llm - 2025_02
Navigation
Papers
| Paper | Date | Comment |
|---|---|---|
| DHP Benchmark: Are LLMs Good NLG Evaluators? Large Language Models (LLMs) are increasingly serving as evaluators in Natural Language Generation (NLG) tasks; this is often referred to as ``LLM-as-a-judge’’ paradigm. However, the capabilities of LLMs in evaluating NLG quality remain underexplored. Current studies depend on human assessments and simple metrics that fail to capture the discernment of LLMs across diverse NLG tasks. To address this gap, we propose the Discernment of Hierarchical Perturbation (DHP) benchmarking framework, which provides quantitative discernment scores for LLMs. This framework leverages hierarchically perturbed text data and statistical tests to systematically measure the NLG evaluation capabilities of LLMs. We re-established six evaluation datasets for this benchmark, covering four NLG tasks: Summarization, Story Completion, Question Answering, and Translation. Our comprehensive benchmarking of five major LLM families provides critical insight into their strengths and limitations as NLG evaluators. Our dataset is available at https://huggingface.co/datasets/YCWANGVINCE/DHP_Benchmark. |
2025-02-25 | |
| Design and implementation of a distributed security threat detection system integrating federated learning and multimodal LLM Traditional security protection methods struggle to address sophisticated attack vectors in large-scale distributed systems, particularly when balancing detection accuracy with data privacy concerns. This paper presents a novel distributed security threat detection system that integrates federated learning with multimodal large language models (LLMs). Our system leverages federated learning to ensure data privacy while employing multimodal LLMs to process heterogeneous data sources including network traffic, system logs, images, and sensor data. Experimental evaluation on a 10TB distributed dataset demonstrates that our approach achieves 96.4% detection accuracy, outperforming traditional baseline models by 4.1 percentage points. The system reduces both false positive and false negative rates by 1.8 and 2.4 percentage points respectively. Performance analysis shows that our system maintains efficient processing capabilities in distributed environments, requiring 180 seconds for model training and 3.8 seconds for threat detection across the distributed network. These results demonstrate significant improvements in detection accuracy and computational efficiency while preserving data privacy, suggesting strong potential for real-world deployment in large-scale security systems. |
2025-02-25 | |
| Detection of LLM-Paraphrased Code and Identification of the Responsible LLM Using Coding Style Features Recent progress in large language models (LLMs) for code generation has raised serious concerns about intellectual property protection. Malicious users can exploit LLMs to produce paraphrased versions of proprietary code that closely resemble the original. While the potential for LLM-assisted code paraphrasing continues to grow, research on detecting it remains limited, underscoring an urgent need for detection system. We respond to this need by proposing two tasks. The first task is to detect whether code generated by an LLM is a paraphrased version of original human-written code. The second task is to identify which LLM is used to paraphrase the original code. For these tasks, we construct a dataset LPcode consisting of pairs of human-written code and LLM-paraphrased code using various LLMs. We statistically confirm significant differences in the coding styles of human-written and LLM-paraphrased code, particularly in terms of naming consistency, code structure, and readability. Based on these findings, we develop LPcodedec, a detection method that identifies paraphrase relationships between human-written and LLM-generated code, and discover which LLM is used for the paraphrasing. LPcodedec outperforms the best baselines in two tasks, improving F1 scores by 2.64% and 15.17% while achieving speedups of 1,343x and 213x, respectively. |
2025-02-25 | |
| NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models Decoder-only LLM-based embedding models are beginning to outperform BERT or T5-based embedding models in general-purpose text embedding tasks, including dense vector-based retrieval. In this work, we introduce NV-Embed, incorporating architectural designs, training procedures, and curated datasets to significantly enhance the performance of LLM as a versatile embedding model, while maintaining its simplicity and reproducibility. For model architecture, we propose a latent attention layer to obtain pooled embeddings, which consistently improves retrieval and downstream task accuracy compared to mean pooling or using the last |
2025-02-25 | ICLR 2025 (Spotlight). We open-source the model at: https://huggingface.co/nvidia/NV-Embed-v2 |
| Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned. Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment. In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It’s important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work. |
2025-02-25 | 10 pages, 9 figur |
| LUME: LLM Unlearning with Multitask Evaluations Unlearning aims to remove copyrighted, sensitive, or private content from large language models (LLMs) without a full retraining. In this work, we develop a multi-task unlearning benchmark (LUME) which features three tasks: (1) unlearn synthetically generated creative short novels, (2) unlearn synthetic biographies with sensitive information, and (3) unlearn a collection of public biographies. We further release two fine-tuned LLMs of 1B and 7B parameter sizes as the target models. We conduct detailed evaluations of several recently proposed unlearning algorithms and present results on carefully crafted metrics to understand their behavior and limitations. |
2025-02-25 | |
| From Perceptions to Decisions: Wildfire Evacuation Decision Prediction with Behavioral Theory-informed LLMs Evacuation decision prediction is critical for efficient and effective wildfire response by helping emergency management anticipate traffic congestion and bottlenecks, allocate resources, and minimize negative impacts. Traditional statistical methods for evacuation decision prediction fail to capture the complex and diverse behavioral logic of different individuals. In this work, for the first time, we introduce FLARE, short for facilitating LLM for advanced reasoning on wildfire evacuation decision prediction, a Large Language Model (LLM)-based framework that integrates behavioral theories and models to streamline the Chain-of-Thought (CoT) reasoning and subsequently integrate with memory-based Reinforcement Learning (RL) module to provide accurate evacuation decision prediction and understanding. Our proposed method addresses the limitations of using existing LLMs for evacuation behavioral predictions, such as limited survey data, mismatching with behavioral theory, conflicting individual preferences, implicit and complex mental states, and intractable mental state-behavior mapping. Experiments on three post-wildfire survey datasets show an average of 20.47% performance improvement over traditional theory-informed behavioral models, with strong cross-event generalizability. Our complete code is publicly available at https://github.com/SusuXu-s-Lab/FLARE |
2025-02-24 | 24 pages, 9 figur |
| Scheherazade: Evaluating Chain-of-Thought Math Reasoning in LLMs with Chain-of-Problems Benchmarks are critical for measuring Large Language Model (LLM) reasoning capabilities. Some benchmarks have even become the de facto indicator of such capabilities. However, as LLM reasoning capabilities improve, existing widely-used benchmarks such as GSM8K marginally encapsulate model reasoning differentials - most state-of-the-art models for example achieve over 94% accuracy on the GSM8K dataset (paperwithcode, 2024). While constructing harder benchmarks is possible, their creation is often manual, expensive, and unscalable. As such, we present Scheherazade, an automated approach to produce large quantities of challenging mathematical reasoning benchmarks by logically chaining a small starting set of problems. We propose two different chaining methods, forward chaining and backward chaining, which include randomized branching techniques to generate complex reasoning problems. We apply Scheherazade on GSM8K to create GSM8K-Scheherazade and evaluate 3 frontier LLMs and OpenAI’s o1-preview on it. We show that while other frontier models’ performance declines precipitously at only a few questions chained, our evaluation suggests o1-preview’s performance persists, with the flagship OpenAI model the only one to perform better at backward reasoning. Our data and code are available at https://github.com/YoshikiTakashima/scheherazade-code-data. |
2025-02-24 | |
| Visualizing Uncertainty in Translation Tasks: An Evaluation of LLM Performance and Confidence Metrics Large language models (LLMs) are increasingly utilized for machine translation, yet their predictions often exhibit uncertainties that hinder interpretability and user trust. Effectively visualizing these uncertainties can enhance the usability of LLM outputs, particularly in contexts where translation accuracy is critical. This paper addresses two primary objectives: (1) providing users with token-level insights into model confidence and (2) developing a web-based visualization tool to quantify and represent translation uncertainties. To achieve these goals, we utilized the T5 model with the WMT19 dataset for translation tasks and evaluated translation quality using established metrics such as BLEU, METEOR, and ROUGE. We introduced three novel uncertainty quantification (UQ) metrics: (1) the geometric mean of token probabilities, (2) the arithmetic mean of token probabilities, and (3) the arithmetic mean of the kurtosis of token distributions. These metrics provide a simple yet effective framework for evaluating translation performance. Our analysis revealed a linear relationship between the traditional evaluation metrics and our UQ metrics, demonstrating the validity of our approach. Additionally, we developed an interactive web-based visualization that uses a color gradient to represent token confidence. This tool offers users a clear and intuitive understanding of translation quality while providing valuable insights into model performance. Overall, we show that our UQ metrics and visualization are both robust and interpretable, offering practical tools for evaluating and accessing machine translation systems. |
2025-02-24 | We would like to withdraw our paper due to an error in the experimental methodology, which impacts the validity of our results. The error specifically affects the analysis presented in the Discussion, where an incorrect experimental modeling step led to misleading interpretation |
| Semantics-Adaptive Activation Intervention for LLMs via Dynamic Steering Vectors Large language models (LLMs) have achieved remarkable performance across many tasks, yet aligning them with desired behaviors remains challenging. Activation intervention has emerged as an effective and economical method to modify the behavior of LLMs. Despite considerable interest in this area, current intervention methods exclusively employ a fixed steering vector to modify model activations, lacking adaptability to diverse input semantics. To address this limitation, we propose Semantics-Adaptive Dynamic Intervention (SADI), a novel method that constructs a dynamic steering vector to intervene model activations at inference time. More specifically, SADI utilizes activation differences in contrastive pairs to precisely identify critical elements of an LLM (i.e., attention heads, hidden states, and neurons) for targeted intervention. During inference, SADI dynamically steers model behavior by scaling element-wise activations based on the directions of input semantics. Experimental results show that SADI outperforms established baselines by substantial margins, improving task performance without training. SADI’s cost-effectiveness and generalizability across various LLM backbones and tasks highlight its potential as a versatile alignment technique. |
2025-02-24 | |
| Wearable Meets LLM for Stress Management: A Duoethnographic Study Integrating Wearable-Triggered Stressors and LLM Chatbots for Personalized Interventions We use a duoethnographic approach to study how wearable-integrated LLM chatbots can assist with personalized stress management, addressing the growing need for immediacy and tailored interventions. Two researchers interacted with custom chatbots over 22 days, responding to wearable-detected physiological prompts, recording stressor phrases, and using them to seek tailored interventions from their LLM-powered chatbots. They recorded their experiences in autoethnographic diaries and analyzed them during weekly discussions, focusing on the relevance, clarity, and impact of chatbot-generated interventions. Results showed that even though most events triggered by the wearable were meaningful, only one in five warranted an intervention. It also showed that interventions tailored with brief event descriptions were more effective than generic ones. By examining the intersection of wearables and LLM, this research contributes to developing more effective, user-centric mental health tools for real-time stress relief and behavior change. |
2025-02-24 | In CHI ‘25 Proceedings of the CHI Conference on Human Factors in Computing Systems Yokohama, Japan |
| Towards Robust Legal Reasoning: Harnessing Logical LLMs in Law Legal services rely heavily on text processing. While large language models (LLMs) show promise, their application in legal contexts demands higher accuracy, repeatability, and transparency. Logic programs, by encoding legal concepts as structured rules and facts, offer reliable automation, but require sophisticated text extraction. We propose a neuro-symbolic approach that integrates LLMs’ natural language understanding with logic-based reasoning to address these limitations. As a legal document case study, we applied neuro-symbolic AI to coverage-related queries in insurance contracts using both closed and open-source LLMs. While LLMs have improved in legal reasoning, they still lack the accuracy and consistency required for complex contract analysis. In our analysis, we tested three methodologies to evaluate whether a specific claim is covered under a contract: a vanilla LLM, an unguided approach that leverages LLMs to encode both the contract and the claim, and a guided approach that uses a framework for the LLM to encode the contract. We demonstrated the promising capabilities of LLM + Logic in the guided approach. |
2025-02-24 | |
| Craw4LLM: Efficient Web Crawling for LLM Pretraining Web crawl is a main source of large language models’ (LLMs) pretraining data, but the majority of crawled web pages are discarded in pretraining due to low data quality. This paper presents Craw4LLM, an efficient web crawling method that explores the web graph based on the preference of LLM pretraining. Specifically, it leverages the influence of a webpage in LLM pretraining as the priority score of the web crawler’s scheduler, replacing the standard graph connectivity based priority. Our experiments on a web graph containing 900 million webpages from a commercial search engine’s index demonstrate the efficiency of Craw4LLM in obtaining high-quality pretraining data. With just 21% URLs crawled, LLMs pretrained on Craw4LLM data reach the same downstream performances of previous crawls, significantly reducing the crawling waste and alleviating the burdens on websites. Our code is publicly available at https://github.com/cxcscmu/Craw4LLM. |
2025-02-24 | |
| ELMo-Tune-V2: LLM-Assisted Full-Cycle Auto-Tuning to Optimize LSM-Based Key-Value Stores Log-Structured Merge-tree-based Key-Value Store (LSM-KVS) is a foundational storage engine serving diverse modern workloads, systems, and applications. To suit varying use cases, LSM-KVS allows a vast configuration space that controls core parameters like compaction, flush, and cache sizes, each consuming a shared pool of CPU, Memory, and Storage resources. Navigating the LSM-KVS configuration space necessitates knowledge of the impact of each configuration on the expected workload and underlying hardware. Beyond expensive and time-intensive human-expert-based tuning, existing LSM-KVS tuning solutions focus on tuning with specific workload expectations while limited to a narrow subset of parameters. This paper introduces ELMo-Tune-V2, a framework that integrates Large Language Models (LLMs) at its foundation to demonstrate the potential of applying modern LLMs in data system optimization problems. ELMo-Tune-V2 leverages the contextual reasoning, cross-domain, and generative capabilities of LLMs to perform 1) self-navigated characterization and modeling of LSM-KVS workloads, 2) automatic tuning across a broad parameter space using cross-domain knowledge, and 3) real-time dynamic configuration adjustments for LSM-KVS. ELMo-Tune-V2 integrates three innovations: LLM-based workload synthesis for adaptive benchmark generation, feedback-driven iterative fine-tuning for configuration refinement, and real-time tuning to handle evolving workloads. Through detailed evaluation using RocksDB under several real-world applications across diverse scenarios, ELMo-Tune-V2 achieves performance improvements up to ~14X our YCSB benchmarks compared against default RocksDB configurations, and our end-to-end tests with upper-level applications, NebulaGraph and Kvrocks, demonstrate performance gains of 34% and 26%, respectively. |
2025-02-24 | |
| Hallucination Detection in LLMs Using Spectral Features of Attention Maps Large Language Models (LLMs) have demonstrated remarkable performance across various tasks but remain prone to hallucinations. Detecting hallucinations is essential for safety-critical applications, and recent methods leverage attention map properties to this end, though their effectiveness remains limited. In this work, we investigate the spectral features of attention maps by interpreting them as adjacency matrices of graph structures. We propose the $\text{LapEigvals}$ method, which utilises the top-$k$ eigenvalues of the Laplacian matrix derived from the attention maps as an input to hallucination detection probes. Empirical evaluations demonstrate that our approach achieves state-of-the-art hallucination detection performance among attention-based methods. Extensive ablation studies further highlight the robustness and generalisation of $\text{LapEigvals}$, paving the way for future advancements in the hallucination detection domain. |
2025-02-24 | Preprint, under review |
| LLM-Enhanced Dialogue Management for Full-Duplex Spoken Dialogue Systems Achieving full-duplex communication in spoken dialogue systems (SDS) requires real-time coordination between listening, speaking, and thinking. This paper proposes a semantic voice activity detection (VAD) module as a dialogue manager (DM) to efficiently manage turn-taking in full-duplex SDS. Implemented as a lightweight (0.5B) LLM fine-tuned on full-duplex conversation data, the semantic VAD predicts four control tokens to regulate turn-switching and turn-keeping, distinguishing between intentional and unintentional barge-ins while detecting query completion for handling user pauses and hesitations. By processing input speech in short intervals, the semantic VAD enables real-time decision-making, while the core dialogue engine (CDE) is only activated for response generation, reducing computational overhead. This design allows independent DM optimization without retraining the CDE, balancing interaction accuracy and inference efficiency for scalable, next-generation full-duplex SDS. |
2025-02-24 | In submission to INTERSPEECH 2025 |
| Empowering LLMs with Logical Reasoning: A Comprehensive Survey Large language models (LLMs) have achieved remarkable successes on various natural language tasks. However, recent studies have found that there are still significant challenges to the logical reasoning abilities of LLMs. This paper summarizes and categorizes the main challenges into two aspects: (1) Logical question answering, LLMs often fail to generate the correct answer within complex logical problem which requires sophisticated deductive, inductive or abductive reasoning given a collection of premises and constrains. (2) Logical consistency, LLMs are prone to producing responses contradicting themselves across different questions. For example, a state-of-the-art Macaw question-answering LLM answers Yes to both questions Is a magpie a bird? and Does a bird have wings? but answers No to Does a magpie have wings?. To facilitate this research direction, we comprehensively investigate the most cutting-edge methods and propose detailed taxonomies of these methods. Specifically, to accurately answer complex logic questions, previous methods can be categorized based on reliance on external solvers, prompts, pretraining, and fine-tuning. To avoid logical contradictions, we discuss concepts and solutions of various logical consistencies, including implication, negation, transitivity, factuality consistency, and their composites. In addition, we review commonly used benchmark datasets and evaluation metrics, and discuss promising research directions, such as extensions to modal logic to account for uncertainty, and efficient algorithms satisfying multiple logical consistencies simultaneously. |
2025-02-24 | |
| The Lottery LLM Hypothesis, Rethinking What Abilities Should LLM Compression Preserve? Motivated by reducing the computational and storage costs of LLMs, model compression and KV cache compression have attracted much attention from researchers. However, current methods predominantly emphasize maintaining the performance of compressed LLMs, as measured by perplexity or simple accuracy on tasks of common sense knowledge QA and basic arithmetic reasoning. In this blog, we present a brief review of recent advancements in LLMs related to retrieval-augmented generation, multi-step reasoning, external tools, and computational expressivity, all of which substantially enhance LLM performance. Then, we propose a lottery LLM hypothesis suggesting that for a given LLM and task, there exists a smaller lottery LLM capable of producing the same performance as the original LLM with the assistance of multi-step reasoning and external tools. Based on the review of current progress in LLMs, we discuss and summarize the essential capabilities that the lottery LLM and KV cache compression must possess, which are currently overlooked in existing methods. |
2025-02-24 | |
| ConvoyLLM: Dynamic Multi-Lane Convoy Control Using LLMs This paper proposes a novel method for multi-lane convoy formation control that uses large language models (LLMs) to tackle coordination challenges in dynamic highway environments. Each connected and autonomous vehicle in the convoy uses a knowledge-driven approach to make real-time adaptive decisions based on various scenarios. Our method enables vehicles to dynamically perform tasks, including obstacle avoidance, convoy joining/leaving, and escort formation switching, all while maintaining the overall convoy structure. We design a Interlaced formation control strategy based on locally dynamic distributed graphs, ensuring the convoy remains stable and flexible. We conduct extensive experiments in the SUMO simulation platform across multiple traffic scenarios, and the results demonstrate that the proposed method is effective, robust, and adaptable to dynamic environments. The code is available at: https://github.com/chuduanfeng/ConvoyLLM. |
2025-02-24 | |
| Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned. Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment. In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It’s important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work. |
2025-02-24 | 10 pages, 9 figur |
| COSMOS: A Hybrid Adaptive Optimizer for Memory-Efficient Training of LLMs Large Language Models (LLMs) have demonstrated remarkable success across various domains, yet their optimization remains a significant challenge due to the complex and high-dimensional loss landscapes they inhabit. While adaptive optimizers such as AdamW are widely used, they suffer from critical limitations, including an inability to capture interdependencies between coordinates and high memory consumption. Subsequent research, exemplified by SOAP, attempts to better capture coordinate interdependence but incurs greater memory overhead, limiting scalability for massive LLMs. An alternative approach aims to reduce memory consumption through low-dimensional projection, but this leads to substantial approximation errors, resulting in less effective optimization (e.g., in terms of per-token efficiency). In this paper, we propose COSMOS, a novel hybrid optimizer that leverages the varying importance of eigensubspaces in the gradient matrix to achieve memory efficiency without compromising optimization performance. The design of COSMOS is motivated by our empirical insights and practical considerations. Specifically, COSMOS applies SOAP to the leading eigensubspace, which captures the primary optimization dynamics, and MUON to the remaining eigensubspace, which is less critical but computationally expensive to handle with SOAP. This hybrid strategy significantly reduces memory consumption while maintaining robust optimization performance, making it particularly suitable for massive LLMs. Numerical experiments on various datasets and transformer architectures are provided to demonstrate the effectiveness of COSMOS. Our code is available at https://github.com/lliu606/COSMOS. |
2025-02-24 | 23 pages, 9 figures, 6 tab |
| HYBRIDMIND: Meta Selection of Natural Language and Symbolic Language for Enhanced LLM Reasoning LLMs approach logical and mathematical reasoning through natural or symbolic languages. While natural language offers human-accessible flexibility but suffers from ambiguity, symbolic reasoning provides precise, machine-executable inferences at the cost of strict domain constraints. We introduce HYBRIDMIND, an adaptive strategy that selects the optimal reasoning approach for each reasoning problem. Through extensive experiments, we evaluate both prompting-based approaches with state-of-the-art LLMs and fine-tuned open-source models. We find that fine-tuning LLaMA-3.1-8B-Instruct as a meta-selector outperforms GPT-4o’s natural language reasoning by 4.4\% on FOLIO and 1.3\% on MATH. More notably, using GPT-3.5-turbo as a prompted meta-selector yields a 10\% improvement on FOLIO’s challenging subset compared to GPT-4o. We will release our code and data to support future research. |
2025-02-24 | |
| Aligned at the Start: Conceptual Groupings in LLM Embeddings This paper shifts focus to the often-overlooked input embeddings - the initial representations fed into transformer blocks. Using fuzzy graph, k-nearest neighbor (k-NN), and community detection, we analyze embeddings from diverse LLMs, finding significant categorical community structure aligned with predefined concepts and categories aligned with humans. We observe these groupings exhibit within-cluster organization (such as hierarchies, topological ordering, etc.), hypothesizing a fundamental structure that precedes contextual processing. To further investigate the conceptual nature of these groupings, we explore cross-model alignments across different LLM categories within their input embeddings, observing a medium to high degree of alignment. Furthermore, provide evidence that manipulating these groupings can play a functional role in mitigating ethnicity bias in LLM tasks. |
2025-02-24 | |
| Non-Halting Queries: Exploiting Fixed Points in LLMs We introduce a new vulnerability that exploits fixed points in autoregressive models and use it to craft queries that never halt. More precisely, for non-halting queries, the LLM never samples the end-of-string token |
2025-02-24 | |
| Time series forecasting based on optimized LLM for fault prediction in distribution power grid insulators Surface contamination on electrical grid insulators leads to an increase in leakage current until an electrical discharge occurs, which can result in a power system shutdown. To mitigate the possibility of disruptive faults resulting in a power outage, monitoring contamination and leakage current can help predict the progression of faults. Given this need, this paper proposes a hybrid deep learning (DL) model for predicting the increase in leakage current in high-voltage insulators. The hybrid structure considers a multi-criteria optimization using tree-structured Parzen estimation, an input stage filter for signal noise attenuation combined with a large language model (LLM) applied for time series forecasting. The proposed optimized LLM outperforms state-of-the-art DL models with a root-mean-square error equal to 2.24$\times10^{-4}$ for a short-term horizon and 1.21$\times10^{-3}$ for a medium-term horizon. |
2025-02-24 | |
| Mutual Reinforcement of LLM Dialogue Synthesis and Summarization Capabilities for Few-Shot Dialogue Summarization In this work, we propose Mutual Reinforcing Data Synthesis (MRDS) within LLMs to improve few-shot dialogue summarization task. Unlike prior methods that require external knowledge, we mutually reinforce the LLM's dialogue synthesis and summarization capabilities, allowing them to complement each other during training and enhance overall performances. The dialogue synthesis capability is enhanced by directed preference optimization with preference scoring from summarization capability. The summarization capability is enhanced by the additional high quality dialogue-summary paired data produced by the dialogue synthesis capability. By leveraging the proposed MRDS mechanism, we elicit the internal knowledge of LLM in the format of synthetic data, and use it to augment the few-shot real training dataset. Empirical results demonstrate that our method improves dialogue summarization, achieving a 1.5% increase in ROUGE scores and a 0.3% improvement in BERT scores in few-shot settings. Furthermore, our method attains the highest average scores in human evaluations, surpassing both the pre-trained models and the baselines fine-tuned solely for summarization tasks. |
2025-02-24 | NAACL 2025 Finding |
| Child vs. machine language learning: Can the logical structure of human language unleash LLMs? We argue that human language learning proceeds in a manner that is different in nature from current approaches to training LLMs, predicting a difference in learning biases. We then present evidence from German plural formation by LLMs that confirm our hypothesis that even very powerful implementations produce results that miss aspects of the logic inherent to language that humans have no problem with. We conclude that attention to the different structures of human language and artificial neural networks is likely to be an avenue to improve LLM performance. |
2025-02-24 | ISCA/ITG Workshop on Diversity in Large Speech and Language Mo |
| KVTuner: Sensitivity-Aware Layer-wise Mixed Precision KV Cache Quantization for Efficient and Nearly Lossless LLM Inference KV cache quantization can improve Large Language Models (LLMs) inference throughput and latency in long contexts and large batch-size scenarios while preserving LLMs effectiveness. However, current methods have three unsolved issues: overlooking layer-wise sensitivity to KV cache quantization, high overhead of online fine-grained decision-making, and low flexibility to different LLMs and constraints. Therefore, we thoroughly analyze the inherent correlation of layer-wise transformer attention patterns to KV cache quantization errors and study why key cache is more important than value cache for quantization error reduction. We further propose a simple yet effective framework KVTuner to adaptively search for the optimal hardware-friendly layer-wise KV quantization precision pairs for coarse-grained KV cache with multi-objective optimization and directly utilize the offline searched configurations during online inference. To reduce the computational cost of offline calibration, we utilize the intra-layer KV precision pair pruning and inter-layer clustering to reduce the search space. Experimental results show that we can achieve nearly lossless 3.25-bit mixed precision KV cache quantization for LLMs like Llama-3.1-8B-Instruct and 4.0-bit for sensitive models like Qwen2.5-7B-Instruct on mathematical reasoning tasks. The maximum inference throughput can be improved by 38.3% compared with KV8 quantization over various context lengths. Our code and searched configurations are available at https://github.com/cmd2001/KVTuner. |
2025-02-24 | 36 pages. Code: https://github.com/cmd2001/KVTuner |
| Capability Instruction Tuning: A New Paradigm for Dynamic LLM Routing Large Language Models (LLMs) have demonstrated human-like instruction-following abilities, particularly those exceeding 100 billion parameters. The combined capability of some smaller, resource-friendly LLMs can address most of the instructions that larger LLMs excel at. In this work, we explore how to route the best-performing LLM for each instruction to achieve better overall performance. We develop a new paradigm, constructing capability instructions with model capability representation, user instruction, and performance inquiry prompts to assess the performance. To learn from capability instructions, we introduce a new end-to-end framework called Model Selection with Aptitude Test (Model-SAT), which generates positive and negative samples based on what different models perform well or struggle with. Model-SAT uses a model capability encoder that extends its model representation to a lightweight LLM. Our experiments show that Model-SAT understands the performance dimensions of candidate models and provides the probabilities of their capability to handle various instructions. Additionally, during deployment, a new model can quickly infer its aptitude test results across 50 tasks, each with 20 shots. Model-SAT performs state-of-the-art model routing without candidate inference and in real-world new model-released scenarios. The code is available at https://github.com/Now-Join-Us/CIT-LLM-Routing |
2025-02-24 | AAAI 2025; Project Page: https://cit-llm-routing.github.io |
| PersonalLLM: Tailoring LLMs to Individual Preferences As LLMs become capable of complex tasks, there is growing potential for personalized interactions tailored to the subtle and idiosyncratic preferences of the user. We present a public benchmark, PersonalLLM, focusing on adapting LLMs to provide maximal benefits for a particular user. Departing from existing alignment benchmarks that implicitly assume uniform preferences, we curate open-ended prompts paired with many high-quality answers over which users would be expected to display heterogeneous latent preferences. Instead of persona-prompting LLMs based on high-level attributes (e.g., user’s race or response length), which yields homogeneous preferences relative to humans, we develop a method that can simulate a large user base with diverse preferences from a set of pre-trained reward models. Our dataset and generated personalities offer an innovative testbed for developing personalization algorithms that grapple with continual data sparsity–few relevant feedback from the particular user–by leveraging historical data from other (similar) users. We explore basic in-context learning and meta-learning baselines to illustrate the utility of PersonalLLM and highlight the need for future methodological development. Our dataset is available at https://huggingface.co/datasets/namkoong-lab/PersonalLLM |
2025-02-24 | 28 pages, 6 figur |
| Unveiling Downstream Performance Scaling of LLMs: A Clustering-Based Perspective The rapid advancements in computing dramatically increase the scale and cost of training Large Language Models (LLMs). Accurately predicting downstream task performance prior to model training is crucial for efficient resource allocation, yet remains challenging due to two primary constraints: (1) the “emergence phenomenon”, wherein downstream performance metrics become meaningful only after extensive training, which limits the ability to use smaller models for prediction; (2) Uneven task difficulty distributions and the absence of consistent scaling laws, resulting in substantial metric variability. Existing performance prediction methods suffer from limited accuracy and reliability, thereby impeding the assessment of potential LLM capabilities. To address these challenges, we propose a Clustering-On-Difficulty (COD) downstream performance prediction framework. COD first constructs a predictable support subset by clustering tasks based on difficulty features, strategically excluding non-emergent and non-scalable clusters. The scores on the selected subset serve as effective intermediate predictors of downstream performance on the full evaluation set. With theoretical support, we derive a mapping function that transforms performance metrics from the predictable subset to the full evaluation set, thereby ensuring accurate extrapolation of LLM downstream performance. The proposed method has been applied to predict performance scaling for a 70B LLM, providing actionable insights for training resource allocation and assisting in monitoring the training process. Notably, COD achieves remarkable predictive accuracy on the 70B LLM by leveraging an ensemble of small models, demonstrating an absolute mean deviation of 1.36% across eight important LLM evaluation benchmarks. |
2025-02-24 | 21 pages,6 figur |
| Making LLMs Reason? The Intermediate Language Problem in Neurosymbolic Approaches Logical reasoning tasks manifest themselves as a challenge to Large Language Models (LLMs). Neurosymbolic approaches use LLMs to translate logical reasoning problems formulated in natural language into a formal intermediate language. Subsequently, the usage of symbolic reasoners yields reliable solving thereof. However, LLMs often fail in translation due to poorly chosen intermediate languages. We introduce the intermediate language problem, which is the problem of choosing a suitable formal language representation for neurosymbolic approaches. Theoretically, we argue that its origins lie in the inability of LLMs to distinguish syntax from semantics and the relative independence of the problem from its representation. We showcase its existence experimentally by contrasting two intermediate languages, Answer Set Programming and the Python Knowledge Engine. In addition, we demonstrate the effects of varying degrees of supplementary context information. Our results show a maximum difference in overall-accuracy of 53.20% and 49.26% in execution-accuracy. When using the GPT4o-mini LLM we beat the state-of-the-art in overall-accuracy on the ProntoQA dataset by 21.20% and by 50.50% on the ProofWriter dataset. |
2025-02-24 | |
| CoT-UQ: Improving Response-wise Uncertainty Quantification in LLMs with Chain-of-Thought Large language models (LLMs) excel in many tasks but struggle to accurately quantify uncertainty in their generated responses. This limitation makes it challenging to detect misinformation and ensure reliable decision-making. Existing uncertainty quantification (UQ) methods for LLMs are primarily prompt-wise rather than response-wise, often requiring multiple response samples, which incurs high computational costs. Moreover, LLMs have been shown to be overconfident, particularly when using reasoning steps to derive their answers. In this work, we propose CoT-UQ, a response-wise UQ framework that integrates LLMs’ inherent reasoning capabilities through Chain-of-Thought (CoT) into the UQ process. CoT-UQ captures critical information during inference by extracting keywords from each reasoning step and assessing their importance to the final answer. This key reasoning information is then aggregated to produce a final uncertainty estimate. We conduct extensive experiments based on LLaMA Family with model sizes varying from 8B to 13B across logical and mathematical reasoning tasks. Experimental results demonstrate that CoT-UQ significantly outperforms existing UQ methods, achieving an average improvement of 5.9% AUROC compared to current UQ methods. The code is available at: https://github.com/ZBox1005/CoT-UQ. |
2025-02-24 | |
| Evaluating Expert Contributions in a MoE LLM for Quiz-Based Tasks Recently, Large Language Models (LLMs) with Mixture of Experts (MoE) layers have gained significant attention. Currently, state-of-the-art LLMs utilize this architecture. There is a substantial amount of research on how to train such models and how to select hyperparameters for this architecture. However, there is a lack of studies focusing on post-evaluation analysis of MoE layer properties. In this paper, we take a first step toward closing this gap by evaluating expert contributions on the quiz-based MMLU benchmark. We show that most experts were never activated during inference on this benchmark. Additionally, the output distribution of gating networks is much closer to uniform than sparse. Finally, we demonstrate that the average performance of some experts within the same layer varies significantly. |
2025-02-24 | preprint, short paper |
| Multilingual LLMs Inherently Reward In-Language Time-Sensitive Semantic Alignment for Low-Resource Languages The unwavering disparity in labeled resources between resource-rich languages and those considered low-resource remains a significant impediment for Large Language Models (LLMs). Recent strides in cross-lingual in-context learning (X-ICL), mainly through semantically aligned examples retrieved from multilingual pre-trained transformers, have shown promise in mitigating this issue. However, our investigation reveals that LLMs intrinsically reward in-language semantically aligned cross-lingual instances over direct cross-lingual semantic alignments, with a pronounced disparity in handling time-sensitive queries in the X-ICL setup. Such queries demand sound temporal reasoning ability from LLMs, yet the advancements have predominantly focused on English. This study aims to bridge this gap by improving temporal reasoning capabilities in low-resource languages. To this end, we introduce mTEMPREASON, a temporal reasoning dataset aimed at the varied degrees of low-resource languages and propose Cross-Lingual Time-Sensitive Semantic Alignment (CLiTSSA), a novel method to improve temporal reasoning in these contexts. To facilitate this, we construct an extension of mTEMPREASON comprising pairs of parallel cross-language temporal queries along with their anticipated in-language semantic similarity scores. Our empirical evidence underscores the superior performance of CLiTSSA compared to established baselines across three languages – Romanian, German, and French, encompassing three temporal tasks and including a diverse set of four contemporaneous LLMs. This marks a significant step forward in addressing resource disparity in the context of temporal reasoning across languages. |
2025-02-24 | |
| DeltaLLM: Compress LLMs with Low-Rank Deltas between Shared Weights We introduce DeltaLLM, a new post-training compression technique to reduce the memory footprint of LLMs. We propose an alternative way of structuring LLMs with weight sharing between layers in subsequent Transformer blocks, along with additional low-rank difference matrices between them. For training, we adopt the progressing module replacement method and show that the lightweight training of the low-rank modules with approximately 30M-40M tokens is sufficient to achieve performance on par with LLMs of comparable sizes trained from scratch. We release the resultant models, DeltaLLAMA and DeltaPHI, with a 12% parameter reduction, retaining 90% of the performance of the base Llama and Phi models on common knowledge and reasoning benchmarks. Our method also outperforms compression techniques JointDrop, LaCo, ShortGPT and SliceGPT with the same number of parameters removed. For example, DeltaPhi 2.9B with a 24% reduction achieves similar average zero-shot accuracies as recovery fine-tuned SlicedPhi 3.3B with a 12% reduction, despite being approximately 400M parameters smaller with no fine-tuning applied. This work provides new insights into LLM architecture design and compression methods when storage space is critical. |
2025-02-24 | |
| Round Attention: A Novel Round-Level Attention Mechanism to Accelerate LLM Inference The increasing context window size in large language models (LLMs) has improved their ability to handle complex, long-text tasks. However, as the conversation rounds continue, it is required to store a large amount of KV cache in GPU memory, which significantly affects the efficiency and even availability of the model serving systems. This paper analyzes dialogue data from real users and discovers that the LLM inference manifests a watershed layer, after which the distribution of round-level attention shows notable similarity. We propose Round Attention, a novel round-level attention mechanism that only recalls and computes the KV cache of the most relevant rounds. The experiments show that our method saves 55\% memory usage without compromising model performance. |
2025-02-24 | |
| CodeSwift: Accelerating LLM Inference for Efficient Code Generation Code generation is a latency-sensitive task that demands high timeliness, but the autoregressive decoding mechanism of Large Language Models (LLMs) leads to poor inference efficiency. Existing LLM inference acceleration methods mainly focus on standalone functions using only built-in components. Moreover, they treat code like natural language sequences, ignoring its unique syntax and semantic characteristics. As a result, the effectiveness of these approaches in code generation tasks remains limited and fails to align with real-world programming scenarios. To alleviate this issue, we propose CodeSwift, a simple yet highly efficient inference acceleration approach specifically designed for code generation, without comprising the quality of the output. CodeSwift constructs a multi-source datastore, providing access to both general and project-specific knowledge, facilitating the retrieval of high-quality draft sequences. Moreover, CodeSwift reduces retrieval cost by controlling retrieval timing, and enhances efficiency through parallel retrieval and a context- and LLM preference-aware cache. Experimental results show that CodeSwift can reach up to 2.53x and 2.54x speedup compared to autoregressive decoding in repository-level and standalone code generation tasks, respectively, outperforming state-of-the-art inference acceleration approaches by up to 88%. |
2025-02-24 | |
| WildFrame: Comparing Framing in Humans and LLMs on Naturally Occurring Texts Humans are influenced by how information is presented, a phenomenon known as the framing effect. Previous work has shown that LLMs may also be susceptible to framing but has done so on synthetic data and did not compare to human behavior. We introduce WildFrame, a dataset for evaluating LLM responses to positive and negative framing, in naturally-occurring sentences, and compare humans on the same data. WildFrame consists of 1,000 texts, first selecting real-world statements with clear sentiment, then reframing them in either positive or negative light, and lastly, collecting human sentiment annotations. By evaluating eight state-of-the-art LLMs on WildFrame, we find that all models exhibit framing effects similar to humans ($r\geq0.57$), with both humans and models being more influenced by positive rather than negative reframing. Our findings benefit model developers, who can either harness framing or mitigate its effects, depending on the downstream application. |
2025-02-24 | |
| CKnowEdit: A New Chinese Knowledge Editing Dataset for Linguistics, Facts, and Logic Error Correction in LLMs Chinese, as a linguistic system rich in depth and complexity, is characterized by distinctive elements such as ancient poetry, proverbs, idioms, and other cultural constructs. However, current Large Language Models (LLMs) face limitations in these specialized domains, highlighting the need for the development of comprehensive datasets that can assess, continuously update, and progressively improve these culturally-grounded linguistic competencies through targeted training optimizations. To address this gap, we introduce CKnowEdit, the first-ever Chinese knowledge editing dataset designed to correct linguistic, factual, and logical errors in LLMs. We collect seven types of knowledge from a wide range of sources, including classical texts, idioms, and content from Baidu Tieba Ruozhiba, taking into account the unique polyphony, antithesis, and logical structures inherent in the Chinese language. By analyzing this dataset, we highlight the challenges current LLMs face in mastering Chinese. Furthermore, our evaluation of state-of-the-art knowledge editing techniques reveals opportunities to advance the correction of Chinese knowledge. Code and dataset are available at https://github.com/zjunlp/EasyEdit. |
2025-02-24 | Ongoing work; project website is available at https://zjunlp.github.io/project/CKnowEdit code and dataset are available at https://github.com/zjunlp/EasyE |
| BFS-Prover: Scalable Best-First Tree Search for LLM-based Automatic Theorem Proving Recent advancements in large language models (LLMs) have spurred growing interest in automatic theorem proving using Lean4, where effective tree search methods are crucial for navigating the underlying large proof search spaces. While the existing approaches primarily rely on value functions and/or Monte Carlo Tree Search (MCTS), the potential of simpler methods like Best-First Tree Search (BFS) remains underexplored. In this paper, we investigate whether BFS can achieve competitive performance in large-scale theorem proving tasks. We present BFS-Prover, a scalable expert iteration framework, featuring three key innovations. First, we implement strategic data filtering at each expert iteration round, excluding problems solvable via beam search node expansion to focus on harder cases. Second, we improve the sample efficiency of BFS through Direct Preference Optimization (DPO) applied to state-tactic pairs automatically annotated with compiler error feedback, refining the LLM’s policy to prioritize productive expansions. Third, we employ length normalization in BFS to encourage exploration of deeper proof paths. BFS-Prover achieves a state-of-the-art score of $72.95\%$ on the MiniF2F test set and therefore challenges the perceived necessity of complex tree search methods, demonstrating that BFS can achieve competitive performance when properly scaled. To facilitate further research and development in this area, we have open-sourced our model at https://huggingface.co/bytedance-research/BFS-Prover. |
2025-02-24 | |
| Understanding the Uncertainty of LLM Explanations: A Perspective Based on Reasoning Topology Understanding the uncertainty in large language model (LLM) explanations is important for evaluating their faithfulness and reasoning consistency, and thus provides insights into the reliability of LLM’s output regarding a question. In this work, we propose a novel framework that quantifies uncertainty in LLM explanations through a reasoning topology perspective. By designing a structural elicitation strategy, we guide the LLMs to frame the explanations of an answer into a graph topology. This process decomposes the explanations into the knowledge related sub-questions and topology-based reasoning structures, which allows us to quantify uncertainty not only at the semantic level but also from the reasoning path. It further brings convenience to assess knowledge redundancy and provide interpretable insights into the reasoning process. Our method offers a systematic way to interpret the LLM reasoning, analyze limitations, and provide guidance for enhancing robustness and faithfulness. This work pioneers the use of graph-structured uncertainty measurement in LLM explanations and demonstrates the potential of topology-based quantification. |
2025-02-24 | 15 pages, 6 figur |
| Predicting Liquidity-Aware Bond Yields using Causal GANs and Deep Reinforcement Learning with LLM Evaluation Financial bond yield forecasting is challenging due to data scarcity, nonlinear macroeconomic dependencies, and evolving market conditions. In this paper, we propose a novel framework that leverages Causal Generative Adversarial Networks (CausalGANs) and Soft Actor-Critic (SAC) reinforcement learning (RL) to generate high-fidelity synthetic bond yield data for four major bond categories (AAA, BAA, US10Y, Junk). By incorporating 12 key macroeconomic variables, we ensure statistical fidelity by preserving essential market properties. To transform this market dependent synthetic data into actionable insights, we employ a finetuned Large Language Model (LLM) Qwen2.5-7B that generates trading signals (BUY/HOLD/SELL), risk assessments, and volatility projections. We use automated, human and LLM evaluations, all of which demonstrate that our framework improves forecasting performance over existing methods, with statistical validation via predictive accuracy, MAE evaluation(0.103%), profit/loss evaluation (60% profit rate), LLM evaluation (3.37/5) and expert assessments scoring 4.67 out of 5. The reinforcement learning-enhanced synthetic data generation achieves the least Mean Absolute Error of 0.103, demonstrating its effectiveness in replicating real-world bond market dynamics. We not only enhance data-driven trading strategies but also provides a scalable, high-fidelity synthetic financial data pipeline for risk & volatility management and investment decision-making. This work establishes a bridge between synthetic data generation, LLM driven financial forecasting, and language model evaluation, contributing to AI-driven financial decision-making. |
2025-02-24 | |
| Reinforcement Learning Enhanced LLMs: A Survey Reinforcement learning (RL) enhanced large language models (LLMs), particularly exemplified by DeepSeek-R1, have exhibited outstanding performance. Despite the effectiveness in improving LLM capabilities, its implementation remains highly complex, requiring complex algorithms, reward modeling strategies, and optimization techniques. This complexity poses challenges for researchers and practitioners in developing a systematic understanding of RL-enhanced LLMs. Moreover, the absence of a comprehensive survey summarizing existing research on RL-enhanced LLMs has limited progress in this domain, hindering further advancements. In this work, we are going to make a systematic review of the most up-to-date state of knowledge on RL-enhanced LLMs, attempting to consolidate and analyze the rapidly growing research in this field, helping researchers understand the current challenges and advancements. Specifically, we (1) detail the basics of RL; (2) introduce popular RL-enhanced LLMs; (3) review researches on two widely-used reward model-based RL techniques: Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning from AI Feedback (RLAIF); and (4) explore Direct Preference Optimization (DPO), a set of methods that bypass the reward model to directly use human preference data for aligning LLM outputs with human expectations. We will also point out current challenges and deficiencies of existing methods and suggest some avenues for further improvements. Project page of this work can be found at https://github.com/ShuheWang1998/Reinforcement-Learning-Enhanced-LLMs-A-Survey. |
2025-02-24 | |
| DynamicNER: A Dynamic, Multilingual, and Fine-Grained Dataset for LLM-based Named Entity Recognition With the advancement of Large Language Models (LLMs), more and more researchers apply LLMs for Named Entity Recognition (NER) methods, bringing vitality to this classical Natural Language Processing task. However, existing datasets are designed for traditional machine learning methods, inadequate for LLM-based methods in terms of corpus selection, entity categorization, and design logic. This limitation leads to less effective evaluation and model fine-tuning. To address this issue, we propose DynamicNER, the first NER dataset specifically designed for LLMs and with dynamic categorization, transcending the limitations of fixed categorization in existing datasets. It is also multi-lingual and multi-granular, covering 8 languages and 155 entity types, with corpus spanning multiple specialized domains. Furthermore, in response to the limitations demonstrated by existing LLM-based methods during DynamicNER testing, we develop CascadeNER, a novel NER method based on a two-stage strategy and lightweight LLMs, addressing the problems in current methods. Experiments show that DynamicNER is an effective benchmark for LLM-based NER methods, and CascadeNER outperforms existing methods with fewer computational resources. Our work is opened at https://github.com/CascadeNER/CascadeNER. |
2025-02-24 | |
| Make LLM Inference Affordable to Everyone: Augmenting GPU Memory with NDP-DIMM The billion-scale Large Language Models (LLMs) need deployment on expensive server-grade GPUs with large-storage HBMs and abundant computation capability. As LLM-assisted services become popular, achieving cost-effective LLM inference on budget-friendly hardware becomes the trend. Extensive researches relocate LLM parameters from expensive GPUs to host memory. However, the restricted bandwidth between the host and GPU memory limits the inference performance. This work introduces Hermes, a budget-friendly system that leverages the near-data processing (NDP) within commodity DRAM DIMMs to enhance the performance of a single consumer-grade GPU, achieving efficient LLM inference. The inherent activation sparsity in LLMs naturally divides weight parameters into two categories, termed hot" andcold” neurons, respectively. Hot neurons, which consist of only approximately 20\% of all weight parameters, account for 80\% of the total computational load, while cold neurons make up the other 80\% of parameters but are responsible for just 20\% of the computational load. Therefore, we propose a heterogeneous computing strategy: mapping hot neurons to a single computation-efficient GPU, while offloading cold neurons to NDP-DIMMs, which offer large memory size but limited computation capabilities. Meanwhile, the dynamic nature of activation sparsity needs a real-time partition of hot/cold neurons and adaptive remapping of cold neurons across multiple NDP-DIMM modules. Therefore, we introduce a lightweight predictor optimizing real-time neuron partition and adjustment between GPU and NDP-DIMMs. We also utilize a window-based online scheduling mechanism to maintain load balance among NDP-DIMM modules. Hermes facilitates the deployment of LLaMA2-70B on consumer-grade hardware at 13.75 tokens/s and realizes an average 75.24$\times$ speedup over the state-of-the-art offloading-based inference system. |
2025-02-24 | 15 pages, 17 figures, accepted by HPCA 2025 |
| User Experience with LLM-powered Conversational Recommendation Systems: A Case of Music Recommendation The advancement of large language models (LLMs) now allows users to actively interact with conversational recommendation systems (CRS) and build their own personalized recommendation services tailored to their unique needs and goals. This experience offers users a significantly higher level of controllability compared to traditional RS, enabling an entirely new dimension of recommendation experiences. Building on this context, this study explored the unique experiences that LLM-powered CRS can provide compared to traditional RS. Through a three-week diary study with 12 participants using custom GPTs for music recommendations, we found that LLM-powered CRS can (1) help users clarify implicit needs, (2) support unique exploration, and (3) facilitate a deeper understanding of musical preferences. Based on these findings, we discuss the new design space enabled by LLM-powered CRS and highlight its potential to support more personalized, user-driven recommendation experiences. |
2025-02-24 | |
| EoRA: Training-free Compensation for Compressed LLM with Eigenspace Low-Rank Approximation In this work, we re-formulate the model compression problem into the customized compensation problem: Given a compressed model, we aim to introduce residual low-rank paths to compensate for compression errors under customized requirements from users (e.g., tasks, compression ratios), resulting in greater flexibility in balancing accuracy and overhead(inference and model size) without being bound to fixed compression formats. However, naively applying SVD to derive residual paths causes suboptimal utilization of the low-rank representation capacity. Instead, we propose Training-free Eigenspace Low-Rank Approximation (EoRA), a method that directly minimizes compression-induced errors without requiring gradient-based training, achieving fast optimization in minutes using a small amount of calibration data. EoRA projects compression errors into the eigenspace of input activations, leveraging eigenvalues to effectively prioritize the reconstruction of high-importance error components. Moreover, EoRA can be seamlessly integrated with fine-tuning and quantization to further improve effectiveness and efficiency. EoRA consistently outperforms previous methods in compensating errors for compressed LLaMA2/3 models on various tasks, such as language generation, commonsense reasoning, and math reasoning tasks (e.g., 31.31%/12.88% and 9.69% improvements on ARC-Easy/ARC-Challenge and MathQA when compensating LLaMA3-8B that is quantized to 4-bit and pruned to 2:4 sparsity). EoRA offers a scalable, training-free solution to compensate for compression errors, making it a powerful tool to deploy LLMs more flexibly. Code is available at https://github.com/NVlabs/EoRA. |
2025-02-24 | |
| FilterLLM: Text-To-Distribution LLM for Billion-Scale Cold-Start Recommendation Large Language Model (LLM)-based cold-start recommendation systems continue to face significant computational challenges in billion-scale scenarios, as they follow a “Text-to-Judgment” paradigm. This approach processes user-item content pairs as input and evaluates each pair iteratively. To maintain efficiency, existing methods rely on pre-filtering a small candidate pool of user-item pairs. However, this severely limits the inferential capabilities of LLMs by reducing their scope to only a few hundred pre-filtered candidates. To overcome this limitation, we propose a novel “Text-to-Distribution” paradigm, which predicts an item’s interaction probability distribution for the entire user set in a single inference. Specifically, we present FilterLLM, a framework that extends the next-word prediction capabilities of LLMs to billion-scale filtering tasks. FilterLLM first introduces a tailored distribution prediction and cold-start framework. Next, FilterLLM incorporates an efficient user-vocabulary structure to train and store the embeddings of billion-scale users. Finally, we detail the training objectives for both distribution prediction and user-vocabulary construction. The proposed framework has been deployed on the Alibaba platform, where it has been serving cold-start recommendations for two months, processing over one billion cold items. Extensive experiments demonstrate that FilterLLM significantly outperforms state-of-the-art methods in cold-start recommendation tasks, achieving over 30 times higher efficiency. Furthermore, an online A/B test validates its effectiveness in billion-scale recommendation systems. |
2025-02-24 | 12 pag |
| Unlocking Scientific Concepts: How Effective Are LLM-Generated Analogies for Student Understanding and Classroom Practice? Teaching scientific concepts is essential but challenging, and analogies help students connect new concepts to familiar ideas. Advancements in large language models (LLMs) enable generating analogies, yet their effectiveness in education remains underexplored. In this paper, we first conducted a two-stage study involving high school students and teachers to assess the effectiveness of LLM-generated analogies in biology and physics through a controlled in-class test and a classroom field study. Test results suggested that LLM-generated analogies could enhance student understanding particularly in biology, but require teachers’ guidance to prevent over-reliance and overconfidence. Classroom experiments suggested that teachers could refine LLM-generated analogies to their satisfaction and inspire new analogies from generated ones, encouraged by positive classroom feedback and homework performance boosts. Based on findings, we developed and evaluated a practical system to help teachers generate and refine teaching analogies. We discussed future directions for developing and evaluating LLM-supported teaching and learning by analogy. |
2025-02-24 | 19 pages, conditionally accepted by CHI 2025 |
| GSQ-Tuning: Group-Shared Exponents Integer in Fully Quantized Training for LLMs On-Device Fine-tuning Large Language Models (LLMs) fine-tuning technologies have achieved remarkable results. However, traditional LLM fine-tuning approaches face significant challenges: they require large Floating Point (FP) computation, raising privacy concerns when handling sensitive data, and are impractical for resource-constrained edge devices. While Parameter-Efficient Fine-Tuning (PEFT) techniques reduce trainable parameters, their reliance on floating-point arithmetic creates fundamental incompatibilities with edge hardware. In this work, we introduce a novel framework for on-device LLM fine-tuning that eliminates the need for floating-point operations in both inference and training, named GSQ-Tuning. At its core is the Group-Shared Exponents Integer format, which efficiently represents model parameters in integer format using shared exponents among parameter groups. When combined with LoRA-like adapters, this enables fully integer-based fine-tuning that is both memory and compute efficient. We demonstrate that our approach achieves accuracy comparable to BF16-based fine-tuning while significantly reducing 1.85x memory usage. Moreover, compared to FP8, our method can reduce 5x power consumption and 11x chip area with same performance, making large-scale model adaptation feasible on edge devices. |
2025-02-24 | |
| Applying LLMs to Active Learning: Towards Cost-Efficient Cross-Task Text Classification without Manually Labeled Data Machine learning-based classifiers have been used for text classification, such as sentiment analysis, news classification, and toxic comment classification. However, supervised machine learning models often require large amounts of labeled data for training, and manual annotation is both labor-intensive and requires domain-specific knowledge, leading to relatively high annotation costs. To address this issue, we propose an approach that integrates large language models (LLMs) into an active learning framework. Our approach combines the Robustly Optimized BERT Pretraining Approach (RoBERTa), Generative Pre-trained Transformer (GPT), and active learning, achieving high cross-task text classification performance without the need for any manually labeled data. Furthermore, compared to directly applying GPT for classification tasks, our approach retains over 93% of its classification performance while requiring only approximately 6% of the computational time and monetary cost, effectively balancing performance and resource efficiency. These findings provide new insights into the efficient utilization of LLMs and active learning algorithms in text classification tasks, paving the way for their broader application. |
2025-02-24 | Statement in Accordance with IEEE Preprint Policy: This work is intended for submission to the IEEE for possible publication |
| Krutrim LLM: Multilingual Foundational Model for over a Billion People India is a diverse society with unique challenges in developing AI systems, including linguistic diversity, oral traditions, data accessibility, and scalability. Existing foundation models are primarily trained on English, limiting their effectiveness for India’s population. Indic languages comprise only 1 percent of Common Crawl corpora despite India representing 18 percent of the global population, leading to linguistic biases. Thousands of regional languages, dialects, and code mixing create additional representation challenges due to sparse training data. We introduce Krutrim LLM, a 2 trillion token multilingual model designed for India’s linguistic landscape. It incorporates the largest known Indic dataset, mitigating data scarcity and ensuring balanced performance across dialects. Krutrim outperforms or matches state-of-the-art models on Indic benchmarks while maintaining competitive English performance. Despite being significantly smaller in training flops, Krutrim LLM matches or exceeds models like LLAMA-2 on 10 out of 16 tasks, with an average score of 0.57 versus 0.55. This evidences Krutrim’s flexible multilingual fluency across diverse linguistic contexts. Krutrim is integrated with real-time search to improve factual accuracy in conversational AI applications. This enhances accessibility for over 1 billion users worldwide. Through intentional design choices addressing data imbalances, Krutrim LLM signifies meaningful progress in building ethical, globally representative AI models. |
2025-02-24 | |
| A Multi-LLM-Agent-Based Framework for Economic and Public Policy Analysis This paper pioneers a novel approach to economic and public policy analysis by leveraging multiple Large Language Models (LLMs) as heterogeneous artificial economic agents. We first evaluate five LLMs’ economic decision-making capabilities in solving two-period consumption allocation problems under two distinct scenarios: with explicit utility functions and based on intuitive reasoning. While previous research has often simulated heterogeneity by solely varying prompts, our approach harnesses the inherent variations in analytical capabilities across different LLMs to model agents with diverse cognitive traits. Building on these findings, we construct a Multi-LLM-Agent-Based (MLAB) framework by mapping these LLMs to specific educational groups and corresponding income brackets. Using interest-income taxation as a case study, we demonstrate how the MLAB framework can simulate policy impacts across heterogeneous agents, offering a promising new direction for economic and public policy analysis by leveraging LLMs’ human-like reasoning capabilities and computational power. |
2025-02-24 | |
| 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. |
2025-02-24 | Accepted to NAACL 2025 main |
| INDIC QA BENCHMARK: A Multilingual Benchmark to Evaluate Question Answering capability of LLMs for Indic Languages Large Language Models (LLMs) perform well on unseen tasks in English, but their abilities in non English languages are less explored due to limited benchmarks and training data. To bridge this gap, we introduce the Indic QA Benchmark, a large dataset for context grounded question answering in 11 major Indian languages, covering both extractive and abstractive tasks. Evaluations of multilingual LLMs, including instruction finetuned versions, revealed weak performance in low resource languages due to a strong English language bias in their training data. We also investigated the Translate Test paradigm,where inputs are translated to English for processing and the results are translated back into the source language for output. This approach outperformed multilingual LLMs, particularly in low resource settings. By releasing Indic QA, we aim to promote further research into LLMs question answering capabilities in low resource languages. This benchmark offers a critical resource to address existing limitations and foster multilingual understanding. |
2025-02-24 | |
| Improving LLM General Preference Alignment via Optimistic Online Mirror Descent Reinforcement learning from human feedback (RLHF) has demonstrated remarkable effectiveness in aligning large language models (LLMs) with human preferences. Many existing alignment approaches rely on the Bradley-Terry (BT) model assumption, which assumes the existence of a ground-truth reward for each prompt-response pair. However, this assumption can be overly restrictive when modeling complex human preferences. In this paper, we drop the BT model assumption and study LLM alignment under general preferences, formulated as a two-player game. Drawing on theoretical insights from learning in games, we integrate optimistic online mirror descent into our alignment framework to approximate the Nash policy. Theoretically, we demonstrate that our approach achieves an $O(T^{-1})$ bound on the duality gap, improving upon the previous $O(T^{-1/2})$ result. More importantly, we implement our method and show through experiments that it outperforms state-of-the-art RLHF algorithms across multiple representative benchmarks. |
2025-02-24 | |
| PAPILLON: Efficient and Stealthy Fuzz Testing-Powered Jailbreaks for LLMs Large Language Models (LLMs) have excelled in various tasks but are still vulnerable to jailbreaking attacks, where attackers create jailbreak prompts to mislead the model to produce harmful or offensive content. Current jailbreak methods either rely heavily on manually crafted templates, which pose challenges in scalability and adaptability, or struggle to generate semantically coherent prompts, making them easy to detect. Additionally, most existing approaches involve lengthy prompts, leading to higher query costs.In this paper, to remedy these challenges, we introduce a novel jailbreaking attack framework called PAPILLON, which is an automated, black-box jailbreaking attack framework that adapts the black-box fuzz testing approach with a series of customized designs. Instead of relying on manually crafted templates,PAPILLON starts with an empty seed pool, removing the need to search for any related jailbreaking templates. We also develop three novel question-dependent mutation strategies using an LLM helper to generate prompts that maintain semantic coherence while significantly reducing their length. Additionally, we implement a two-level judge module to accurately detect genuine successful jailbreaks. We evaluated PAPILLON on 7 representative LLMs and compared it with 5 state-of-the-art jailbreaking attack strategies. For proprietary LLM APIs, such as GPT-3.5 turbo, GPT-4, and Gemini-Pro, PAPILLONs achieves attack success rates of over 90%, 80%, and 74%, respectively, exceeding existing baselines by more than 60\%. Additionally, PAPILLON can maintain high semantic coherence while significantly reducing the length of jailbreak prompts. When targeting GPT-4, PAPILLON can achieve over 78% attack success rate even with 100 tokens. Moreover, PAPILLON demonstrates transferability and is robust to state-of-the-art defenses. |
2025-02-24 | |
| AgentFL: Scaling LLM-based Fault Localization to Project-Level Context Fault Localization (FL) is an essential step during the debugging process. With the strong capabilities of code comprehension, the recent Large Language Models (LLMs) have demonstrated promising performance in diagnosing bugs in the code. Nevertheless, due to LLMs’ limited performance in handling long contexts, existing LLM-based fault localization remains on localizing bugs within a small code scope (i.e., a method or a class), which struggles to diagnose bugs for a large code scope (i.e., an entire software system). To address the limitation, this paper presents AgentFL, a multi-agent system based on ChatGPT for automated fault localization. By simulating the behavior of a human developer, AgentFL models the FL task as a three-step process, which involves comprehension, navigation, and confirmation. Within each step, AgentFL hires agents with diversified expertise, each of which utilizes different tools to handle specific tasks. Particularly, we adopt a series of auxiliary strategies such as Test Behavior Tracking, Document-Guided Search, and Multi-Round Dialogue to overcome the challenges in each step. The evaluation on the widely used Defects4J-V1.2.0 benchmark shows that AgentFL can localize 157 out of 395 bugs within Top-1, which outperforms the other LLM-based approaches and exhibits complementarity to the state-of-the-art learning-based techniques. Additionally, we confirm the indispensability of the components in AgentFL with the ablation study and demonstrate the usability of AgentFL through a user study. Finally, the cost analysis shows that AgentFL spends an average of only 0.074 dollars and 97 seconds for a single bug. |
2025-02-24 | Added a comment to refer to the published version |
| AlphaAgent: LLM-Driven Alpha Mining with Regularized Exploration to Counteract Alpha Decay Alpha mining, a critical component in quantitative investment, focuses on discovering predictive signals for future asset returns in increasingly complex financial markets. However, the pervasive issue of alpha decay, where factors lose their predictive power over time, poses a significant challenge for alpha mining. Traditional methods like genetic programming face rapid alpha decay from overfitting and complexity, while approaches driven by Large Language Models (LLMs), despite their promise, often rely too heavily on existing knowledge, creating homogeneous factors that worsen crowding and accelerate decay. To address this challenge, we propose AlphaAgent, an autonomous framework that effectively integrates LLM agents with ad hoc regularizations for mining decay-resistant alpha factors. AlphaAgent employs three key mechanisms: (i) originality enforcement through a similarity measure based on abstract syntax trees (ASTs) against existing alphas, (ii) hypothesis-factor alignment via LLM-evaluated semantic consistency between market hypotheses and generated factors, and (iii) complexity control via AST-based structural constraints, preventing over-engineered constructions that are prone to overfitting. These mechanisms collectively guide the alpha generation process to balance originality, financial rationale, and adaptability to evolving market conditions, mitigating the risk of alpha decay. Extensive evaluations show that AlphaAgent outperforms traditional and LLM-based methods in mitigating alpha decay across bull and bear markets, consistently delivering significant alpha in Chinese CSI 500 and US S&P 500 markets over the past four years. Notably, AlphaAgent showcases remarkable resistance to alpha decay, elevating the potential for yielding powerful factors. |
2025-02-24 | 9 pag |
| MultiOCR-QA: Dataset for Evaluating Robustness of LLMs in Question Answering on Multilingual OCR Texts Optical Character Recognition (OCR) plays a crucial role in digitizing historical and multilingual documents, yet OCR errors – imperfect extraction of the text, including character insertion, deletion and permutation – can significantly impact downstream tasks like question-answering (QA). In this work, we introduce a multilingual QA dataset MultiOCR-QA, designed to analyze the effects of OCR noise on QA systems’ performance. The MultiOCR-QA dataset comprises 60K question-answer pairs covering three languages, English, French, and German. The dataset is curated from OCR-ed old documents, allowing for the evaluation of OCR-induced challenges on question answering. We evaluate MultiOCR-QA on various levels and types of OCR errors to access the robustness of LLMs in handling real-world digitization errors. Our findings show that QA systems are highly prone to OCR induced errors and exhibit performance degradation on noisy OCR text. |
2025-02-24 | |
| Deep Learning and LLM-based Methods Applied to Stellar Lightcurve Classification Light curves serve as a valuable source of information on stellar formation and evolution. With the rapid advancement of machine learning techniques, it can be effectively processed to extract astronomical patterns and information. In this study, we present a comprehensive evaluation of deep-learning and large language model (LLM) based models for the automatic classification of variable star light curves, based on large datasets from the Kepler and K2 missions. Special emphasis is placed on Cepheids, RR Lyrae, and eclipsing binaries, examining the influence of observational cadence and phase distribution on classification precision. Employing AutoDL optimization, we achieve striking performance with the 1D-Convolution+BiLSTM architecture and the Swin Transformer, hitting accuracies of 94\% and 99\% correspondingly, with the latter demonstrating a notable 83\% accuracy in discerning the elusive Type II Cepheids-comprising merely 0.02\% of the total dataset.We unveil StarWhisper LightCurve (LC), an innovative Series comprising three LLM-based models: LLM, multimodal large language model (MLLM), and Large Audio Language Model (LALM). Each model is fine-tuned with strategic prompt engineering and customized training methods to explore the emergent abilities of these models for astronomical data. Remarkably, StarWhisper LC Series exhibit high accuracies around 90\%, significantly reducing the need for explicit feature engineering, thereby paving the way for streamlined parallel data processing and the progression of multifaceted multimodal models in astronomical applications. The study furnishes two detailed catalogs illustrating the impacts of phase and sampling intervals on deep learning classification accuracy, showing that a substantial decrease of up to 14\% in observation duration and 21\% in sampling points can be realized without compromising accuracy by more than 10\%. |
2025-02-24 | 35 pages, 20 figur |
| The Blessing of Reasoning: LLM-Based Contrastive Explanations in Black-Box Recommender Systems Modern recommender systems use ML models to predict consumer preferences from consumption history. Although these “black-box” models achieve impressive predictive performance, they often suffer from a lack of transparency and explainability. Contrary to the presumed tradeoff between explainability and accuracy, we show that integrating large language models (LLMs) with deep neural networks (DNNs) can improve both. We propose LR-Recsys, which augments DNN-based systems with LLM reasoning capabilities. LR-Recsys introduces a contrastive-explanation generator that produces human-readable positive explanations and negative explanations. These explanations are embedded via a fine-tuned autoencoder and combined with consumer and product features to improve predictions. Beyond offering explainability, we show that LR-Recsys also improves learning efficiency and predictive accuracy, as supported by high-dimensional, multi-environment statistical learning theory. LR-Recsys outperforms state-of-the-art recommender systems by 3-14% on three real-world datasets. Importantly, our analysis reveals that these gains primarily derive from LLMs’ reasoning capabilities rather than their external domain knowledge. LR-RecSys presents an effective approach to combine LLMs with traditional DNNs, two of the most widely used ML models today. The explanations generated by LR-Recsys provide actionable insights for consumers, sellers, and platforms, helping to build trust, optimize product offerings, and inform targeting strategies. |
2025-02-24 | |
| LLM Inference Acceleration via Efficient Operation Fusion The rapid development of the Transformer-based Large Language Models (LLMs) in recent years has been closely linked to their ever-growing and already enormous sizes. Many LLMs contain hundreds of billions of parameters and require dedicated hardware resources for training and inference. One of the key challenges inherent to the Transformer architecture is the requirement to support numerous non-linear transformations that involves normalization. For instance, each decoder block typically contains at least one Softmax operation and two Layernorms. The computation of the corresponding normalization scaling factors becomes a major bottleneck as it requires spatial collective operations. In other words, when it comes to the computation of denominators for Softmax and Layernorm, all vector elements must be aggregated into a single location, requiring significant communication. These collective operations slow down inference on Transformers by approximately 20%, defeating the whole purpose of distributed in-memory compute. In this work, we propose an extremely efficient technique that can completely hide the overhead caused by such collective operations. Note that each Softmax and Layernorm operation is typically followed by a linear layer. Since non-linear and linear operations are performed on different hardware engines, they can be easily parallelized once the algebra allows such commutation. By leveraging the inherent properties of linear operations, we can defer the normalization of the preceding Softmax and Layernorm until after the linear layer is computed. Now we can compute the collective scaling factors concurrently with the matrix multiplication and completely hide the latency of the former behind the latter. Such parallelization preserves the numerical accuracy while significantly improving the hardware utilization and reducing the overall latency. |
2025-02-24 | |
| SQLong: Enhanced NL2SQL for Longer Contexts with LLMs Open-weight large language models (LLMs) have significantly advanced performance in the Natural Language to SQL (NL2SQL) task. However, their effectiveness diminishes when dealing with large database schemas, as the context length increases. To address this limitation, we present SQLong, a novel and efficient data augmentation framework designed to enhance LLM performance in long-context scenarios for the NL2SQL task. SQLong generates augmented datasets by extending existing database schemas with additional synthetic CREATE TABLE commands and corresponding data rows, sampled from diverse schemas in the training data. This approach effectively simulates long-context scenarios during finetuning and evaluation. Through experiments on the Spider and BIRD datasets, we demonstrate that LLMs finetuned with SQLong-augmented data significantly outperform those trained on standard datasets. These imply SQLong’s practical implementation and its impact on improving NL2SQL capabilities in real-world settings with complex database schemas. |
2025-02-23 | |
| Next-Generation Database Interfaces: A Survey of LLM-based Text-to-SQL Generating accurate SQL from users’ natural language questions (text-to-SQL) remains a long-standing challenge due to the complexities involved in user question understanding, database schema comprehension, and SQL generation. Traditional text-to-SQL systems, which combine human engineering and deep neural networks, have made significant progress. Subsequently, pre-trained language models (PLMs) have been developed for text-to-SQL tasks, achieving promising results. However, as modern databases and user questions grow more complex, PLMs with a limited parameter size often produce incorrect SQL. This necessitates more sophisticated and tailored optimization methods, which restricts the application of PLM-based systems. Recently, large language models (LLMs) have shown significant capabilities in natural language understanding as model scale increases. Thus, integrating LLM-based solutions can bring unique opportunities, improvements, and solutions to text-to-SQL research. In this survey, we provide a comprehensive review of existing LLM-based text-to-SQL studies. Specifically, we offer a brief overview of the technical challenges and evolutionary process of text-to-SQL. Next, we introduce the datasets and metrics designed to evaluate text-to-SQL systems. Subsequently, we present a systematic analysis of recent advances in LLM-based text-to-SQL. Finally, we make a summarization and discuss the remaining challenges in this field and suggest expectations for future research directions. |
2025-02-23 | |
| RapidPen: Fully Automated IP-to-Shell Penetration Testing with LLM-based Agents We present RapidPen, a fully automated penetration testing (pentesting) framework that addresses the challenge of achieving an initial foothold (IP-to-Shell) without human intervention. Unlike prior approaches that focus primarily on post-exploitation or require a human-in-the-loop, RapidPen leverages large language models (LLMs) to autonomously discover and exploit vulnerabilities, starting from a single IP address. By integrating advanced ReAct-style task planning (Re) with retrieval-augmented knowledge bases of successful exploits, along with a command-generation and direct execution feedback loop (Act), RapidPen systematically scans services, identifies viable attack vectors, and executes targeted exploits in a fully automated manner. In our evaluation against a vulnerable target from the Hack The Box platform, RapidPen achieved shell access within 200-400 seconds at a per-run cost of approximately $0.3-$0.6, demonstrating a 60\% success rate when reusing prior “success-case” data. These results underscore the potential of truly autonomous pentesting for both security novices and seasoned professionals. Organizations without dedicated security teams can leverage RapidPen to quickly identify critical vulnerabilities, while expert pentesters can offload repetitive tasks and focus on complex challenges. Ultimately, our work aims to make penetration testing more accessible and cost-efficient, thereby enhancing the overall security posture of modern software ecosystems. |
2025-02-23 | |
| Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyond The LLM unlearning technique has recently been introduced to comply with data regulations and address the safety and ethical concerns of LLMs by removing the undesired data-model influence. However, state-of-the-art unlearning methods face a critical vulnerability: they are susceptible to ``relearning’’ the removed information from a small number of forget data points, known as relearning attacks. In this paper, we systematically investigate how to make unlearned models robust against such attacks. For the first time, we establish a connection between robust unlearning and sharpness-aware minimization (SAM) through a unified robust optimization framework, in an analogy to adversarial training designed to defend against adversarial attacks. Our analysis for SAM reveals that smoothness optimization plays a pivotal role in mitigating relearning attacks. Thus, we further explore diverse smoothing strategies to enhance unlearning robustness. Extensive experiments on benchmark datasets, including WMDP and MUSE, demonstrate that SAM and other smoothness optimization approaches consistently improve the resistance of LLM unlearning to relearning attacks. Notably, smoothness-enhanced unlearning also helps defend against (input-level) jailbreaking attacks, broadening our proposal’s impact in robustifying LLM unlearning. Codes are available at https://github.com/OPTML-Group/Unlearn-Smooth. |
2025-02-23 | |
| Dynamic LLM Routing and Selection based on User Preferences: Balancing Performance, Cost, and Ethics With the widespread deployment of large language models (LLMs) such as GPT4, BART, and LLaMA, the need for a system that can intelligently select the most suitable model for specific tasks while balancing cost, latency, accuracy, and ethical considerations has become increasingly important. Recognizing that not all tasks necessitate models with over 100 billion parameters, we introduce OptiRoute, an advanced model routing engine designed to dynamically select and route tasks to the optimal LLM based on detailed user-defined requirements. OptiRoute captures both functional (e.g., accuracy, speed, cost) and non-functional (e.g., helpfulness, harmlessness, honesty) criteria, leveraging lightweight task analysis and complexity estimation to efficiently match tasks with the best-fit models from a diverse array of LLMs. By employing a hybrid approach combining k-nearest neighbors (kNN) search and hierarchical filtering, OptiRoute optimizes for user priorities while minimizing computational overhead. This makes it ideal for real-time applications in cloud-based ML platforms, personalized AI services, and regulated industries. |
2025-02-23 | |
| From Text to Space: Mapping Abstract Spatial Models in LLMs during a Grid-World Navigation Task Understanding how large language models (LLMs) represent and reason about spatial information is crucial for building robust agentic systems that can navigate real and simulated environments. In this work, we investigate the influence of different text-based spatial representations on LLM performance and internal activations in a grid-world navigation task. By evaluating models of various sizes on a task that requires navigating toward a goal, we examine how the format used to encode spatial information impacts decision-making. Our experiments reveal that cartesian representations of space consistently yield higher success rates and path efficiency, with performance scaling markedly with model size. Moreover, probing LLaMA-3.1-8B revealed subsets of internal units, primarily located in intermediate layers, that robustly correlate with spatial features, such as the position of the agent in the grid or action correctness, regardless of how that information is represented, and are also activated by unrelated spatial reasoning tasks. This work advances our understanding of how LLMs process spatial information and provides valuable insights for developing more interpretable and robust agentic AI systems. |
2025-02-23 | |
| TradingAgents: Multi-Agents LLM Financial Trading Framework Significant progress has been made in automated problem-solving using societies of agents powered by large language models (LLMs). In finance, efforts have largely focused on single-agent systems handling specific tasks or multi-agent frameworks independently gathering data. However, multi-agent systems’ potential to replicate real-world trading firms’ collaborative dynamics remains underexplored. TradingAgents proposes a novel stock trading framework inspired by trading firms, featuring LLM-powered agents in specialized roles such as fundamental analysts, sentiment analysts, technical analysts, and traders with varied risk profiles. The framework includes Bull and Bear researcher agents assessing market conditions, a risk management team monitoring exposure, and traders synthesizing insights from debates and historical data to make informed decisions. By simulating a dynamic, collaborative trading environment, this framework aims to improve trading performance. Detailed architecture and extensive experiments reveal its superiority over baseline models, with notable improvements in cumulative returns, Sharpe ratio, and maximum drawdown, highlighting the potential of multi-agent LLM frameworks in financial trading. TradingAgents is available at https://github.com/TradingAgents-AI. |
2025-02-23 | Multi-Agent AI in the Real World @ AAAI 2025 |
| Are the Values of LLMs Structurally Aligned with Humans? A Causal Perspective As large language models (LLMs) become increasingly integrated into critical applications, aligning their behavior with human values presents significant challenges. Current methods, such as Reinforcement Learning from Human Feedback (RLHF), typically focus on a limited set of coarse-grained values and are resource-intensive. Moreover, the correlations between these values remain implicit, leading to unclear explanations for value-steering outcomes. Our work argues that a latent causal value graph underlies the value dimensions of LLMs and that, despite alignment training, this structure remains significantly different from human value systems. We leverage these causal value graphs to guide two lightweight value-steering methods: role-based prompting and sparse autoencoder (SAE) steering, effectively mitigating unexpected side effects. Furthermore, SAE provides a more fine-grained approach to value steering. Experiments on Gemma-2B-IT and Llama3-8B-IT demonstrate the effectiveness and controllability of our methods. |
2025-02-23 | |
| Reasoning about Affordances: Causal and Compositional Reasoning in LLMs With the rapid progress of Large Language Models (LLMs), it becomes increasingly important to understand their abilities and limitations. In two experiments, we investigate the causal and compositional reasoning abilities of LLMs and humans in the domain of object affordances, an area traditionally linked to embodied cognition. The tasks, designed from scratch to avoid data contamination, require decision-makers to select unconventional objects to replace a typical tool for a particular purpose, such as using a table tennis racket to dig a hole. In Experiment 1, we evaluated GPT-3.5 and GPT-4o, finding that GPT-4o, when given chain-of-thought prompting, performed on par with human participants, while GPT-3.5 lagged significantly. In Experiment 2, we introduced two new conditions, Distractor (more object choices, increasing difficulty) and Image (object options presented visually), and evaluated Claude 3 Sonnet and Claude 3.5 Sonnet in addition to the GPT models. The Distractor condition significantly impaired performance across humans and models, although GPT-4o and Claude 3.5 still performed well above chance. Surprisingly, the Image condition had little impact on humans or GPT-4o, but significantly lowered Claude 3.5’s accuracy. Qualitative analysis showed that GPT-4o and Claude 3.5 have a stronger ability than their predecessors to identify and flexibly apply causally relevant object properties. The improvement from GPT-3.5 and Claude 3 to GPT-4o and Claude 3.5 suggests that models are increasingly capable of causal and compositional reasoning in some domains, although further mechanistic research is necessary to understand how LLMs reason. |
2025-02-23 | 21 pages, 7 figures, 3 tab |
| Augmenting Black-box LLMs with Medical Textbooks for Biomedical Question Answering Large-scale language models (LLMs) like ChatGPT have demonstrated impressive abilities in generating responses based on human instructions. However, their use in the medical field can be challenging due to their lack of specific, in-depth knowledge. In this study, we present a system called LLMs Augmented with Medical Textbooks (LLM-AMT) designed to enhance the proficiency of LLMs in specialized domains. LLM-AMT integrates authoritative medical textbooks into the LLMs’ framework using plug-and-play modules. These modules include a Query Augmenter, a Hybrid Textbook Retriever, and a Knowledge Self-Refiner. Together, they incorporate authoritative medical knowledge. Additionally, an LLM Reader aids in contextual understanding. Our experimental results on three medical QA tasks demonstrate that LLMAMT significantly improves response quality, with accuracy gains ranging from 11.6% to 16.6%. Notably, with GPT-4-Turbo as the base model, LLM-AMT outperforms the specialized Med-PaLM 2 model pre-trained on a massive amount of medical corpus by 2-3%. We found that despite being 100x smaller in size, medical textbooks as a retrieval corpus is proven to be a more effective knowledge database than Wikipedia in the medical domain, boosting performance by 7.8%-13.7%. |
2025-02-23 | This version has been accepted and published at EMNLP Findings 2024 |
| Reasoning About Persuasion: Can LLMs Enable Explainable Propaganda Detection? There has been significant research on propagandistic content detection across different modalities and languages. However, most studies have primarily focused on detection, with little attention given to explanations justifying the predicted label. This is largely due to the lack of resources that provide explanations alongside annotated labels. To address this issue, we propose a multilingual (i.e., Arabic and English) explanation-enhanced dataset, the first of its kind. Additionally, we introduce an explanation-enhanced LLM for both label detection and rationale-based explanation generation. Our findings indicate that the model performs comparably while also generating explanations. We will make the dataset and experimental resources publicly available for the research community. |
2025-02-23 | |
| SciLitLLM: How to Adapt LLMs for Scientific Literature Understanding Scientific literature understanding is crucial for extracting targeted information and garnering insights, thereby significantly advancing scientific discovery. Despite the remarkable success of Large Language Models (LLMs), they face challenges in scientific literature understanding, primarily due to (1) a lack of scientific knowledge and (2) unfamiliarity with specialized scientific tasks. To develop an LLM specialized in scientific literature understanding, we propose a hybrid strategy that integrates continual pre-training (CPT) and supervised fine-tuning (SFT), to simultaneously infuse scientific domain knowledge and enhance instruction-following capabilities for domain-specific tasks.cIn this process, we identify two key challenges: (1) constructing high-quality CPT corpora, and (2) generating diverse SFT instructions. We address these challenges through a meticulous pipeline, including PDF text extraction, parsing content error correction, quality filtering, and synthetic instruction creation. Applying this strategy, we present a suite of LLMs: SciLitLLM, specialized in scientific literature understanding. These models demonstrate promising performance on scientific literature understanding benchmarks. Our contributions are threefold: (1) We present an effective framework that integrates CPT and SFT to adapt LLMs to scientific literature understanding, which can also be easily adapted to other domains. (2) We propose an LLM-based synthesis method to generate diverse and high-quality scientific instructions, resulting in a new instruction set – SciLitIns – for supervised fine-tuning in less-represented scientific domains. (3) SciLitLLM achieves promising performance improvements on scientific literature understanding benchmarks. |
2025-02-23 | |
| Multilingual != Multicultural: Evaluating Gaps Between Multilingual Capabilities and Cultural Alignment in LLMs Large Language Models (LLMs) are becoming increasingly capable across global languages. However, the ability to communicate across languages does not necessarily translate to appropriate cultural representations. A key concern is US-centric bias, where LLMs reflect US rather than local cultural values. We propose a novel methodology that compares LLM-generated response distributions against population-level opinion data from the World Value Survey across four languages (Danish, Dutch, English, and Portuguese). Using a rigorous linear mixed-effects regression framework, we compare two families of models: Google’s Gemma models (2B–27B parameters) and successive iterations of OpenAI’s turbo-series. Across the families of models, we find no consistent relationships between language capabilities and cultural alignment. While the Gemma models have a positive correlation between language capability and cultural alignment across languages, the OpenAI models do not. Importantly, we find that self-consistency is a stronger predictor of multicultural alignment than multilingual capabilities. Our results demonstrate that achieving meaningful cultural alignment requires dedicated effort beyond improving general language capabilities. |
2025-02-23 | |
| On Calibration of LLM-based Guard Models for Reliable Content Moderation Large language models (LLMs) pose significant risks due to the potential for generating harmful content or users attempting to evade guardrails. Existing studies have developed LLM-based guard models designed to moderate the input and output of threat LLMs, ensuring adherence to safety policies by blocking content that violates these protocols upon deployment. However, limited attention has been given to the reliability and calibration of such guard models. In this work, we empirically conduct comprehensive investigations of confidence calibration for 9 existing LLM-based guard models on 12 benchmarks in both user input and model output classification. Our findings reveal that current LLM-based guard models tend to 1) produce overconfident predictions, 2) exhibit significant miscalibration when subjected to jailbreak attacks, and 3) demonstrate limited robustness to the outputs generated by different types of response models. Additionally, we assess the effectiveness of post-hoc calibration methods to mitigate miscalibration. We demonstrate the efficacy of temperature scaling and, for the first time, highlight the benefits of contextual calibration for confidence calibration of guard models, particularly in the absence of validation sets. Our analysis and experiments underscore the limitations of current LLM-based guard models and provide valuable insights for the future development of well-calibrated guard models toward more reliable content moderation. We also advocate for incorporating reliability evaluation of confidence calibration when releasing future LLM-based guard models. |
2025-02-23 | Accepted to ICLR 2025 |
| RouteLLM: Learning to Route LLMs with Preference Data Large language models (LLMs) exhibit impressive capabilities across a wide range of tasks, yet the choice of which model to use often involves a trade-off between performance and cost. More powerful models, though effective, come with higher expenses, while less capable models are more cost-effective. To address this dilemma, we propose several efficient router models that dynamically select between a stronger and a weaker LLM during inference, aiming to optimize the balance between cost and response quality. We develop a training framework for these routers leveraging human preference data and data augmentation techniques to enhance performance. Our evaluation on widely-recognized benchmarks shows that our approach significantly reduces costs-by over 2 times in certain cases-without compromising the quality of responses. Interestingly, our router models also demonstrate significant transfer learning capabilities, maintaining their performance even when the strong and weak models are changed at test time. This highlights the potential of these routers to provide a cost-effective yet high-performance solution for deploying LLMs. |
2025-02-23 | |
| TerEffic: Highly Efficient Ternary LLM Inference on FPGA Large Language Model (LLM) deployment on edge devices is typically constrained by the need for off-chip memory access, leading to high power consumption and limited throughput. Ternary quantization for LLMs is promising in maintaining model accuracy while reducing memory footprint. However, existing accelerators have not exploited this potential for on-chip inference. We present TerEffic, an FPGA-based accelerator that carefully co-designs memory architecture and computational units to unlock highly efficient LLM inference with fully on-chip execution. Through weight compression, custom computational units, and memory hierarchy optimization, we achieve unprecedented efficiency by eliminating off-chip memory bandwidth bottlenecks. We propose two architectural variants: a fully on-chip design for smaller models and an HBM-assisted design for larger ones. When evaluated on a 370M parameter model with single-batch inference, our on-chip design achieves 12,700 tokens/sec (149 times higher than NVIDIA’s Jetson Orin Nano) with a power efficiency of 467 tokens/sec/W (19 times better than Jetson Orin Nano). The HBM-assisted design provides 521 tokens/sec on a 2.7B parameter model (2 times higher than NVIDIA’s A100) with 33W power consumption, achieving a power efficiency of 16 tokens/sec/W (8 times better than A100). |
2025-02-23 | |
| Towards Fully-Automated Materials Discovery via Large-Scale Synthesis Dataset and Expert-Level LLM-as-a-Judge Materials synthesis is vital for innovations such as energy storage, catalysis, electronics, and biomedical devices. Yet, the process relies heavily on empirical, trial-and-error methods guided by expert intuition. Our work aims to support the materials science community by providing a practical, data-driven resource. We have curated a comprehensive dataset of 17K expert-verified synthesis recipes from open-access literature, which forms the basis of our newly developed benchmark, AlchemyBench. AlchemyBench offers an end-to-end framework that supports research in large language models applied to synthesis prediction. It encompasses key tasks, including raw materials and equipment prediction, synthesis procedure generation, and characterization outcome forecasting. We propose an LLM-as-a-Judge framework that leverages large language models for automated evaluation, demonstrating strong statistical agreement with expert assessments. Overall, our contributions offer a supportive foundation for exploring the capabilities of LLMs in predicting and guiding materials synthesis, ultimately paving the way for more efficient experimental design and accelerated innovation in materials science. |
2025-02-23 | under review |
| Unveiling LLM Mechanisms Through Neural ODEs and Control Theory This paper proposes a framework combining Neural Ordinary Differential Equations (Neural ODEs) and robust control theory to enhance the interpretability and control of large language models (LLMs). By utilizing Neural ODEs to model the dynamic evolution of input-output relationships and introducing control mechanisms to optimize output quality, we demonstrate the effectiveness of this approach across multiple question-answer datasets. Experimental results show that the integration of Neural ODEs and control theory significantly improves output consistency and model interpretability, advancing the development of explainable AI technologies. |
2025-02-23 | |
| LLMs Are Biased Towards Output Formats! Systematically Evaluating and Mitigating Output Format Bias of LLMs We present the first systematic evaluation examining format bias in performance of large language models (LLMs). Our approach distinguishes between two categories of an evaluation metric under format constraints to reliably and accurately assess performance: one measures performance when format constraints are adhered to, while the other evaluates performance regardless of constraint adherence. We then define a metric for measuring the format bias of LLMs and establish effective strategies to reduce it. Subsequently, we present our empirical format bias evaluation spanning four commonly used categories – multiple-choice question-answer, wrapping, list, and mapping – covering 15 widely-used formats. Our evaluation on eight generation tasks uncovers significant format bias across state-of-the-art LLMs. We further discover that improving the format-instruction following capabilities of LLMs across formats potentially reduces format bias. Based on our evaluation findings, we study prompting and fine-tuning with synthesized format data techniques to mitigate format bias. Our methods successfully reduce the variance in ChatGPT’s performance among wrapping formats from 235.33 to 0.71 (%$^2$). |
2025-02-23 | NAACL 2025 Main Conferenc |
| Does Unlearning Truly Unlearn? A Black Box Evaluation of LLM Unlearning Methods Large language model unlearning aims to remove harmful information that LLMs have learnt to prevent their use for malicious purposes. LLMU and RMU have been proposed as two methods for LLM unlearning, achieving impressive results on unlearning benchmarks. We study in detail the impact of unlearning on LLM performance metrics using the WMDP dataset as well as a new biology dataset we create. We show that unlearning has a notable impact on general model capabilities, with the performance degradation being more significant in general for LLMU. We further test the robustness of the two methods and find that doing 5-shot prompting or rephrasing the question in simple ways can lead to an over ten-fold increase in accuracy on unlearning benchmarks. Finally, we show that training on unrelated data can almost completely recover pre-unlearning performance, demonstrating that these methods fail at truly unlearning. Our methodology serves as an evaluation framework for LLM unlearning methods. The code is available at: https://github.com/JaiDoshi/Knowledge-Erasure. |
2025-02-23 | 9 pages, 2 figur |
| Leveraging Edge Intelligence and LLMs to Advance 6G-Enabled Internet of Automated Defense Vehicles The evolution of Artificial Intelligence (AI) and its subset Deep Learning (DL), has profoundly impacted numerous domains, including autonomous driving. The integration of autonomous driving in military settings reduces human casualties and enables precise and safe execution of missions in hazardous environments while allowing for reliable logistics support without the risks associated with fatigue-related errors. However, relying on autonomous driving solely requires an advanced decision-making model that is adaptable and optimum in any situation. Considering the presence of numerous interconnected autonomous vehicles in mission-critical scenarios, Ultra-Reliable Low Latency Communication (URLLC) is vital for ensuring seamless coordination, real-time data exchange, and instantaneous response to dynamic driving environments. The advent of 6G strengthens the Internet of Automated Defense Vehicles (IoADV) concept within the realm of Internet of Military Defense Things (IoMDT) by enabling robust connectivity, crucial for real-time data exchange, advanced navigation, and enhanced safety features through IoADV interactions. On the other hand, a critical advancement in this space is using pre-trained Generative Large Language Models (LLMs) for decision-making and communication optimization for autonomous driving. Hence, this work presents opportunities and challenges with a vision of realizing the full potential of these technologies in critical defense applications, especially through the advancement of IoADV and its role in enhancing autonomous military operations. |
2025-02-23 | 8 pages, 5 figures, accepted to IEEE Internet of Things Magazin |
| Ensemble ToT of LLMs and Its Application to Automatic Grading System for Supporting Self-Learning Providing students with detailed and timely grading feedback is essential for self-learning. While existing LLM-based grading systems are promising, most of them rely on one single model, which limits their performance. To address this, we propose Ensemble Tree-of-Thought (ToT), a framework that enhances LLM outputs by integrating multiple models. Using this framework, we develop a grading system. Ensemble ToT follows three steps: (1) analyzing LLM performance, (2) generating candidate answers, and (3) refining them into a final result. Based on this, our grading system first evaluates the grading tendencies of LLMs, then generates multiple results, and finally integrates them via a simulated debate. Experimental results demonstrate our approach’s ability to provide accurate and explainable grading by effectively coordinating multiple LLMs. |
2025-02-23 | 33 pages, 25 figur |
| An Analyst-Inspector Framework for Evaluating Reproducibility of LLMs in Data Science Large Language Models (LLMs) have demonstrated potential for data science tasks via code generation. However, the exploratory nature of data science, alongside the stochastic and opaque outputs of LLMs, raise concerns about their reliability. While prior work focuses on benchmarking LLM accuracy, reproducibility remains underexplored, despite being critical to establishing trust in LLM-driven analysis. We propose a novel analyst-inspector framework to automatically evaluate and enforce the reproducibility of LLM-generated data science workflows - the first rigorous approach to the best of our knowledge. Defining reproducibility as the sufficiency and completeness of workflows for reproducing functionally equivalent code, this framework enforces computational reproducibility principles, ensuring transparent, well-documented LLM workflows while minimizing reliance on implicit model assumptions. Using this framework, we systematically evaluate five state-of-the-art LLMs on 1,032 data analysis tasks across three diverse benchmark datasets. We also introduce two novel reproducibility-enhancing prompting strategies. Our results show that higher reproducibility strongly correlates with improved accuracy and reproducibility-enhancing prompts are effective, demonstrating structured prompting’s potential to enhance automated data science workflows and enable transparent, robust AI-driven analysis. Our code is publicly available. |
2025-02-23 | |
| On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback As LLMs become more widely deployed, there is increasing interest in directly optimizing for feedback from end users (e.g. thumbs up) in addition to feedback from paid annotators. However, training to maximize human feedback creates a perverse incentive structure for the AI to resort to manipulative or deceptive tactics to obtain positive feedback from users who are vulnerable to such strategies. We study this phenomenon by training LLMs with Reinforcement Learning with simulated user feedback in environments of practical LLM usage. In our settings, we find that: 1) Extreme forms of “feedback gaming” such as manipulation and deception are learned reliably; 2) Even if only 2% of users are vulnerable to manipulative strategies, LLMs learn to identify and target them while behaving appropriately with other users, making such behaviors harder to detect; 3) To mitigate this issue, it may seem promising to leverage continued safety training or LLM-as-judges during training to filter problematic outputs. Instead, we found that while such approaches help in some of our settings, they backfire in others, sometimes even leading to subtler manipulative behaviors. We hope our results can serve as a case study which highlights the risks of using gameable feedback sources – such as user feedback – as a target for RL. |
2025-02-22 | Accepted to ICLR 2025 |
| LLMKey: LLM-Powered Wireless Key Generation Scheme for Next-Gen IoV Systems Wireless key generation holds significant promise for establishing cryptographic keys in Next-Gen Internet of Vehicles (IoV) systems. However, existing approaches often face inefficiencies and performance limitations caused by frequent channel probing and ineffective quantization. To address these challenges, this paper introduces LLMKey, a novel key generation system designed to enhance efficiency and security. We identify excessive channel probing and suboptimal quantization as critical bottlenecks in current methods. To mitigate these issues, we propose an innovative large language model (LLM)-based channel probing technique that leverages the capabilities of LLMs to reduce probing rounds while preserving crucial channel information. Instead of conventional quantization, LLMKey adopts a perturbed compressed sensing-based key delivery mechanism, improving both robustness and security. Extensive evaluations are conducted in four real-world scenarios, encompassing V2I (Vehicle-to-Infrastructure) and V2V (Vehicle-to-Vehicle) settings in both urban and rural environments. The results show that LLMKey achieves an average key agreement rate of 98.78\%, highlighting its effectiveness and reliability across diverse conditions. |
2025-02-22 | |
| A Federated Framework for LLM-based Recommendation Large Language Models (LLMs) have empowered generative recommendation systems through fine-tuning user behavior data. However, utilizing the user data may pose significant privacy risks, potentially leading to ethical dilemmas and violations of data protection regulations. To address the privacy concerns, Federated Learning for Recommendation (Fed4Rec) has been identified as a promising solution. However, directly applying Fed4Rec in the LLM context introduces two challenges: 1) exacerbated client performance imbalance, which ultimately impacts the system’s long-term effectiveness, and 2) substantial client resource costs, posing a high demand for clients’ both computational and storage capability to locally train and infer LLMs. To tackle these challenges, we propose a federated framework for LLM-based recommendation (shorted as FELLRec). Generally, FELLRec designs two key strategies. 1) Dynamic balance strategy, which designs dynamic parameter aggregation and learning speed for different clients, aiming to ensure balanced performance across clients. 2) Flexible storage strategy, which selectively retains certain sensitive LLM layers on the client side, while offloading other layers to the server, aiming to preserve privacy while saving resources. Experiment results show that FELLRec can achieve a more balanced client performance and improved overall performance in a computational and storage-efficient way while safeguarding user privacy well. |
2025-02-22 | Accepted by NAACL 2025 Finding |
| Mojito: LLM-Aided Motion Instructor with Jitter-Reduced Inertial Tokens Human bodily movements convey critical insights into action intentions and cognitive processes, yet existing multimodal systems primarily focused on understanding human motion via language, vision, and audio, which struggle to capture the dynamic forces and torques inherent in 3D motion. Inertial measurement units (IMUs) present a promising alternative, offering lightweight, wearable, and privacy-conscious motion sensing. However, processing of streaming IMU data faces challenges such as wireless transmission instability, sensor noise, and drift, limiting their utility for long-term real-time motion capture (MoCap), and more importantly, online motion analysis. To address these challenges, we introduce Mojito, an intelligent motion agent that integrates inertial sensing with large language models (LLMs) for interactive motion capture and behavioral analysis. |
2025-02-22 | First three authors contribute equally. Project page: https://koyui.github.io/mojito |
| Maybe I Should Not Answer That, but… Do LLMs Understand The Safety of Their Inputs? Ensuring the safety of the Large Language Model (LLM) is critical, but currently used methods in most cases sacrifice the model performance to obtain increased safety or perform poorly on data outside of their adaptation distribution. We investigate existing methods for such generalization and find them insufficient. Surprisingly, while even plain LLMs recognize unsafe prompts, they may still generate unsafe responses. To avoid performance degradation and preserve safe performance, we advocate for a two-step framework, where we first identify unsafe prompts via a lightweight classifier, and apply a “safe” model only to such prompts. In particular, we explore the design of the safety detector in more detail, investigating the use of different classifier architectures and prompting techniques. Interestingly, we find that the final hidden state for the last token is enough to provide robust performance, minimizing false positives on benign data while performing well on malicious prompt detection. Additionally, we show that classifiers trained on the representations from different model layers perform comparably on the latest model layers, indicating that safety representation is present in the LLMs’ hidden states at most model stages. Our work is a step towards efficient, representation-based safety mechanisms for LLMs. |
2025-02-22 | |
| Patterns Over Principles: The Fragility of Inductive Reasoning in LLMs under Noisy Observations Inductive reasoning, a cornerstone of human cognition, enables generalization from limited data but hasn’t yet been fully achieved by large language models (LLMs). While modern LLMs excel at reasoning tasks, their ability to maintain stable and consistent rule abstraction under imperfect observations remains underexplored. To fill this gap, in this work, we introduce Robust Rule Induction, a task that evaluates LLMs’ capability in inferring rules from data that are fused with noisy examples. To address this task, we further propose Sample-steered Rule Refinement (SRR), a method enhancing reasoning stability via observation diversification and execution-guided feedback. Experiments across arithmetic, cryptography, and list functions reveal: (1) SRR outperforms other methods with minimal performance degradation under noise; (2) Despite slight accuracy variation, LLMs exhibit instability under noise (e.g., 0% accuracy change with only 70% consistent score); (3) Counterfactual task gaps highlight LLMs’ reliance on memorized patterns over genuine abstraction. Our findings challenge LLMs’ reasoning robustness, revealing susceptibility to hypothesis drift and pattern overfitting, while providing empirical evidence critical for developing human-like inductive systems. Code and data are available at \href{https://github.com/lcy2723/Robust-Rule-Induction}{https://github.com/lcy2723/Robust-Rule-Induction}. |
2025-02-22 | |
| Number Representations in LLMs: A Computational Parallel to Human Perception Humans are believed to perceive numbers on a logarithmic mental number line, where smaller values are represented with greater resolution than larger ones. This cognitive bias, supported by neuroscience and behavioral studies, suggests that numerical magnitudes are processed in a sublinear fashion rather than on a uniform linear scale. Inspired by this hypothesis, we investigate whether large language models (LLMs) exhibit a similar logarithmic-like structure in their internal numerical representations. By analyzing how numerical values are encoded across different layers of LLMs, we apply dimensionality reduction techniques such as PCA and PLS followed by geometric regression to uncover latent structures in the learned embeddings. Our findings reveal that the model’s numerical representations exhibit sublinear spacing, with distances between values aligning with a logarithmic scale. This suggests that LLMs, much like humans, may encode numbers in a compressed, non-uniform manner. |
2025-02-22 | The number line of LLM |
| The Law of Knowledge Overshadowing: Towards Understanding, Predicting, and Preventing LLM Hallucination Hallucination is a persistent challenge in large language models (LLMs), where even with rigorous quality control, models often generate distorted facts. This paradox, in which error generation continues despite high-quality training data, calls for a deeper understanding of the underlying LLM mechanisms. To address it, we propose a novel concept: knowledge overshadowing, where model’s dominant knowledge can obscure less prominent knowledge during text generation, causing the model to fabricate inaccurate details. Building on this idea, we introduce a novel framework to quantify factual hallucinations by modeling knowledge overshadowing. Central to our approach is the log-linear law, which predicts that the rate of factual hallucination increases linearly with the logarithmic scale of (1) Knowledge Popularity, (2) Knowledge Length, and (3) Model Size. The law provides a means to preemptively quantify hallucinations, offering foresight into their occurrence even before model training or inference. Built on overshadowing effect, we propose a new decoding strategy CoDa, to mitigate hallucinations, which notably enhance model factuality on Overshadow (27.9%), MemoTrap (13.1%) and NQ-Swap (18.3%). Our findings not only deepen understandings of the underlying mechanisms behind hallucinations but also provide actionable insights for developing more predictable and controllable language models. |
2025-02-22 | 19 pages, 5 figur |
| Understanding Zero-shot Rare Word Recognition Improvements Through LLM Integration In this study, we investigate the integration of a large language model (LLM) with an automatic speech recognition (ASR) system, specifically focusing on enhancing rare word recognition performance. Using a 190,000-hour dataset primarily sourced from YouTube, pre-processed with Whisper V3 pseudo-labeling, we demonstrate that the LLM-ASR architecture outperforms traditional Zipformer-Transducer models in the zero-shot rare word recognition task, after training on a large dataset. Our analysis reveals that the LLM contributes significantly to improvements in rare word error rate (R-WER), while the speech encoder primarily determines overall transcription performance (Orthographic Word Error Rate, O-WER, and Normalized Word Error Rate, N-WER). Through extensive ablation studies, we highlight the importance of adapter integration in aligning speech encoder outputs with the LLM’s linguistic capabilities. Furthermore, we emphasize the critical role of high-quality labeled data in achieving optimal performance. These findings provide valuable insights into the synergy between LLM-based ASR architectures, paving the way for future advancements in large-scale LLM-based speech recognition systems. |
2025-02-22 | |
| Mixture Compressor for Mixture-of-Experts LLMs Gains More Mixture-of-Experts large language models (MoE-LLMs) marks a significant step forward of language models, however, they encounter two critical challenges in practice: 1) expert parameters lead to considerable memory consumption and loading latency; and 2) the current activated experts are redundant, as many tokens may only require a single expert. Motivated by these issues, we investigate the MoE-LLMs and make two key observations: a) different experts exhibit varying behaviors on activation reconstruction error, routing scores, and activated frequencies, highlighting their differing importance, and b) not all tokens are equally important – only a small subset is critical. Building on these insights, we propose MC, a training-free Mixture-Compressor for MoE-LLMs, which leverages the significance of both experts and tokens to achieve an extreme compression. First, to mitigate storage and loading overheads, we introduce Pre-Loading Mixed-Precision Quantization, which formulates the adaptive bit-width allocation as a Linear Programming problem, where the objective function balances multi-factors reflecting the importance of each expert. Additionally, we develop Online Dynamic Pruning, which identifies important tokens to retain and dynamically select activated experts for other tokens during inference to optimize efficiency while maintaining performance. Our MC integrates static quantization and dynamic pruning to collaboratively achieve extreme compression for MoE-LLMs with less accuracy loss, ensuring an optimal trade-off between performance and efficiency. Extensive experiments confirm the effectiveness of our approach. For instance, at 2.54 bits, MC compresses 76.6% of the model, with only a 3.8% average accuracy loss. During dynamic inference, we further reduce activated parameters by 15%, with a performance drop of less than 0.6%. |
2025-02-22 | ICLR 2025 |
| Humanizing the Machine: Proxy Attacks to Mislead LLM Detectors The advent of large language models (LLMs) has revolutionized the field of text generation, producing outputs that closely mimic human-like writing. Although academic and industrial institutions have developed detectors to prevent the malicious usage of LLM-generated texts, other research has doubt about the robustness of these systems. To stress test these detectors, we introduce a proxy-attack strategy that effortlessly compromises LLMs, causing them to produce outputs that align with human-written text and mislead detection systems. Our method attacks the source model by leveraging a reinforcement learning (RL) fine-tuned humanized small language model (SLM) in the decoding phase. Through an in-depth analysis, we demonstrate that our attack strategy is capable of generating responses that are indistinguishable to detectors, preventing them from differentiating between machine-generated and human-written text. We conduct systematic evaluations on extensive datasets using proxy-attacked open-source models, including Llama2-13B, Llama3-70B, and Mixtral-8*7B in both white- and black-box settings. Our findings show that the proxy-attack strategy effectively deceives the leading detectors, resulting in an average AUROC drop of 70.4% across multiple datasets, with a maximum drop of 90.3% on a single dataset. Furthermore, in cross-discipline scenarios, our strategy also bypasses these detectors, leading to a significant relative decrease of up to 90.9%, while in cross-language scenario, the drop reaches 91.3%. Despite our proxy-attack strategy successfully bypassing the detectors with such significant relative drops, we find that the generation quality of the attacked models remains preserved, even within a modest utility budget, when compared to the text produced by the original, unattacked source model. |
2025-02-22 | 29 pag |
| Merger-as-a-Stealer: Stealing Targeted PII from Aligned LLMs with Model Merging Model merging has emerged as a promising approach for updating large language models (LLMs) by integrating multiple domain-specific models into a cross-domain merged model. Despite its utility and plug-and-play nature, unmonitored mergers can introduce significant security vulnerabilities, such as backdoor attacks and model merging abuse. In this paper, we identify a novel and more realistic attack surface where a malicious merger can extract targeted personally identifiable information (PII) from an aligned model with model merging. Specifically, we propose \texttt{Merger-as-a-Stealer}, a two-stage framework to achieve this attack: First, the attacker fine-tunes a malicious model to force it to respond to any PII-related queries. The attacker then uploads this malicious model to the model merging conductor and obtains the merged model. Second, the attacker inputs direct PII-related queries to the merged model to extract targeted PII. Extensive experiments demonstrate that \texttt{Merger-as-a-Stealer} successfully executes attacks against various LLMs and model merging methods across diverse settings, highlighting the effectiveness of the proposed framework. Given that this attack enables character-level extraction for targeted PII without requiring any additional knowledge from the attacker, we stress the necessity for improved model alignment and more robust defense mechanisms to mitigate such threats. |
2025-02-22 | 17 pages, 3 figur |
| CitaLaw: Enhancing LLM with Citations in Legal Domain In this paper, we propose CitaLaw, the first benchmark designed to evaluate LLMs’ ability to produce legally sound responses with appropriate citations. CitaLaw features a diverse set of legal questions for both laypersons and practitioners, paired with a comprehensive corpus of law articles and precedent cases as a reference pool. This framework enables LLM-based systems to retrieve supporting citations from the reference corpus and align these citations with the corresponding sentences in their responses. Moreover, we introduce syllogism-inspired evaluation methods to assess the legal alignment between retrieved references and LLM-generated responses, as well as their consistency with user questions. Extensive experiments on 2 open-domain and 7 legal-specific LLMs demonstrate that integrating legal references substantially enhances response quality. Furthermore, our proposed syllogism-based evaluation method exhibits strong agreement with human judgments. |
2025-02-22 |