, the predicted errors do not align well with those annotated by human, limiting their interpretability as feedback signals. To enhance the quality of error annotations predicted by LLM evaluators, we introduce a universal and training-free framework, $\textbf{MQM-APE}$, based on the idea of filtering out non-impactful errors by Automatically Post-Editing (APE) the original translation based on each error, leaving only those errors that contribute to quality improvement. Specifically, we prompt the LLM to act as 1) $\textit{evaluator}$ to provide error annotations, 2) $\textit{post-editor}$ to determine whether errors impact quality improvement and 3) $\textit{pairwise quality verifier}$ as the error filter. Experiments show that our approach consistently improves both the reliability and quality of error spans against GEMBA-MQM, across eight LLMs in both high- and low-resource languages. Orthogonal to trained approaches, MQM-APE complements translation-specific evaluators such as Tower, highlighting its broad applicability. Further analysis confirms the effectiveness of each module and offers valuable insights into evaluator design and LLMs selection. |
2024-12-16 | COLING 2025 |
| OpenFactCheck: Building, Benchmarking Customized Fact-Checking Systems and Evaluating the Factuality of Claims and LLMs The increased use of large language models (LLMs) across a variety of real-world applications calls for mechanisms to verify the factual accuracy of their outputs. Difficulties lie in assessing the factuality of free-form responses in open domains. Also, different papers use disparate evaluation benchmarks and measurements, which renders them hard to compare and hampers future progress. To mitigate these issues, we propose OpenFactCheck, a unified framework for building customized automatic fact-checking systems, benchmarking their accuracy, evaluating factuality of LLMs, and verifying claims in a document. OpenFactCheck consists of three modules: (i) CUSTCHECKER allows users to easily customize an automatic fact-checker and verify the factual correctness of documents and claims, (ii) LLMEVAL, a unified evaluation framework assesses LLM’s factuality ability from various perspectives fairly, and (iii) CHECKEREVAL is an extensible solution for gauging the reliability of automatic fact-checkers’ verification results using human-annotated datasets. Data and code are publicly available at https://github.com/yuxiaw/openfactcheck. |
2024-12-16 | 22 pages, 8 tables, 11 figures |
| Glimpse: Enabling White-Box Methods to Use Proprietary Models for Zero-Shot LLM-Generated Text Detection Advanced large language models (LLMs) can generate text almost indistinguishable from human-written text, highlighting the importance of LLM-generated text detection. However, current zero-shot techniques face challenges as white-box methods are restricted to use weaker open-source LLMs, and black-box methods are limited by partial observation from stronger proprietary LLMs. It seems impossible to enable white-box methods to use proprietary models because API-level access to the models neither provides full predictive distributions nor inner embeddings. To traverse the divide, we propose Glimpse, a probability distribution estimation approach, predicting the full distributions from partial observations. Despite the simplicity of Glimpse, we successfully extend white-box methods like Entropy, Rank, Log-Rank, and Fast-DetectGPT to latest proprietary models. Experiments show that Glimpse with Fast-DetectGPT and GPT-3.5 achieves an average AUROC of about 0.95 in five latest source models, improving the score by 51% relative to the remaining space of the open source baseline (Table 1). It demonstrates that the latest LLMs can effectively detect their own outputs, suggesting that advanced LLMs may be the best shield against themselves. |
2024-12-16 | 10 pages, 9 figures, 10 tables |
| Embodied CoT Distillation From LLM To Off-the-shelf Agents We address the challenge of utilizing large language models (LLMs) for complex embodied tasks, in the environment where decision-making systems operate timely on capacity-limited, off-the-shelf devices. We present DeDer, a framework for decomposing and distilling the embodied reasoning capabilities from LLMs to efficient, small language model (sLM)-based policies. In DeDer, the decision-making process of LLM-based strategies is restructured into a hierarchy with a reasoning-policy and planning-policy. The reasoning-policy is distilled from the data that is generated through the embodied in-context learning and self-verification of an LLM, so it can produce effective rationales. The planning-policy, guided by the rationales, can render optimized plans efficiently. In turn, DeDer allows for adopting sLMs for both policies, deployed on off-the-shelf devices. Furthermore, to enhance the quality of intermediate rationales, specific to embodied tasks, we devise the embodied knowledge graph, and to generate multiple rationales timely through a single inference, we also use the contrastively prompted attention model. Our experiments with the ALFRED benchmark demonstrate that DeDer surpasses leading language planning and distillation approaches, indicating the applicability and efficiency of sLM-based embodied policies derived through DeDer. |
2024-12-16 | Accepted at ICML 2024 |
| Game Development as Human-LLM Interaction Game development is a highly specialized task that relies on a complex game engine powered by complex programming languages, preventing many gaming enthusiasts from handling it. This paper introduces the Chat Game Engine (ChatGE) powered by LLM, which allows everyone to develop a custom game using natural language through Human-LLM interaction. To enable an LLM to function as a ChatGE, we instruct it to perform the following processes in each turn: (1) $P_{script}$: configure the game script segment based on the user’s input; (2) $P_{code}$: generate the corresponding code snippet based on the game script segment; (3) $P_{utter}$: interact with the user, including guidance and feedback. We propose a data synthesis pipeline based on LLM to generate game script-code pairs and interactions from a few manually crafted seed data. We propose a three-stage progressive training strategy to transfer the dialogue-based LLM to our ChatGE smoothly. We construct a ChatGE for poker games as a case study and comprehensively evaluate it from two perspectives: interaction quality and code correctness. |
2024-12-16 | |
| Are LLMs Rigorous Logical Reasoner? Empowering Natural Language Proof Generation with Contrastive Stepwise Decoding Logical reasoning remains a pivotal component within the realm of artificial intelligence. The recent evolution of large language models (LLMs) has marked significant progress in this domain. The adoption of strategies like chain-of-thought (CoT) has enhanced the performance of LLMs across diverse reasoning tasks. Nonetheless, logical reasoning that involves proof planning, specifically those that necessitate the validation of explanation accuracy, continues to present stumbling blocks. In this study, we first evaluate the efficacy of LLMs with advanced CoT strategies concerning such tasks. Our analysis reveals that LLMs still struggle to navigate complex reasoning chains, which demand the meticulous linkage of premises to derive a cogent conclusion. To address this issue, we finetune a smaller-scale language model, equipping it to decompose proof objectives into more manageable subgoals. We also introduce contrastive decoding to stepwise proof generation, making use of negative reasoning paths to strengthen the model’s capacity for logical deduction. Experiments on EntailmentBank underscore the success of our method in augmenting the proof planning abilities of language models. |
2024-12-16 | The paper is currently undergoing extensive revisions and improvements |
| Look Ahead Text Understanding and LLM Stitching This paper proposes a look ahead text understanding problem with look ahead section identification (LASI) as an example. This problem may appear in generative AI as well as human interactions, where we want to understand the direction of a developing text or conversation. We tackle the problem using transformer-based LLMs. We show that LASI is more challenging than classic section identification (SI). We argue that both bidirectional contextual information (e.g., BERT) and unidirectional predictive ability (e.g., GPT) will benefit the task. We propose two approaches to stitch together BERT and GPT. Experiments show that our approach outperforms the established models, especially when there is noise in the text (which is often the case for developing text in generative AI). Our paper sheds light on other look ahead text understanding tasks that are important to social media, such as look ahead sentiment classification, and points out the opportunities to leverage pre-trained LLMs through stitching. |
2024-12-16 | 9 pages, 6 figures, 4 tables, published in Vol. 18 (2024): Proceedings of the Eighteenth International AAAI Conference on Web and Social Media |
| InverseCoder: Self-improving Instruction-Tuned Code LLMs with Inverse-Instruct Recent advancements in open-source code large language models (LLMs) have been driven by fine-tuning on the data generated from powerful closed-source LLMs, which are expensive to obtain. This paper explores whether it is possible to use a fine-tuned open-source model to generate additional data to augment its instruction-tuning dataset. We make two observations: (1) A code snippet can serve as the response to different instructions. (2) Instruction-tuned code LLMs perform better at translating code into instructions than the reverse. Based on these observations, we propose Inverse-Instruct, a data augmentation technique that uses a fine-tuned LLM to generate additional instructions of code responses from its own training dataset. The additional instruction-response pairs are added to the original dataset, and a stronger code LLM can be obtained by fine-tuning on the augmented dataset. We empirically validate Inverse-Instruct on a range of open-source code models (e.g. CodeLlama-Python and DeepSeek-Coder) and benchmarks (e.g., HumanEval(+), MBPP(+), DS-1000 and MultiPL-E), showing it consistently improves the base models. |
2024-12-16 | Accepted for publication at AAAI 2025. Extended version with full appendix, 18 pages |
| Structured Extraction of Real World Medical Knowledge using LLMs for Summarization and Search Creation and curation of knowledge graphs can accelerate disease discovery and analysis in real-world data. While disease ontologies aid in biological data annotation, codified categories (SNOMED-CT, ICD10, CPT) may not capture patient condition nuances or rare diseases. Multiple disease definitions across data sources complicate ontology mapping and disease clustering. We propose creating patient knowledge graphs using large language model extraction techniques, allowing data extraction via natural language rather than rigid ontological hierarchies. Our method maps to existing ontologies (MeSH, SNOMED-CT, RxNORM, HPO) to ground extracted entities. Using a large ambulatory care EHR database with 33.6M patients, we demonstrate our method through the patient search for Dravet syndrome, which received ICD10 recognition in October 2020. We describe our construction of patient-specific knowledge graphs and symptom-based patient searches. Using confirmed Dravet syndrome ICD10 codes as ground truth, we employ LLM-based entity extraction to characterize patients in grounded ontologies. We then apply this method to identify Beta-propeller protein-associated neurodegeneration (BPAN) patients, demonstrating real-world discovery where no ground truth exists. |
2024-12-16 | 10 pages, 3 figures, Work published in 4th Workshop on Knowledge Graphs and Big Data (In Conjunction with IEEE Big Data 2024) |
| How Can LLMs and Knowledge Graphs Contribute to Robot Safety? A Few-Shot Learning Approach Large Language Models (LLMs) are transforming the robotics domain by enabling robots to comprehend and execute natural language instructions. The cornerstone benefits of LLM include processing textual data from technical manuals, instructions, academic papers, and user queries based on the knowledge provided. However, deploying LLM-generated code in robotic systems without safety verification poses significant risks. This paper outlines a safety layer that verifies the code generated by ChatGPT before executing it to control a drone in a simulated environment. The safety layer consists of a fine-tuned GPT-4o model using Few-Shot learning, supported by knowledge graph prompting (KGP). Our approach improves the safety and compliance of robotic actions, ensuring that they adhere to the regulations of drone operations. |
2024-12-16 | |
| Zero-Shot Prompting Approaches for LLM-based Graphical User Interface Generation Graphical user interface (GUI) prototyping represents an essential activity in the development of interactive systems, which are omnipresent today. GUI prototypes facilitate elicitation of requirements and help to test, evaluate, and validate ideas with users and the development team. However, creating GUI prototypes is a time-consuming process and often requires extensive resources. While existing research for automatic GUI generation focused largely on resource-intensive training and fine-tuning of LLMs, mainly for low-fidelity GUIs, we investigate the potential and effectiveness of Zero-Shot (ZS) prompting for high-fidelity GUI generation. We propose a Retrieval-Augmented GUI Generation (RAGG) approach, integrated with an LLM-based GUI retrieval re-ranking and filtering mechanism based on a large-scale GUI repository. In addition, we adapt Prompt Decomposition (PDGG) and Self-Critique (SCGG) for GUI generation. To evaluate the effectiveness of the proposed ZS prompting approaches for GUI generation, we extensively evaluated the accuracy and subjective satisfaction of the generated GUI prototypes. Our evaluation, which encompasses over 3,000 GUI annotations from over 100 crowd-workers with UI/UX experience, shows that SCGG, in contrast to PDGG and RAGG, can lead to more effective GUI generation, and provides valuable insights into the defects that are produced by the LLMs in the generated GUI prototypes. |
2024-12-15 | |
| Aligning LLMs with Individual Preferences via Interaction As large language models (LLMs) demonstrate increasingly advanced capabilities, aligning their behaviors with human values and preferences becomes crucial for their wide adoption. While previous research focuses on general alignment to principles such as helpfulness, harmlessness, and honesty, the need to account for individual and diverse preferences has been largely overlooked, potentially undermining customized human experiences. To address this gap, we train LLMs that can ‘‘interact to align’’, essentially cultivating the meta-skill of LLMs to implicitly infer the unspoken personalized preferences of the current user through multi-turn conversations, and then dynamically align their following behaviors and responses to these inferred preferences. Our approach involves establishing a diverse pool of 3,310 distinct user personas by initially creating seed examples, which are then expanded through iterative self-generation and filtering. Guided by distinct user personas, we leverage multi-LLM collaboration to develop a multi-turn preference dataset containing 3K+ multi-turn conversations in tree structures. Finally, we apply supervised fine-tuning and reinforcement learning to enhance LLMs using this dataset. For evaluation, we establish the ALOE (ALign With CustOmized PrEferences) benchmark, consisting of 100 carefully selected examples and well-designed metrics to measure the customized alignment performance during conversations. Experimental results demonstrate the effectiveness of our method in enabling dynamic, personalized alignment via interaction. |
2024-12-15 | Accepted to COLING 2025. The code and dataset are made public at https://github.com/ShujinWu-0814/ALOE |
| LLM Based Input Space Partitioning Testing for Library APIs Automated library APIs testing is difficult as it requires exploring a vast space of parameter inputs that may involve objects with complex data types. Existing search based approaches, with limited knowledge of relations between object states and program branches, often suffer from the low efficiency issue, i.e., tending to generate invalid inputs. Symbolic execution based approaches can effectively identify such relations, but fail to scale to large programs. In this work, we present an LLM-based input space partitioning testing approach, LISP, for library APIs. The approach leverages LLMs to understand the code of a library API under test and perform input space partitioning based on its understanding and rich common knowledge. Specifically, we provide the signature and code of the API under test to LLMs, with the expectation of obtaining a text description of each input space partition of theAPI under test. Then, we generate inputs through employing the generated text description to sample inputs from each partition, ultimately resulting in test suites that systematically explore the program behavior of the API. We evaluate LISP on more than 2,205 library API methods taken from 10 popular open-source Java libraries (e.g.,apache/commonslang with 2.6k stars, guava with 48.8k stars on GitHub). Our experiment results show that LISP is effective in library API testing. It significantly outperforms state-of-the-art tool EvoSuite in terms of edge coverage. On average, LISP achieves 67.82% branch coverage, surpassing EvoSuite by 1.21 times. In total, LISP triggers 404 exceptions or errors in the experiments, and discovers 13 previously unknown vulnerabilities during evaluation, which have been assigned CVE IDs. |
2024-12-15 | 11 pages, 10 figures. Proceedings of the 47th IEEE/ACM International Conference on Software Engineering (ICSE 2025) |
| CATER: Leveraging LLM to Pioneer a Multidimensional, Reference-Independent Paradigm in Translation Quality Evaluation This paper introduces the Comprehensive AI-assisted Translation Edit Ratio (CATER), a novel and fully prompt-driven framework for evaluating machine translation (MT) quality. Leveraging large language models (LLMs) via a carefully designed prompt-based protocol, CATER expands beyond traditional reference-bound metrics, offering a multidimensional, reference-independent evaluation that addresses linguistic accuracy, semantic fidelity, contextual coherence, stylistic appropriateness, and information completeness. CATER’s unique advantage lies in its immediate implementability: by providing the source and target texts along with a standardized prompt, an LLM can rapidly identify errors, quantify edit effort, and produce category-level and overall scores. This approach eliminates the need for pre-computed references or domain-specific resources, enabling instant adaptation to diverse languages, genres, and user priorities through adjustable weights and prompt modifications. CATER’s LLM-enabled strategy supports more nuanced assessments, capturing phenomena such as subtle omissions, hallucinations, and discourse-level shifts that increasingly challenge contemporary MT systems. By uniting the conceptual rigor of frameworks like MQM and DQF with the scalability and flexibility of LLM-based evaluation, CATER emerges as a valuable tool for researchers, developers, and professional translators worldwide. The framework and example prompts are openly available, encouraging community-driven refinement and further empirical validation. |
2024-12-15 | 17pages,1sample prompt |
| Propulsion: Steering LLM with Tiny Fine-Tuning The rapid advancements in Large Language Models (LLMs) have revolutionized natural language processing (NLP) and related fields. However, fine-tuning these models for specific tasks remains computationally expensive and risks degrading pre-learned features. To address these challenges, we propose Propulsion, a novel parameter efficient fine-tuning (PEFT) method designed to optimize task-specific performance while drastically reducing computational overhead. Inspired by the concept of controlled adjustments in physical motion, Propulsion selectively re-scales specific dimensions of a pre-trained model, guiding output predictions toward task objectives without modifying the model’s parameters. By introducing lightweight, trainable Propulsion parameters at the pre-trained layer, we minimize the number of parameters updated during fine-tuning, preventing overfitting or overwriting of existing knowledge. Our theoretical analysis, supported by Neural Tangent Kernel (NTK) theory, shows that Propulsion approximates the performance of full fine-tuning with far fewer trainable parameters. Empirically, Propulsion reduces the parameter count from 355.3 million to just 0.086 million, achieving over a 10x reduction compared to standard approaches like LoRA while maintaining competitive performance across benchmarks. |
2024-12-15 | 26 pages, 11 figures accepted paper |
| Fast On-device LLM Inference with NPUs On-device inference for Large Language Models (LLMs), driven by increasing privacy concerns and advancements of mobile-sized models, has gained significant interest. However, even mobile-sized LLMs (e.g., Gemma-2B) encounter unacceptably high inference latency, often bottlenecked by the prefill stage in tasks like screen UI understanding. We present llm.npu, the first LLM inference system utilizing on-device Neural Processing Unit (NPU) offloading to reduce prefill latency. llm.npu enhances NPU offloading efficiency by re-constructing the prompt and model in three levels: (1) At prompt level, it divides variable-length prompts into multiple fixed-sized chunks while maintaining data dependencies; (2) At tensor level, it identifies and extracts significant outliers to run on the CPU/GPU in parallel with minimal overhead; (3) At block level, it schedules Transformer blocks in an out-of-order manner to the CPU/GPU and NPU based on their hardware affinity and sensitivity to accuracy. Compared to competitive baselines, llm.npu achieves 22.4x faster prefill speed and 30.7$\times$ energy savings on average, and up to 32.8x speedup in an end-to-end real-world application. For the first time, llm.npu achieves more than 1,000 tokens/sec prefilling for a billion-sized model. |
2024-12-15 | |
| Towards Trustworthy LLMs for Code: A Data-Centric Synergistic Auditing Framework LLM-powered coding and development assistants have become prevalent to programmers’ workflows. However, concerns about the trustworthiness of LLMs for code persist despite their widespread use. Much of the existing research focused on either training or evaluation, raising questions about whether stakeholders in training and evaluation align in their understanding of model trustworthiness and whether they can move toward a unified direction. In this paper, we propose a vision for a unified trustworthiness auditing framework, DataTrust, which adopts a data-centric approach that synergistically emphasizes both training and evaluation data and their correlations. DataTrust aims to connect model trustworthiness indicators in evaluation with data quality indicators in training. It autonomously inspects training data and evaluates model trustworthiness using synthesized data, attributing potential causes from specific evaluation data to corresponding training data and refining indicator connections. Additionally, a trustworthiness arena powered by DataTrust will engage crowdsourced input and deliver quantitative outcomes. We outline the benefits that various stakeholders can gain from DataTrust and discuss the challenges and opportunities it presents. |
2024-12-15 | Short Vision Paper, Accepted by ICSE’25-NIER |
| Teola: Towards End-to-End Optimization of LLM-based Applications Large language model (LLM)-based applications consist of both LLM and non-LLM components, each contributing to the end-to-end latency. Despite great efforts to optimize LLM inference, end-to-end workflow optimization has been overlooked. Existing frameworks employ coarse-grained orchestration with task modules, which confines optimizations to within each module and yields suboptimal scheduling decisions. We propose fine-grained end-to-end orchestration, which utilizes task primitives as the basic units and represents each query’s workflow as a primitive-level dataflow graph. This explicitly exposes a much larger design space, enables optimizations in parallelization and pipelining across primitives of different modules, and enhances scheduling to improve application-level performance. We build Teola, a novel orchestration framework for LLM-based applications that implements this scheme. Comprehensive experiments show that Teola can achieve up to 2.09x speedup over existing systems across various popular LLM applications. |
2024-12-15 | |
| GaLore$+$: Boosting Low-Rank Adaptation for LLMs with Cross-Head Projection Recent low-rank training methods, such as GaLore, have significantly reduced the memory required to optimize large language models (LLMs). However, these methods often suffer from time-consuming low-rank projection estimations. In particular, the singular value decomposition (SVD) in GaLore can consume more than 80\% of the total training time. To address this issue, we propose GaLore$+$, which uses cross-head low-rank projection to reduce the substantial time consumption in estimating low-rank projections for multi-head attention. In addition, we employ randomized subspace iteration to achieve fast SVD. To further enhance performance, we propose sparsely coded residuals to reduce the errors caused by low-rank approximation on the first- and second-order moments of the optimizers and weight updates. We evaluate GaLore$+$ on arithmetic reasoning and natural language generation datasets. Our experiments demonstrate that GaLore$+$ delivers superior performance while achieving approximately $4\times$ fine-tuning speed compared to vanilla GaLore. |
2024-12-15 | |
| BlockLLM: Memory-Efficient Adaptation of LLMs by Selecting and Optimizing the Right Coordinate Blocks Training large language models (LLMs) for pretraining or adapting to new tasks and domains has become increasingly critical as their applications expand. However, as the model and the data sizes grow, the training process presents significant memory challenges, often requiring a prohibitive amount of GPU memory that may not be readily available. Existing methods such as low-rank adaptation (LoRA) add trainable low-rank matrix factorizations, altering the training dynamics and limiting the model’s parameter search to a low-rank subspace. GaLore, a more recent method, employs Gradient Low-Rank Projection to reduce the memory footprint, in the full parameter training setting. However GaLore can only be applied to a subset of the LLM layers that satisfy the “reversibility” property, thus limiting their applicability. In response to these challenges, we introduce BlockLLM, an approach inspired by block coordinate descent. Our method carefully selects and updates a very small subset of the trainable parameters without altering any part of its architecture and training procedure. BlockLLM achieves state-of-the-art performance in both finetuning and pretraining tasks, while reducing the memory footprint of the underlying optimization process. Our experiments demonstrate that fine-tuning with only less than 5% of the parameters, BlockLLM achieves state-of-the-art perplexity scores on the GLUE benchmarks. On Llama model pretrained on C4 dataset, BlockLLM is able to train with significantly less memory than the state-of-the-art, while still maintaining competitive performance. |
2024-12-15 | 18 pages, 7 figures |
| Judging the Judges: A Systematic Study of Position Bias in LLM-as-a-Judge LLM-as-a-Judge presents a promising alternative to human evaluators across various tasks, but inherent biases, especially position bias - a tendency to favor solutions based on their position in the prompt - have compromised its effectiveness. Our study introduces a systematic framework to examine position bias in pairwise comparisons, focusing on repetition stability, position consistency, and preference fairness. This research significantly contributes to the field by introducing new concepts for understanding position bias and providing a multi-dimensional framework for evaluations. We conducted experiments with 12 LLM judges across MTBench and DevBench, covering 22 tasks and approximately 40 solution-generating models - candidates, resulting in over 100,000 evaluation instances. Our findings confirm that position bias in capable LLM judges is not due to random chances, along with notable variations observed across judges and tasks. Moreover, position bias is weakly influenced by the length of prompt components but significantly impacted by the quality gap between solutions. These insights can help optimize judge model selections, improve benchmark design, and inform future research on debiasing strategies, ultimately enhancing the reliability of LLM judges. |
2024-12-15 | |
| An Efficient Sign Language Translation Using Spatial Configuration and Motion Dynamics with LLMs Gloss-free Sign Language Translation (SLT) converts sign videos directly into spoken language sentences without relying on glosses. Recently, Large Language Models (LLMs) have shown remarkable translation performance in gloss-free methods by harnessing their powerful natural language generation capabilities. However, these methods often rely on domain-specific fine-tuning of visual encoders to achieve optimal results. By contrast, this paper emphasizes the importance of capturing the spatial configurations and motion dynamics inherent in sign language. With this in mind, we introduce Spatial and Motion-based Sign Language Translation (SpaMo), a novel LLM-based SLT framework. The core idea of SpaMo is simple yet effective. We first extract spatial and motion features using off-the-shelf visual encoders and then input these features into an LLM with a language prompt. Additionally, we employ a visual-text alignment process as a warm-up before the SLT supervision. Our experiments demonstrate that SpaMo achieves state-of-the-art performance on two popular datasets, PHOENIX14T and How2Sign. |
2024-12-15 | Under Review |
| NITRO: LLM Inference on Intel Laptop NPUs Large Language Models (LLMs) have become essential tools in natural language processing, finding large usage in chatbots such as ChatGPT and Gemini, and are a central area of research. A particular area of interest includes designing hardware specialized for these AI applications, with one such example being the neural processing unit (NPU). In 2023, Intel released the Intel Core Ultra processor with codename Meteor Lake, featuring a CPU, GPU, and NPU system-on-chip. However, official software support for the NPU through Intel’s OpenVINO framework is limited to static model inference. The dynamic nature of autoregressive token generation in LLMs is therefore not supported out of the box. To address this shortcoming, we present NITRO (NPU Inference for Transformers Optimization), a Python-based framework built on top of OpenVINO to support text and chat generation on NPUs. In this paper, we discuss in detail the key modifications made to the transformer architecture to enable inference, some performance benchmarks, and future steps towards improving the package. The code repository for NITRO can be found here: https://github.com/abdelfattah-lab/nitro. |
2024-12-15 | 11 pages, 7 figures |
| SceneLLM: Implicit Language Reasoning in LLM for Dynamic Scene Graph Generation Dynamic scenes contain intricate spatio-temporal information, crucial for mobile robots, UAVs, and autonomous driving systems to make informed decisions. Parsing these scenes into semantic triplets |
2024-12-15 | 27 pages, 4 figures |
| PromptV: Leveraging LLM-powered Multi-Agent Prompting for High-quality Verilog Generation Recent advances in agentic LLMs have demonstrated remarkable automated Verilog code generation capabilities. However, existing approaches either demand substantial computational resources or rely on LLM-assisted single-agent prompt learning techniques, which we observe for the first time has a degeneration issue - characterized by deteriorating generative performance and diminished error detection and correction capabilities. This paper proposes a novel multi-agent prompt learning framework to address these limitations and enhance code generation quality. We show for the first time that multi-agent architectures can effectively mitigate the degeneration risk while improving code error correction capabilities, resulting in higher-quality Verilog code generation. Experimental results show that the proposed method could achieve 96.4% and 96.5% pass@10 scores on VerilogEval Machine and Human benchmarks, respectively while attaining 100% Syntax and 99.9% Functionality pass@5 metrics on the RTLLM benchmark. |
2024-12-15 | |
| LLMs for Literature Review: Are we there yet? Literature reviews are an essential component of scientific research, but they remain time-intensive and challenging to write, especially due to the recent influx of research papers. This paper explores the zero-shot abilities of recent Large Language Models (LLMs) in assisting with the writing of literature reviews based on an abstract. We decompose the task into two components: 1. Retrieving related works given a query abstract, and 2. Writing a literature review based on the retrieved results. We analyze how effective LLMs are for both components. For retrieval, we introduce a novel two-step search strategy that first uses an LLM to extract meaningful keywords from the abstract of a paper and then retrieves potentially relevant papers by querying an external knowledge base. Additionally, we study a prompting-based re-ranking mechanism with attribution and show that re-ranking doubles the normalized recall compared to naive search methods, while providing insights into the LLM’s decision-making process. In the generation phase, we propose a two-step approach that first outlines a plan for the review and then executes steps in the plan to generate the actual review. To evaluate different LLM-based literature review methods, we create test sets from arXiv papers using a protocol designed for rolling use with newly released LLMs to avoid test set contamination in zero-shot evaluations. We release this evaluation protocol to promote additional research and development in this regard. Our empirical results suggest that LLMs show promising potential for writing literature reviews when the task is decomposed into smaller components of retrieval and planning. Further, we demonstrate that our planning-based approach achieves higher-quality reviews by minimizing hallucinated references in the generated review by 18-26% compared to existing simpler LLM-based generation methods. |
2024-12-15 | |
| Can LLMs Help Create Grammar?: Automating Grammar Creation for Endangered Languages with In-Context Learning Yes! In the present-day documenting and preserving endangered languages, the application of Large Language Models (LLMs) presents a promising approach. This paper explores how LLMs, particularly through in-context learning, can assist in generating grammatical information for low-resource languages with limited amount of data. We takes Moklen as a case study to evaluate the efficacy of LLMs in producing coherent grammatical rules and lexical entries using only bilingual dictionaries and parallel sentences of the unknown language without building the model from scratch. Our methodology involves organising the existing linguistic data and prompting to efficiently enable to generate formal XLE grammar. Our results demonstrate that LLMs can successfully capture key grammatical structures and lexical information, although challenges such as the potential for English grammatical biases remain. This study highlights the potential of LLMs to enhance language documentation efforts, providing a cost-effective solution for generating linguistic data and contributing to the preservation of endangered languages. |
2024-12-14 | Preprint manuscript. Under revision. Accepted to COLING 2025 |
| Attribute Structuring Improves LLM-Based Evaluation of Clinical Text Summaries Summarizing clinical text is crucial in health decision-support and clinical research. Large language models (LLMs) have shown the potential to generate accurate clinical text summaries, but still struggle with issues regarding grounding and evaluation, especially in safety-critical domains such as health. Holistically evaluating text summaries is challenging because they may contain unsubstantiated information. Here, we explore a general mitigation framework using Attribute Structuring (AS), which structures the summary evaluation process. It decomposes the evaluation process into a grounded procedure that uses an LLM for relatively simple structuring and scoring tasks, rather than the full task of holistic summary evaluation. Experiments show that AS consistently improves the correspondence between human annotations and automated metrics in clinical text summarization. Additionally, AS yields interpretations in the form of a short text span corresponding to each output, which enables efficient human auditing, paving the way towards trustworthy evaluation of clinical information in resource-constrained scenarios. We release our code, prompts, and an open-source benchmark at https://github.com/microsoft/attribute-structuring. |
2024-12-14 | Published in ML4H Findings 2024, 4 pages |
| LLMs-in-the-Loop Part 2: Expert Small AI Models for Anonymization and De-identification of PHI Across Multiple Languages The rise of chronic diseases and pandemics like COVID-19 has emphasized the need for effective patient data processing while ensuring privacy through anonymization and de-identification of protected health information (PHI). Anonymized data facilitates research without compromising patient confidentiality. This paper introduces expert small AI models developed using the LLM-in-the-loop methodology to meet the demand for domain-specific de-identification NER models. These models overcome the privacy risks associated with large language models (LLMs) used via APIs by eliminating the need to transmit or store sensitive data. More importantly, they consistently outperform LLMs in de-identification tasks, offering superior performance and reliability. Our de-identification NER models, developed in eight languages (English, German, Italian, French, Romanian, Turkish, Spanish, and Arabic) achieved f1-micro score averages of 0.966, 0.975, 0.976, 0.970, 0.964, 0.974, 0.978, and 0.953 respectively. These results establish them as the most accurate healthcare anonymization solutions, surpassing existing small models and even general-purpose LLMs such as GPT-4o. While Part-1 of this series introduced the LLM-in-the-loop methodology for bio-medical document translation, this second paper showcases its success in developing cost-effective expert small NER models in de-identification tasks. Our findings lay the groundwork for future healthcare AI innovations, including biomedical entity and relation extraction, demonstrating the value of specialized models for domain-specific challenges. |
2024-12-14 | 21 pages, 7 tables |
| SusGen-GPT: A Data-Centric LLM for Financial NLP and Sustainability Report Generation The rapid growth of the financial sector and the rising focus on Environmental, Social, and Governance (ESG) considerations highlight the need for advanced NLP tools. However, open-source LLMs proficient in both finance and ESG domains remain scarce. To address this gap, we introduce SusGen-30K, a category-balanced dataset comprising seven financial NLP tasks and ESG report generation, and propose TCFD-Bench, a benchmark for evaluating sustainability report generation. Leveraging this dataset, we developed SusGen-GPT, a suite of models achieving state-of-the-art performance across six adapted and two off-the-shelf tasks, trailing GPT-4 by only 2% despite using 7-8B parameters compared to GPT-4’s 1,700B. Based on this, we propose the SusGen system, integrated with Retrieval-Augmented Generation (RAG), to assist in sustainability report generation. This work demonstrates the efficiency of our approach, advancing research in finance and ESG. |
2024-12-14 | |
| CEKER: A Generalizable LLM Framework for Literature Analysis with a Case Study in Unikernel Security Literature reviews are a critical component of formulating and justifying new research, but are a manual and often time-consuming process. This research introduces a novel, generalizable approach to literature analysis called CEKER which uses a three-step process to streamline the collection of literature, the extraction of key insights, and the summarized analysis of key trends and gaps. Leveraging Large Language Models (LLMs), this methodology represents a significant shift from traditional manual literature reviews, offering a scalable, flexible, and repeatable approach that can be applied across diverse research domains. A case study on unikernel security illustrates CEKER’s ability to generate novel insights validated against previous manual methods. CEKER’s analysis highlighted reduced attack surface as the most prominent theme. Key security gaps included the absence of Address Space Layout Randomization, missing debugging tools, and limited entropy generation, all of which represent important challenges to unikernel security. The study also revealed a reliance on hypervisors as a potential attack vector and emphasized the need for dynamic security adjustments to address real-time threats. |
2024-12-14 | 7 pages, 2 figures |
| Generating executable oracles to check conformance of client code to requirements of JDK Javadocs using LLMs Software testing remains the most widely used methodology for validating quality of code. However, effectiveness of testing critically depends on the quality of test suites used. Test cases in a test suite consist of two fundamental parts: (1) input values for the code under test, and (2) correct checks for the outputs it produces. These checks are commonly written as assertions, and termed test oracles. The last couple of decades have seen much progress in automated test input generation, e.g., using fuzzing and symbolic execution. However, automating test oracles remains a relatively less explored problem area. Indeed, a test oracle by its nature requires knowledge of expected behavior, which may only be known to the developer and may not not exist in a formal language that supports automated reasoning. Our focus in this paper is automation of test oracles for clients of widely used Java libraries, e.g., java.lang and java.util packages. Our key insight is that Javadocs that provide a rich source of information can enable automated generation of test oracles. Javadocs of the core Java libraries are fairly detailed documents that contain natural language descriptions of not only how the libraries behave but also how the clients must (not) use them. We use large language models as an enabling technology to embody our insight into a framework for test oracle automation, and evaluate it experimentally. Our experiments demonstrate that LLMs can generate oracles for checking normal and exceptional behaviors from Javadocs, with 98.8% of these oracles being compilable and 96.4% accurately reflecting intended properties. Even for the few incorrect oracles, errors are minor and can be easily corrected with the help of additional comment information generated by the LLMs. |
2024-12-14 | |
| From Words to Worth: Newborn Article Impact Prediction with LLM As the academic landscape expands, the challenge of efficiently identifying impactful newly published articles grows increasingly vital. This paper introduces a promising approach, leveraging the capabilities of LLMs to predict the future impact of newborn articles solely based on titles and abstracts. Moving beyond traditional methods heavily reliant on external information, the proposed method employs LLM to discern the shared semantic features of highly impactful papers from a large collection of title-abstract pairs. These semantic features are further utilized to predict the proposed indicator, TNCSI_SP, which incorporates favorable normalization properties across value, field, and time. To facilitate parameter-efficient fine-tuning of the LLM, we have also meticulously curated a dataset containing over 12,000 entries, each annotated with titles, abstracts, and their corresponding TNCSI_SP values. The quantitative results, with an MAE of 0.216 and an NDCG@20 of 0.901, demonstrate that the proposed approach achieves state-of-the-art performance in predicting the impact of newborn articles when compared to several promising methods. Finally, we present a real-world application example for predicting the impact of newborn journal articles to demonstrate its noteworthy practical value. Overall, our findings challenge existing paradigms and propose a shift towards a more content-focused prediction of academic impact, offering new insights for article impact prediction. |
2024-12-14 | Accepted by AAAI 2025. Code, dataset are released at https://sway.cloud.microsoft/KOH09sPR21Ubojbc |
| FinGPT: Enhancing Sentiment-Based Stock Movement Prediction with Dissemination-Aware and Context-Enriched LLMs Financial sentiment analysis is crucial for understanding the influence of news on stock prices. Recently, large language models (LLMs) have been widely adopted for this purpose due to their advanced text analysis capabilities. However, these models often only consider the news content itself, ignoring its dissemination, which hampers accurate prediction of short-term stock movements. Additionally, current methods often lack sufficient contextual data and explicit instructions in their prompts, limiting LLMs’ ability to interpret news. In this paper, we propose a data-driven approach that enhances LLM-powered sentiment-based stock movement predictions by incorporating news dissemination breadth, contextual data, and explicit instructions. We cluster recent company-related news to assess its reach and influence, enriching prompts with more specific data and precise instructions. This data is used to construct an instruction tuning dataset to fine-tune an LLM for predicting short-term stock price movements. Our experimental results show that our approach improves prediction accuracy by 8\% compared to existing methods. |
2024-12-14 | 1st Workshop on Preparing Good Data for Generative AI: Challenges and Approaches@ AAAI 2025 |
| TrendSim: Simulating Trending Topics in Social Media Under Poisoning Attacks with LLM-based Multi-agent System Trending topics have become a significant part of modern social media, attracting users to participate in discussions of breaking events. However, they also bring in a new channel for poisoning attacks, resulting in negative impacts on society. Therefore, it is urgent to study this critical problem and develop effective strategies for defense. In this paper, we propose TrendSim, an LLM-based multi-agent system to simulate trending topics in social media under poisoning attacks. Specifically, we create a simulation environment for trending topics that incorporates a time-aware interaction mechanism, centralized message dissemination, and an interactive system. Moreover, we develop LLM-based human-like agents to simulate users in social media, and propose prototype-based attackers to replicate poisoning attacks. Besides, we evaluate TrendSim from multiple aspects to validate its effectiveness. Based on TrendSim, we conduct simulation experiments to study four critical problems about poisoning attacks on trending topics for social benefit. |
2024-12-14 | 19 pages, 9 tables, 8 figure |
| Exploring Critical Testing Scenarios for Decision-Making Policies: An LLM Approach Recent advances in decision-making policies have led to significant progress in fields such as autonomous driving and robotics. However, testing these policies remains crucial with the existence of critical scenarios that may threaten their reliability. Despite ongoing research, challenges such as low testing efficiency and limited diversity persist due to the complexity of the decision-making policies and their environments. To address these challenges, this paper proposes an adaptable Large Language Model (LLM)-driven online testing framework to explore critical and diverse testing scenarios for decision-making policies. Specifically, we design a “generate-test-feedback” pipeline with templated prompt engineering to harness the world knowledge and reasoning abilities of LLMs. Additionally, a multi-scale scenario generation strategy is proposed to address the limitations of LLMs in making fine-grained adjustments, further enhancing testing efficiency. Finally, the proposed LLM-driven method is evaluated on five widely recognized benchmarks, and the experimental results demonstrate that our method significantly outperforms baseline methods in uncovering both critical and diverse scenarios. These findings suggest that LLM-driven methods hold significant promise for advancing the testing of decision-making policies. |
2024-12-14 | 16 pages, 13 figures |
| WEPO: Web Element Preference Optimization for LLM-based Web Navigation The rapid advancement of autonomous web navigation has significantly benefited from grounding pretrained Large Language Models (LLMs) as agents. However, current research has yet to fully leverage the redundancy of HTML elements for contrastive training. This paper introduces a novel approach to LLM-based web navigation tasks, called Web Element Preference Optimization (WEPO). WEPO utilizes unsupervised preference learning by sampling distance-based non-salient web elements as negative samples, optimizing maximum likelihood objective within Direct Preference Optimization (DPO). We evaluate WEPO on the Mind2Web benchmark and empirically demonstrate that WEPO aligns user high-level intent with output actions more effectively. The results show that our method achieved the state-of-the-art, with an improvement of 13.8% over WebAgent and 5.3% over the visual language model CogAgent baseline. Our findings underscore the potential of preference optimization to enhance web navigation and other web page based tasks, suggesting a promising direction for future research. |
2024-12-14 | Published at AAAI 2025 |
| InstructPipe: Building Visual Programming Pipelines with Human Instructions Using LLMs Visual programming has the potential of providing novice programmers with a low-code experience to build customized processing pipelines. Existing systems typically require users to build pipelines from scratch, implying that novice users are expected to set up and link appropriate nodes from a blank workspace. In this paper, we introduce InstructPipe, an AI assistant for prototyping machine learning (ML) pipelines with text instructions. We contribute two large language model (LLM) modules and a code interpreter as part of our framework. The LLM modules generate pseudocode for a target pipeline, and the interpreter renders the pipeline in the node-graph editor for further human-AI collaboration. Both technical and user evaluation (N=16) shows that InstructPipe empowers users to streamline their ML pipeline workflow, reduce their learning curve, and leverage open-ended commands to spark innovative ideas. |
2024-12-14 | |
| Learning to Verify Summary Facts with Fine-Grained LLM Feedback Training automatic summary fact verifiers often faces the challenge of a lack of human-labeled data. In this paper, we explore alternative way of leveraging Large Language Model (LLM) generated feedback to address the inherent limitation of using human-labeled data. We introduce FineSumFact, a large-scale dataset containing fine-grained factual feedback on summaries. We employ 10 distinct LLMs for diverse summary generation and Llama-3-70B-Instruct for feedback. We utilize this dataset to fine-tune the lightweight open-source model Llama-3-8B-Instruct, optimizing resource efficiency while maintaining high performance. Our experimental results reveal that the model trained on extensive LLM-generated datasets surpasses that trained on smaller human-annotated datasets when evaluated using human-generated test sets. Fine-tuning fact verification models with LLM feedback can be more effective and cost-efficient than using human feedback. The dataset is available at https://github.com/DISL-Lab/FineSumFact. |
2024-12-14 | Accepted at COLING 2025 |
| Chasing Progress, Not Perfection: Revisiting Strategies for End-to-End LLM Plan Generation The capability of Large Language Models (LLMs) to plan remains a topic of debate. Some critics argue that strategies to boost LLMs’ reasoning skills are ineffective in planning tasks, while others report strong outcomes merely from training models on a planning corpus. This study reassesses recent strategies by developing an end-to-end LLM planner and employing diverse metrics for a thorough evaluation. We find that merely fine-tuning LLMs on a corpus of planning instances does not lead to robust planning skills, as indicated by poor performance on out-of-distribution test sets. At the same time, we find that various strategies, including Chain-of-Thought, do enhance the probability of a plan being executable. This indicates progress towards better plan quality, despite not directly enhancing the final validity rate. Among the strategies we evaluated, reinforcement learning with our novel `Longest Contiguous Common Subsequence’ reward emerged as the most effective, contributing to both plan validity and executability. Overall, our research addresses key misconceptions in the LLM-planning literature; we validate incremental progress in plan executability, although plan validity remains a challenge. Hence, future strategies should focus on both these aspects, drawing insights from our findings. |
2024-12-14 | 8 pages main body, 10 pages appendix, accepted by Workshop on Planning in the Era of LLMs (LM4Plan @ AAAI 2025) |
| Thinking with Knowledge Graphs: Enhancing LLM Reasoning Through Structured Data Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation. However, they often struggle with complex reasoning tasks and are prone to hallucination. Recent research has shown promising results in leveraging knowledge graphs (KGs) to enhance LLM performance. KGs provide a structured representation of entities and their relationships, offering a rich source of information that can enhance the reasoning capabilities of LLMs. For this work, we have developed different techniques that tightly integrate KG structures and semantics into LLM representations. Our results show that we are able to significantly improve the performance of LLMs in complex reasoning scenarios, and ground the reasoning process with KGs. We are the first to represent KGs with programming language and fine-tune pretrained LLMs with KGs. This integration facilitates more accurate and interpretable reasoning processes, paving the way for more advanced reasoning capabilities of LLMs. |
2024-12-14 | |
| KVDirect: Distributed Disaggregated LLM Inference Large Language Models (LLMs) have become the new foundation for many applications, reshaping human society like a storm. Disaggregated inference, which separates prefill and decode stages, is a promising approach to improving hardware utilization and service quality. However, due to inefficient inter-node communication, existing systems restrict disaggregated inference to a single node, limiting resource allocation flexibility and reducing service capacity. This paper introduces KVDirect, which optimizes KV cache transfer to enable a distributed disaggregated LLM inference. KVDirect achieves this through the following contributions. First, we propose a novel tensor-centric communication mechanism that reduces the synchronization overhead in traditional distributed GPU systems. Second, we design a custom communication library to support dynamic GPU resource scheduling and efficient KV cache transfer. Third, we introduce a pull-based KV cache transfer strategy that reduces GPU resource idling and improves latency. Finally, we implement KVDirect as an open-source LLM inference framework. Our evaluation demonstrates that KVDirect reduces per-request latency by 55% compared to the baseline across diverse workloads under the same resource constraints. |
2024-12-13 | |
| AdvPrefix: An Objective for Nuanced LLM Jailbreaks Many jailbreak attacks on large language models (LLMs) rely on a common objective: making the model respond with the prefix “Sure, here is (harmful request)”. While straightforward, this objective has two limitations: limited control over model behaviors, often resulting in incomplete or unrealistic responses, and a rigid format that hinders optimization. To address these limitations, we introduce AdvPrefix, a new prefix-forcing objective that enables more nuanced control over model behavior while being easy to optimize. Our objective leverages model-dependent prefixes, automatically selected based on two criteria: high prefilling attack success rates and low negative log-likelihood. It can further simplify optimization by using multiple prefixes for a single user request. AdvPrefix can integrate seamlessly into existing jailbreak attacks to improve their performance for free. For example, simply replacing GCG attack’s target prefixes with ours on Llama-3 improves nuanced attack success rates from 14% to 80%, suggesting that current alignment struggles to generalize to unseen prefixes. Our work demonstrates the importance of jailbreak objectives in achieving nuanced jailbreaks. |
2024-12-13 | |
| One world, one opinion? The superstar effect in LLM responses As large language models (LLMs) are shaping the way information is shared and accessed online, their opinions have the potential to influence a wide audience. This study examines who the LLMs view as the most prominent figures across various fields, using prompts in ten different languages to explore the influence of linguistic diversity. Our findings reveal low diversity in responses, with a small number of figures dominating recognition across languages (also known as the “superstar effect”). These results highlight the risk of narrowing global knowledge representation when LLMs retrieve subjective information. |
2024-12-13 | |
| Cultural Evolution of Cooperation among LLM Agents Large language models (LLMs) provide a compelling foundation for building generally-capable AI agents. These agents may soon be deployed at scale in the real world, representing the interests of individual humans (e.g., AI assistants) or groups of humans (e.g., AI-accelerated corporations). At present, relatively little is known about the dynamics of multiple LLM agents interacting over many generations of iterative deployment. In this paper, we examine whether a “society” of LLM agents can learn mutually beneficial social norms in the face of incentives to defect, a distinctive feature of human sociality that is arguably crucial to the success of civilization. In particular, we study the evolution of indirect reciprocity across generations of LLM agents playing a classic iterated Donor Game in which agents can observe the recent behavior of their peers. We find that the evolution of cooperation differs markedly across base models, with societies of Claude 3.5 Sonnet agents achieving significantly higher average scores than Gemini 1.5 Flash, which, in turn, outperforms GPT-4o. Further, Claude 3.5 Sonnet can make use of an additional mechanism for costly punishment to achieve yet higher scores, while Gemini 1.5 Flash and GPT-4o fail to do so. For each model class, we also observe variation in emergent behavior across random seeds, suggesting an understudied sensitive dependence on initial conditions. We suggest that our evaluation regime could inspire an inexpensive and informative new class of LLM benchmarks, focussed on the implications of LLM agent deployment for the cooperative infrastructure of society. |
2024-12-13 | 15 pages, 6 figures |
| Detecting LLM Hallucination Through Layer-wise Information Deficiency: Analysis of Unanswerable Questions and Ambiguous Prompts Large language models (LLMs) frequently generate confident yet inaccurate responses, introducing significant risks for deployment in safety-critical domains. We present a novel approach to detecting model hallucination through systematic analysis of information flow across model layers when processing inputs with insufficient or ambiguous context. Our investigation reveals that hallucination manifests as usable information deficiencies in inter-layer transmissions. While existing approaches primarily focus on final-layer output analysis, we demonstrate that tracking cross-layer information dynamics ($\mathcal{L}$I) provides robust indicators of model reliability, accounting for both information gain and loss during computation. $\mathcal{L}$I improves model reliability by immediately integrating with universal LLMs without additional training or architectural modifications. |
2024-12-13 | |
| Efficient Continual Pre-training of LLMs for Low-resource Languages Open-source Large Language models (OsLLMs) propel the democratization of natural language research by giving the flexibility to augment or update model parameters for performance improvement. Nevertheless, like proprietary LLMs, Os-LLMs offer poorer performance on low-resource languages (LRLs) than high-resource languages (HRLs), owing to smaller amounts of training data and underrepresented vocabulary. On the other hand, continual pre-training (CPT) with large amounts of language-specific data is a costly proposition in terms of data acquisition and computational resources. Our goal is to drastically reduce CPT cost. To that end, we first develop a new algorithm to select a subset of texts from a larger corpus. We show the effectiveness of our technique using very little CPT data. In search of further improvement, we design a new algorithm to select tokens to include in the LLM vocabulary. We experiment with the recent Llama-3 model and nine Indian languages with diverse scripts and extent of resource availability. For evaluation, we use IndicGenBench, a generation task benchmark dataset for Indic languages. We experiment with various CPT corpora and augmented vocabulary size and offer insights across language families. |
2024-12-13 | |
| How good is my story? Towards quantitative metrics for evaluating LLM-generated XAI narratives A rapidly developing application of LLMs in XAI is to convert quantitative explanations such as SHAP into user-friendly narratives to explain the decisions made by smaller prediction models. Evaluating the narratives without relying on human preference studies or surveys is becoming increasingly important in this field. In this work we propose a framework and explore several automated metrics to evaluate LLM-generated narratives for explanations of tabular classification tasks. We apply our approach to compare several state-of-the-art LLMs across different datasets and prompt types. As a demonstration of their utility, these metrics allow us to identify new challenges related to LLM hallucinations for XAI narratives. |
2024-12-13 | |
| RTL-Breaker: Assessing the Security of LLMs against Backdoor Attacks on HDL Code Generation Large language models (LLMs) have demonstrated remarkable potential with code generation/completion tasks for hardware design. In fact, LLM-based hardware description language (HDL) code generation has enabled the industry to realize complex designs more quickly, reducing the time and effort required in the development cycle. However, the increased reliance on such automation introduces critical security risks. Notably, given that LLMs have to be trained on vast datasets of codes that are typically sourced from publicly available repositories (often without thorough validation), LLMs are susceptible to so-called data poisoning or backdoor attacks. Here, attackers inject malicious code for the training data, which can be carried over into the HDL code generated by LLMs. This threat vector can compromise the security and integrity of entire hardware systems. In this work, we propose RTL-Breaker, a novel backdoor attack framework on LLM-based HDL code generation. RTL-Breaker provides an in-depth analysis for essential aspects of this novel problem: 1) various trigger mechanisms versus their effectiveness for inserting malicious modifications, and 2) side-effects by backdoor attacks on code generation in general, i.e., impact on code quality. RTL-Breaker emphasizes the urgent need for more robust measures to safeguard against such attacks. Toward that end, we open-source our framework and all data. |
2024-12-13 | Accepted at 2025 Design, Automation & Test in Europe (DATE) Conference |
| From Allies to Adversaries: Manipulating LLM Tool-Calling through Adversarial Injection Tool-calling has changed Large Language Model (LLM) applications by integrating external tools, significantly enhancing their functionality across diverse tasks. However, this integration also introduces new security vulnerabilities, particularly in the tool scheduling mechanisms of LLM, which have not been extensively studied. To fill this gap, we present ToolCommander, a novel framework designed to exploit vulnerabilities in LLM tool-calling systems through adversarial tool injection. Our framework employs a well-designed two-stage attack strategy. Firstly, it injects malicious tools to collect user queries, then dynamically updates the injected tools based on the stolen information to enhance subsequent attacks. These stages enable ToolCommander to execute privacy theft, launch denial-of-service attacks, and even manipulate business competition by triggering unscheduled tool-calling. Notably, the ASR reaches 91.67% for privacy theft and hits 100% for denial-of-service and unscheduled tool calling in certain cases. Our work demonstrates that these vulnerabilities can lead to severe consequences beyond simple misuse of tool-calling systems, underscoring the urgent need for robust defensive strategies to secure LLM Tool-calling systems. |
2024-12-13 | |
| MPPO: Multi Pair-wise Preference Optimization for LLMs with Arbitrary Negative Samples Aligning Large Language Models (LLMs) with human feedback is crucial for their development. Existing preference optimization methods such as DPO and KTO, while improved based on Reinforcement Learning from Human Feedback (RLHF), are inherently derived from PPO, requiring a reference model that adds GPU memory resources and relies heavily on abundant preference data. Meanwhile, current preference optimization research mainly targets single-question scenarios with two replies, neglecting optimization with multiple replies, which leads to a waste of data in the application. This study introduces the MPPO algorithm, which leverages the average likelihood of model responses to fit the reward function and maximizes the utilization of preference data. Through a comparison of Point-wise, Pair-wise, and List-wise implementations, we found that the Pair-wise approach achieves the best performance, significantly enhancing the quality of model responses. Experimental results demonstrate MPPO’s outstanding performance across various benchmarks. On MT-Bench, MPPO outperforms DPO, ORPO, and SimPO. Notably, on Arena-Hard, MPPO surpasses DPO and ORPO by substantial margins. These achievements underscore the remarkable advantages of MPPO in preference optimization tasks. |
2024-12-13 | Accepted by COLING2025 |
| TACOMORE: Leveraging the Potential of LLMs in Corpus-based Discourse Analysis with Prompt Engineering The capacity of LLMs to carry out automated qualitative analysis has been questioned by corpus linguists, and it has been argued that corpus-based discourse analysis incorporating LLMs is hindered by issues of unsatisfying performance, hallucination, and irreproducibility. Our proposed method, TACOMORE, aims to address these concerns by serving as an effective prompting framework in this domain. The framework consists of four principles, i.e., Task, Context, Model and Reproducibility, and specifies five fundamental elements of a good prompt, i.e., Role Description, Task Definition, Task Procedures, Contextual Information and Output Format. We conduct experiments on three LLMs, i.e., GPT-4o, Gemini-1.5-Pro and Gemini-1.5.Flash, and find that TACOMORE helps improve LLM performance in three representative discourse analysis tasks, i.e., the analysis of keywords, collocates and concordances, based on an open corpus of COVID-19 research articles. Our findings show the efficacy of the proposed prompting framework TACOMORE in corpus-based discourse analysis in terms of Accuracy, Ethicality, Reasoning, and Reproducibility, and provide novel insights into the application and evaluation of LLMs in automated qualitative studies. |
2024-12-13 | |
| Can LLMs Convert Graphs to Text-Attributed Graphs? Graphs are ubiquitous data structures found in numerous real-world applications, such as drug discovery, recommender systems, and social network analysis. Graph neural networks (GNNs) have become a popular tool to learn node embeddings through message passing on these structures. However, a significant challenge arises when applying GNNs to multiple graphs with different feature spaces, as existing GNN architectures are not designed for cross-graph feature alignment. To address this, recent approaches introduce text-attributed graphs, where each node is associated with a textual description, enabling the use of a shared textual encoder to project nodes from different graphs into a unified feature space. While promising, this method relies heavily on the availability of text-attributed data, which can be difficult to obtain in practice. To bridge this gap, we propose a novel method named Topology-Aware Node description Synthesis (TANS), which leverages large language models (LLMs) to automatically convert existing graphs into text-attributed graphs. The key idea is to integrate topological information with each node’s properties, enhancing the LLMs’ ability to explain how graph topology influences node semantics. We evaluate our TANS on text-rich, text-limited, and text-free graphs, demonstrating that it enables a single GNN to operate across diverse graphs. Notably, on text-free graphs, our method significantly outperforms existing approaches that manually design node features, showcasing the potential of LLMs for preprocessing graph-structured data, even in the absence of textual information. The code and data are available at https://github.com/Zehong-Wang/TANS. |
2024-12-13 | |
| You Name It, I Run It: An LLM Agent to Execute Tests of Arbitrary Projects The ability to execute the test suite of a project is essential in many scenarios, e.g., to assess code quality and code coverage, to validate code changes made by developers or automated tools, and to ensure compatibility with dependencies. Despite its importance, executing the test suite of a project can be challenging in practice because different projects use different programming languages, software ecosystems, build systems, testing frameworks, and other tools. These challenges make it difficult to create a reliable, universal test execution method that works across different projects. This paper presents ExecutionAgent, an automated technique that installs arbitrary projects, configures them to run test cases, and produces project-specific scripts to reproduce the setup. Inspired by the way a human developer would address this task, our approach is a large language model-based agent that autonomously executes commands and interacts with the host system. The agent uses meta-prompting to gather guidelines on the latest technologies related to the given project, and it iteratively refines its process based on feedback from the previous steps. Our evaluation applies ExecutionAgent to 50 open-source projects that use 14 different programming languages and many different build and testing tools. The approach successfully executes the test suites of 33/55 projects, while matching the test results of ground truth test suite executions with a deviation of only 7.5\%. These results improve over the best previously available technique by 6.6x. The costs imposed by the approach are reasonable, with an execution time of 74 minutes and LLM costs of 0.16 dollars, on average per project. We envision ExecutionAgent to serve as a valuable tool for developers, automated programming tools, and researchers that need to execute tests across a wide variety of projects. |
2024-12-13 | |
| Script-Based Dialog Policy Planning for LLM-Powered Conversational Agents: A Basic Architecture for an “AI Therapist” Large Language Model (LLM)-Powered Conversational Agents have the potential to provide users with scaled behavioral healthcare support, and potentially even deliver full-scale “AI therapy’” in the future. While such agents can already conduct fluent and proactive emotional support conversations, they inherently lack the ability to (a) consistently and reliably act by predefined rules to align their conversation with an overarching therapeutic concept and (b) make their decision paths inspectable for risk management and clinical evaluation – both essential requirements for an “AI Therapist”. In this work, we introduce a novel paradigm for dialog policy planning in conversational agents enabling them to (a) act according to an expert-written “script” that outlines the therapeutic approach and (b) explicitly transition through a finite set of states over the course of the conversation. The script acts as a deterministic component, constraining the LLM’s behavior in desirable ways and establishing a basic architecture for an AI Therapist. We implement two variants of Script-Based Dialog Policy Planning using different prompting techniques and synthesize a total of 100 conversations with LLM-simulated patients. The results demonstrate the feasibility of this new technology and provide insights into the efficiency and effectiveness of different implementation variants. |
2024-12-13 | 9 pages, 5 figures, 1 table |
| GAOKAO-Eval: Does high scores truly reflect strong capabilities in LLMs? Large Language Models (LLMs) are commonly evaluated using human-crafted benchmarks, under the premise that higher scores implicitly reflect stronger human-like performance. However, there is growing concern that LLMs may game" these benchmarks due to data leakage, achieving high scores while struggling with tasks simple for humans. To substantively address the problem, we create GAOKAO-Eval, a comprehensive benchmark based on China's National College Entrance Examination (Gaokao), and conductclosed-book” evaluations for representative models released prior to Gaokao. Contrary to prevailing consensus, even after addressing data leakage and comprehensiveness, GAOKAO-Eval reveals that high scores still fail to truly reflect human-aligned capabilities. To better understand this mismatch, We introduce the Rasch model from cognitive psychology to analyze LLM scoring patterns and identify two key discrepancies: 1) anomalous consistent performance across various question difficulties, and 2) high variance in performance on questions of similar difficulty. In addition, We identified inconsistent grading of LLM-generated answers among teachers and recurring mistake patterns. we find that the phenomenons are well-grounded in the motivations behind OpenAI o1, and o1’s reasoning-as-difficulties can mitigate the mismatch. These results show that GAOKAO-Eval can reveal limitations in LLM capabilities not captured by current benchmarks and highlight the need for more LLM-aligned difficulty analysis. |
2024-12-13 | 10 pages, 13 figures |