llm - 2025_02
Navigation
- Part 1
- Part 2
- Part 3
- Part 4
- Part 5
- Part 6
- Part 7
- Part 8
- Part 9
- Part 10
- Part 11
- Part 12
- Part 13
- Part 14
Papers
While large language models (LLMs) excel in generating coherent and contextually rich outputs, their capacity to efficiently handle long-form contexts is limited by fixed-length position embeddings. Additionally, the computational cost of processing long sequences increases quadratically, making it challenging to extend context length. To address these challenges, we propose Long-form Context Injection with Recurrent Compression (LCIRC), a method that enables the efficient processing long-form sequences beyond the model's length limit through recurrent compression without retraining the entire model. We further introduce query dependent context modeling, which selectively compresses query-relevant information, ensuring that the model retains the most pertinent content. Our empirical results demonstrate that Query Dependent LCIRC (QD-LCIRC) significantly improves LLM's ability to manage extended contexts, making it well-suited for tasks that require both comprehensive context understanding and query relevance.
The increasing complexity of computer science research projects demands more effective tools for deploying code repositories. Large Language Models (LLMs), such as Anthropic Claude and Meta Llama, have demonstrated significant advancements across various fields of computer science research, including the automation of diverse software engineering tasks. To evaluate the effectiveness of LLMs in handling complex code development tasks of research projects, particularly for NLP/CV/AI/ML/DM topics, we introduce CSR-Bench, a benchmark for Computer Science Research projects. This benchmark assesses LLMs from various aspects including accuracy, efficiency, and deployment script quality, aiming to explore their potential in conducting computer science research autonomously. We also introduce a novel framework, CSR-Agents, that utilizes multiple LLM agents to automate the deployment of GitHub code repositories of computer science research projects. Specifically, by checking instructions from markdown files and interpreting repository structures, the model generates and iteratively improves bash commands that set up the experimental environments and deploy the code to conduct research tasks. Preliminary results from CSR-Bench indicate that LLM agents can significantly enhance the workflow of repository deployment, thereby boosting developer productivity and improving the management of developmental workflows.
Large language models (LLMs) have shown promise in many natural language understanding tasks, including content moderation. However, these models can be expensive to query in real-time and do not allow for a community-specific approach to content moderation. To address these challenges, we explore the use of open-source small language models (SLMs) for community-specific content moderation tasks. We fine-tune and evaluate SLMs (less than 15B parameters) by comparing their performance against much larger open- and closed-sourced models in both a zero-shot and few-shot setting. Using 150K comments from 15 popular Reddit communities, we find that SLMs outperform zero-shot LLMs at content moderation -- 11.5% higher accuracy and 25.7% higher recall on average across all communities. Moreover, few-shot in-context learning leads to only a marginal increase in the performance of LLMs, still lacking compared to SLMs. We further show the promise of cross-community content moderation, which has implications for new communities and the development of cross-platform moderation techniques. Finally, we outline directions for future work on language model based content moderation. Code and models can be found at https://github.com/AGoyal0512/SLM-Mod.
Conceptual combination is a cognitive process that merges basic concepts, enabling the creation of complex expressions. During this process, the properties of combination (e.g., the whiteness of a peeled apple) can be inherited from basic concepts, newly emerge, or be canceled. However, previous studies have evaluated a limited set of properties and have not examined the generative process. To address this gap, we introduce the Conceptual Combination with Property Type dataset (CCPT), which consists of 12.3K annotated triplets of noun phrases, properties, and property types. Using CCPT, we establish three types of tasks to evaluate LLMs for conceptual combination thoroughly. Our key findings are threefold: (1) Our automatic metric grading property emergence and cancellation closely corresponds with human judgments. (2) LLMs, including OpenAI's o1, struggle to generate noun phrases which possess given emergent properties. (3) Our proposed method, inspired by cognitive psychology model that explains how relationships between concepts are formed, improves performances in all generative tasks. The dataset and experimental code are available at https://github.com/seokwon99/CCPT.git.
Data harmonization is an essential task that entails integrating datasets from diverse sources. Despite years of research in this area, it remains a time-consuming and challenging task due to schema mismatches, varying terminologies, and differences in data collection methodologies. This paper presents the case for agentic data harmonization as a means to both empower experts to harmonize their data and to streamline the process. We introduce Harmonia, a system that combines LLM-based reasoning, an interactive user interface, and a library of data harmonization primitives to automate the synthesis of data harmonization pipelines. We demonstrate Harmonia in a clinical data harmonization scenario, where it helps to interactively create reusable pipelines that map datasets to a standard format. Finally, we discuss challenges and open problems, and suggest research directions for advancing our vision.
The prototyping of computer games, particularly card games, requires extensive human effort in creative ideation and gameplay evaluation. Recent advances in Large Language Models (LLMs) offer opportunities to automate and streamline these processes. However, it remains challenging for LLMs to design novel game mechanics beyond existing databases, generate consistent gameplay environments, and develop scalable gameplay AI for large-scale evaluations. This paper addresses these challenges by introducing a comprehensive automated card game prototyping framework. The approach highlights a graph-based indexing method for generating novel game designs, an LLM-driven system for consistent game code generation validated by gameplay records, and a gameplay AI constructing method that uses an ensemble of LLM-generated action-value functions optimized through self-play. These contributions aim to accelerate card game prototyping, reduce human labor, and lower barriers to entry for game developers.
Large Language Model (LLM) inference, where a trained model generates text one word at a time in response to user prompts, is a computationally intensive process requiring efficient scheduling to optimize latency and resource utilization. A key challenge in LLM inference is the management of the Key-Value (KV) cache, which reduces redundant computations but introduces memory constraints. In this work, we model LLM inference with KV cache constraints theoretically and propose novel batching and scheduling algorithms that minimize inference latency while effectively managing the KV cache's memory. We analyze both semi-online and fully online scheduling models, and our results are threefold. First, we provide a polynomial-time algorithm that achieves exact optimality in terms of average latency in the semi-online prompt arrival model. Second, in the fully online case with a stochastic prompt arrival, we introduce an efficient online scheduling algorithm with constant regret. Third, we prove that no algorithm (deterministic or randomized) can achieve a constant competitive ratio in fully online adversarial settings. Our empirical evaluations on a public LLM inference dataset, using the Llama-70B model on A100 GPUs, show that our approach significantly outperforms benchmark algorithms used currently in practice, achieving lower latency while reducing energy consumption. Overall, our results offer a path toward more sustainable and cost-effective LLM deployment.
As large language models (LLMs) expand into performing as agents for real-world applications beyond traditional NLP tasks, evaluating their robustness becomes increasingly important. However, existing benchmarks often overlook critical dimensions like cultural and social awareness. To address these, we introduce CASA, a benchmark designed to assess LLM agents' sensitivity to cultural and social norms across two web-based tasks: online shopping and social discussion forums. Our approach evaluates LLM agents' ability to detect and appropriately respond to norm-violating user queries and observations. Furthermore, we propose a comprehensive evaluation framework that measures awareness coverage, helpfulness in managing user queries, and the violation rate when facing misleading web content. Experiments show that current LLMs perform significantly better in non-agent than in web-based agent environments, with agents achieving less than 10% awareness coverage and over 40% violation rates. To improve performance, we explore two methods: prompting and fine-tuning, and find that combining both methods can offer complementary advantages -- fine-tuning on culture-specific datasets significantly enhances the agents' ability to generalize across different regions, while prompting boosts the agents' ability to navigate complex tasks. These findings highlight the importance of constantly benchmarking LLM agents' cultural and social awareness during the development cycle.
Automatic analysis of user reviews to understand user sentiments toward app functionality (i.e. app features) helps align development efforts with user expectations and needs. Recent advances in Large Language Models (LLMs) such as ChatGPT have shown impressive performance on several new tasks without updating the model's parameters i.e. using zero or a few labeled examples, but the capabilities of LLMs are yet unexplored for feature-specific sentiment analysis. The goal of our study is to explore the capabilities of LLMs to perform feature-specific sentiment analysis of user reviews. This study compares the performance of state-of-the-art LLMs, including GPT-4, ChatGPT, and different variants of Llama-2 chat, against previous approaches for extracting app features and associated sentiments in zero-shot, 1-shot, and 5-shot scenarios. The results indicate that GPT-4 outperforms the rule-based SAFE by 17% in f1-score for extracting app features in the zero-shot scenario, with 5-shot further improving it by 6%. However, the fine-tuned RE-BERT exceeds GPT-4 by 6% in f1-score. For predicting positive and neutral sentiments, GPT-4 achieves f1-scores of 76% and 45% in the zero-shot setting, which improve by 7% and 23% in the 5-shot setting, respectively. Our study conducts a thorough evaluation of both proprietary and open-source LLMs to provide an objective assessment of their performance in extracting feature-sentiment pairs.
Decision-making is a complex process requiring diverse abilities, making it an excellent framework for evaluating Large Language Models (LLMs). Researchers have examined LLMs' decision-making through the lens of Game Theory. However, existing evaluation mainly focus on two-player scenarios where an LLM competes against another. Additionally, previous benchmarks suffer from test set leakage due to their static design. We introduce GAMA($\gamma$)-Bench, a new framework for evaluating LLMs' Gaming Ability in Multi-Agent environments. It includes eight classical game theory scenarios and a dynamic scoring scheme specially designed to quantitatively assess LLMs' performance. $\gamma$-Bench allows flexible game settings and adapts the scoring system to different game parameters, enabling comprehensive evaluation of robustness, generalizability, and strategies for improvement. Our results indicate that GPT-3.5 demonstrates strong robustness but limited generalizability, which can be enhanced using methods like Chain-of-Thought. We also evaluate 13 LLMs from 6 model families, including GPT-3.5, GPT-4, Gemini, LLaMA-3.1, Mixtral, and Qwen-2. Gemini-1.5-Pro outperforms others, scoring of $69.8$ out of $100$, followed by LLaMA-3.1-70B ($65.9$) and Mixtral-8x22B ($62.4$). Our code and experimental results are publicly available at https://github.com/CUHK-ARISE/GAMABench.
Modern techniques in Content-based Recommendation (CBR) leverage item content information to provide personalized services to users, but suffer from resource-intensive training on large datasets. To address this issue, we explore the dataset condensation for textual CBR in this paper. The goal of dataset condensation is to synthesize a small yet informative dataset, upon which models can achieve performance comparable to those trained on large datasets. While existing condensation approaches are tailored to classification tasks for continuous data like images or embeddings, direct application of them to CBR has limitations. To bridge this gap, we investigate efficient dataset condensation for content-based recommendation. Inspired by the remarkable abilities of large language models (LLMs) in text comprehension and generation, we leverage LLMs to empower the generation of textual content during condensation. To handle the interaction data involving both users and items, we devise a dual-level condensation method: content-level and user-level. At content-level, we utilize LLMs to condense all contents of an item into a new informative title. At user-level, we design a clustering-based synthesis module, where we first utilize LLMs to extract user interests. Then, the user interests and user embeddings are incorporated to condense users and generate interactions for condensed users. Notably, the condensation paradigm of this method is forward and free from iterative optimization on the synthesized dataset. Extensive empirical findings from our study, conducted on three authentic datasets, substantiate the efficacy of the proposed method. Particularly, we are able to approximate up to 97% of the original performance while reducing the dataset size by 95% (i.e., on dataset MIND).
Large language models (LLMs) have facilitated a wide range of applications with distinct service-level objectives (SLOs), from latency-sensitive online tasks like interactive chatbots to throughput-oriented offline workloads like document summarization. The existing deployment model, which dedicates machines to each workload, simplifies SLO management but often leads to poor resource utilization. This paper introduces HyGen, an interference-aware LLM serving system that enables efficient co-location of online and offline workloads while preserving latency requirements. HyGen incorporates two key innovations: (1) performance control mechanisms, including a latency predictor to estimate batch execution time and an SLO-aware profiler to quantify latency interference, and (2) SLO-aware offline scheduling policies that maximize serving throughput and prevent starvation, without compromising online serving latency. Our evaluation on production workloads shows that HyGen achieves up to 3.87x overall throughput and 5.84x offline throughput gains over online and hybrid serving baselines, respectively, while strictly satisfying latency SLOs.
Anomaly event detection plays a crucial role in various real-world applications. However, current approaches predominantly rely on supervised learning, which faces significant challenges: the requirement for extensive labeled training data and lack of interpretability in decision-making processes. To address these limitations, we present a training-free framework that integrates open-set object detection with symbolic regression, powered by Large Language Models (LLMs) for efficient symbolic pattern discovery. The LLMs guide the symbolic reasoning process, establishing logical relationships between detected entities. Through extensive experiments across multiple domains, our framework demonstrates several key advantages: (1) achieving superior detection accuracy through direct reasoning without any training process; (2) providing highly interpretable logical expressions that are readily comprehensible to humans; and (3) requiring minimal annotation effort - approximately 1% of the data needed by traditional training-based methods.To facilitate comprehensive evaluation and future research, we introduce two datasets: a large-scale private dataset containing over 110,000 annotated images covering various anomaly scenarios including construction site safety violations, illegal fishing activities, and industrial hazards, along with a public benchmark dataset of 5,000 samples with detailed anomaly event annotations. Code is available at here.
The growing use of large language models (LLMs) has raised concerns regarding their safety. While many studies have focused on English, the safety of LLMs in Arabic, with its linguistic and cultural complexities, remains under-explored. Here, we aim to bridge this gap. In particular, we present an Arab-region-specific safety evaluation dataset consisting of 5,799 questions, including direct attacks, indirect attacks, and harmless requests with sensitive words, adapted to reflect the socio-cultural context of the Arab world. To uncover the impact of different stances in handling sensitive and controversial topics, we propose a dual-perspective evaluation framework. It assesses the LLM responses from both governmental and opposition viewpoints. Experiments over five leading Arabic-centric and multilingual LLMs reveal substantial disparities in their safety performance. This reinforces the need for culturally specific datasets to ensure the responsible deployment of LLMs.
Low-rank optimization has emerged as a promising approach to enabling memory-efficient training of large language models (LLMs). Existing low-rank optimization methods typically project gradients onto a low-rank subspace, reducing the memory cost of storing optimizer states. A key challenge in these methods is identifying suitable subspaces to ensure an effective optimization trajectory. Most existing approaches select the dominant subspace to preserve gradient information, as this intuitively provides the best approximation. However, we find that in practice, the dominant subspace stops changing during pretraining, thereby constraining weight updates to similar subspaces. In this paper, we propose importance sampling subspace selection (I3S) for low-rank optimization, which theoretically offers a comparable convergence rate to the dominant subspace approach. Empirically, we demonstrate that I3S significantly outperforms previous methods in LLM pretraining tasks.
This paper presents a comprehensive framework for testing and evaluating quality characteristics of Large Language Model (LLM) systems enhanced with Retrieval-Augmented Generation (RAG) in tourism applications. Through systematic empirical evaluation of three different LLM variants across multiple parameter configurations, we demonstrate the effectiveness of our testing methodology in assessing both functional correctness and extra-functional properties. Our framework implements 17 distinct metrics that encompass syntactic analysis, semantic evaluation, and behavioral evaluation through LLM judges. The study reveals significant information about how different architectural choices and parameter configurations affect system performance, particularly highlighting the impact of temperature and top-p parameters on response quality. The tests were carried out on a tourism recommendation system for the V\"armland region, utilizing standard and RAG-enhanced configurations. The results indicate that the newer LLM versions show modest improvements in performance metrics, though the differences are more pronounced in response length and complexity rather than in semantic quality. The research contributes practical insights for implementing robust testing practices in LLM-RAG systems, providing valuable guidance to organizations deploying these architectures in production environments.
With the increasing adoption of large language models (LLMs) in education, concerns about inherent biases in these models have gained prominence. We evaluate LLMs for bias in the personalized educational setting, specifically focusing on the models' roles as "teachers." We reveal significant biases in how models generate and select educational content tailored to different demographic groups, including race, ethnicity, sex, gender, disability status, income, and national origin. We introduce and apply two bias score metrics--Mean Absolute Bias (MAB) and Maximum Difference Bias (MDB)--to analyze 9 open and closed state-of-the-art LLMs. Our experiments, which utilize over 17,000 educational explanations across multiple difficulty levels and topics, uncover that models potentially harm student learning by both perpetuating harmful stereotypes and reversing them. We find that bias is similar for all frontier models, with the highest MAB along income levels while MDB is highest relative to both income and disability status. For both metrics, we find the lowest bias exists for sex/gender and race/ethnicity.
With the rapid development of evaluation datasets to assess LLMs understanding across a wide range of subjects and domains, identifying a suitable language understanding benchmark has become increasingly challenging. In this work, we explore LLM evaluation challenges for low-resource language understanding and introduce \proverbeval, LLM evaluation benchmark for low-resource languages, focusing on low-resource language understanding in culture-specific scenarios. We benchmark various LLMs and explore factors that create variability in the benchmarking process. We observed performance variances of up to 50\%, depending on the order in which answer choices were presented in multiple-choice tasks. Native language proverb descriptions significantly improve tasks such as proverb generation, contributing to improved outcomes. Additionally, monolingual evaluations consistently outperformed their cross-lingual counterparts in generation tasks. We argue that special attention must be given to the order of choices, the choice of prompt language, task variability, and generation tasks when creating LLM evaluation benchmarks. Evaluation data available at https://huggingface.co/datasets/israel/ProverbEval, evaluation code https://github.com/EthioNLP/EthioProverbEval.
The increasing volume of drug combinations in modern therapeutic regimens needs reliable methods for predicting drug-drug interactions (DDIs). While Large Language Models (LLMs) have revolutionized various domains, their potential in pharmaceutical research, particularly in DDI prediction, remains largely unexplored. This study thoroughly investigates LLMs' capabilities in predicting DDIs by uniquely processing molecular structures (SMILES), target organisms, and gene interaction data as raw text input from the latest DrugBank dataset. We evaluated 18 different LLMs, including proprietary models (GPT-4, Claude, Gemini) and open-source variants (from 1.5B to 72B parameters), first assessing their zero-shot capabilities in DDI prediction. We then fine-tuned selected models (GPT-4, Phi-3.5 2.7B, Qwen-2.5 3B, Gemma-2 9B, and Deepseek R1 distilled Qwen 1.5B) to optimize their performance. Our comprehensive evaluation framework included validation across 13 external DDI datasets, comparing against traditional approaches such as l2-regularized logistic regression. Fine-tuned LLMs demonstrated superior performance, with Phi-3.5 2.7B achieving a sensitivity of 0.978 in DDI prediction, with an accuracy of 0.919 on balanced datasets (50% positive, 50% negative cases). This result represents an improvement over both zero-shot predictions and state-of-the-art machine-learning methods used for DDI prediction. Our analysis reveals that LLMs can effectively capture complex molecular interaction patterns and cases where drug pairs target common genes, making them valuable tools for practical applications in pharmaceutical research and clinical settings.
Large language models (LLMs) are increasingly capable of completing knowledge intensive tasks by recalling information from a static pretraining corpus. Here we are concerned with LLMs in the context of evolving data requirements. For instance: batches of new data that are introduced periodically; subsets of data with user-based access controls; or requirements on dynamic removal of documents with guarantees that associated knowledge cannot be recalled. We wish to satisfy these requirements while at the same time ensuring a model does not forget old information when new data becomes available. To address these issues, we introduce AdapterSwap, a training and inference scheme that organizes knowledge from a data collection into a set of low-rank adapters, which are dynamically composed during inference. Our experiments demonstrate AdapterSwap's ability to support efficient continual learning, while also enabling organizations to have fine-grained control over data access and deletion.
African American English (AAE) presents unique challenges in natural language processing (NLP). This research systematically compares the performance of available NLP models--rule-based, transformer-based, and large language models (LLMs)--capable of identifying key grammatical features of AAE, namely Habitual Be and Multiple Negation. These features were selected for their distinct grammatical complexity and frequency of occurrence. The evaluation involved sentence-level binary classification tasks, using both zero-shot and few-shot strategies. The analysis reveals that while LLMs show promise compared to the baseline, they are influenced by biases such as recency and unrelated features in the text such as formality. This study highlights the necessity for improved model training and architectural adjustments to better accommodate AAE's unique linguistic characteristics. Data and code are available.
In this paper, we introduce a black-box prompt optimization method that uses an attacker LLM agent to uncover higher levels of memorization in a victim agent, compared to what is revealed by prompting the target model with the training data directly, which is the dominant approach of quantifying memorization in LLMs. We use an iterative rejection-sampling optimization process to find instruction-based prompts with two main characteristics: (1) minimal overlap with the training data to avoid presenting the solution directly to the model, and (2) maximal overlap between the victim model's output and the training data, aiming to induce the victim to spit out training data. We observe that our instruction-based prompts generate outputs with 23.7% higher overlap with training data compared to the baseline prefix-suffix measurements. Our findings show that (1) instruction-tuned models can expose pre-training data as much as their base-models, if not more so, (2) contexts other than the original training data can lead to leakage, and (3) using instructions proposed by other LLMs can open a new avenue of automated attacks that we should further study and explore. The code can be found at https://github.com/Alymostafa/Instruction_based_attack .
This paper presents HamRaz, a novel Persian-language mental health dataset designed for Person-Centered Therapy (PCT) using Large Language Models (LLMs). Despite the growing application of LLMs in AI-driven psychological counseling, existing datasets predominantly focus on Western and East Asian contexts, overlooking cultural and linguistic nuances essential for effective Persian-language therapy. To address this gap, HamRaz combines script-based dialogues with adaptive LLM role-playing, ensuring coherent and dynamic therapy interactions. We also introduce HamRazEval, a dual evaluation framework that measures conversational quality and therapeutic effectiveness using General Dialogue Metrics and the Barrett-Lennard Relationship Inventory (BLRI). Experimental results show HamRaz outperforms conventional Script Mode and Two-Agent Mode, producing more empathetic, context-aware, and realistic therapy sessions. By releasing HamRaz, we contribute a culturally adapted, LLM-driven resource to advance AI-powered psychotherapy research in diverse communities.
Large language models (LLMs) can store a significant amount of factual knowledge in their parameters. However, their parametric knowledge may conflict with the information provided in the context -- this phenomenon, known as \emph{context-memory knowledge conflicts}, can lead to undesirable model behaviour, such as reliance on outdated or incorrect information. Analysing the internal activations of LLMs, we find that they can internally register the signals of knowledge conflict at mid-layers. Such signals allow us to detect whether a knowledge conflict occurs and use \emph{inference-time} intervention strategies to resolve it. In this work, we propose \textsc{SpARE}, a \emph{training-free} representation engineering method that uses pre-trained sparse auto-encoders (SAEs) to control the knowledge selection behaviour of LLMs. \textsc{SpARE} identifies the functional features that control the knowledge selection behaviours and applies them to edit the internal activations of LLMs at inference time. Our experimental results show that \textsc{SpARE} can effectively control the usage of either knowledge source to resolve knowledge conflict in open-domain question-answering tasks, surpassing existing representation engineering methods ($+10\%$) as well as contrastive decoding methods ($+15\%$).
Large Language Model (LLM) Agents have demonstrated remarkable capabilities in task automation and intelligent decision-making, driving the widespread adoption of agent development frameworks such as LangChain and AutoGen. However, these frameworks predominantly serve developers with extensive technical expertise - a significant limitation considering that only 0.03 % of the global population possesses the necessary programming skills. This stark accessibility gap raises a fundamental question: Can we enable everyone, regardless of technical background, to build their own LLM agents using natural language alone? To address this challenge, we introduce MetaChain-a Fully-Automated and highly Self-Developing framework that enables users to create and deploy LLM agents through Natural Language Alone. Operating as an autonomous Agent Operating System, MetaChain comprises four key components: i) Agentic System Utilities, ii) LLM-powered Actionable Engine, iii) Self-Managing File System, and iv) Self-Play Agent Customization module. This lightweight yet powerful system enables efficient and dynamic creation and modification of tools, agents, and workflows without coding requirements or manual intervention. Beyond its code-free agent development capabilities, MetaChain also serves as a versatile multi-agent system for General AI Assistants. Comprehensive evaluations on the GAIA benchmark demonstrate MetaChain's effectiveness in generalist multi-agent tasks, surpassing existing state-of-the-art methods. Furthermore, MetaChain's Retrieval-Augmented Generation (RAG)-related capabilities have shown consistently superior performance compared to many alternative LLM-based solutions.
Training-free embedding methods directly leverage pretrained large language models (LLMs) to embed text, bypassing the costly and complex procedure of contrastive learning. Previous training-free embedding methods have mainly focused on optimizing embedding prompts and have overlooked the benefits of utilizing the generative abilities of LLMs. We propose a novel method, GenEOL, which uses LLMs to generate diverse transformations of a sentence that preserve its meaning, and aggregates the resulting embeddings of these transformations to enhance the overall sentence embedding. GenEOL significantly outperforms the existing training-free embedding methods by an average of 2.85 points across several LLMs on the sentence semantic text similarity (STS) benchmark. GenEOL also achieves notable gains in clustering, reranking, and pair-classification tasks from the MTEB benchmark. Additionally, GenEOL stabilizes representation quality across LLM layers and remains robust to perturbations of embedding prompts.
In this paper, we introduce and apply Operations Research Question Answering (ORQA), a new benchmark designed to assess the generalization capabilities of Large Language Models (LLMs) in the specialized technical domain of Operations Research (OR). This benchmark evaluates whether LLMs can emulate the knowledge and reasoning skills of OR experts when confronted with diverse and complex optimization problems. The dataset, developed by OR experts, features real-world optimization problems that demand multistep reasoning to construct their mathematical models. Our evaluations of various open source LLMs, such as LLaMA 3.1, DeepSeek, and Mixtral, reveal their modest performance, highlighting a gap in their ability to generalize to specialized technical domains. This work contributes to the ongoing discourse on LLMs generalization capabilities, offering valuable insights for future research in this area. The dataset and evaluation code are publicly available.
Multiple heads decoding accelerates the inference of Large Language Models (LLMs) by predicting next several tokens simultaneously. It generates and verifies multiple candidate sequences in parallel via tree attention with a fixed structure. In this paper, we replace the fixed tree attention with dynamic tree attention on multiple head decoding, specifically in the context of MEDUSA. We propose a simple and low complexity strategy to generate candidates and construct the dynamic tree structure. Preliminary experiments show that the proposed method improves the decoding efficiency of multiple head decoding for LLMs while maintaining the generation quality. This result demonstrates the potential for improvement of multiple head decoding in candidate generation.
Fine-tuning large language models (LLMs) requires significant memory, often exceeding the capacity of a single GPU. A common solution to this memory challenge is offloading compute and data from the GPU to the CPU. However, this approach is hampered by the limited bandwidth of commodity hardware, which constrains communication between the CPU and GPU, and by slower matrix multiplications on the CPU. In this paper, we present an offloading framework, LSP-Offload, that enables near-native speed LLM fine-tuning on commodity hardware through learned sparse projectors. Our data-driven approach involves learning efficient sparse compressors that minimize communication with minimal precision loss. Additionally, we introduce a novel layer-wise communication schedule to maximize parallelism between communication and computation. As a result, our framework can fine-tune a 1.3 billion parameter model on a 4GB laptop GPU and a 6.7 billion parameter model on a 24GB NVIDIA RTX 4090 GPU. Compared to state-of-the-art offloading frameworks, our approach reduces end-to-end fine-tuning time by 33.1%-62.5% when converging to the same accuracy. We open source our framework at https://github.com/gulang2019/LSP-Offload.
We introduce CompAct, a technique that reduces peak memory utilization on GPU by 25-30% for pretraining and 50% for fine-tuning of LLMs. Peak device memory is a major limiting factor in training LLMs, with various recent works aiming to reduce model memory. However most works don't target the largest component of allocated memory during training: the model's compute graph, which is stored for the backward pass. By storing low-rank, compressed activations to be used in the backward pass we greatly reduce the required memory, unlike previous methods which only reduce optimizer overheads or the number of trained parameters. Our compression uses random projection matrices, thus avoiding additional memory overheads. Comparisons with previous techniques for either pretraining or fine-tuning show that CompAct substantially improves existing compute-performance tradeoffs. We expect CompAct's savings to scale even higher for larger models.
One of the most important yet onerous tasks in the academic peer-reviewing process is composing meta-reviews, which involves assimilating diverse opinions from multiple expert peers, formulating one's self-judgment as a senior expert, and then summarizing all these perspectives into a concise holistic overview to make an overall recommendation. This process is time-consuming and can be compromised by human factors like fatigue, inconsistency, missing tiny details, etc. Given the latest major developments in Large Language Models (LLMs), it is very compelling to rigorously study whether LLMs can help metareviewers perform this important task better. In this paper, we perform a case study with three popular LLMs, i.e., GPT-3.5, LLaMA2, and PaLM2, to assist meta-reviewers in better comprehending multiple experts perspectives by generating a controlled multi-perspective summary (MPS) of their opinions. To achieve this, we prompt three LLMs with different types/levels of prompts based on the recently proposed TELeR taxonomy. Finally, we perform a detailed qualitative study of the MPSs generated by the LLMs and report our findings.
Materials discovery and design are essential for advancing technology across various industries by enabling the development of application-specific materials. Recent research has leveraged Large Language Models (LLMs) to accelerate this process. We explore the potential of LLMs to generate viable hypotheses that, once validated, can expedite materials discovery. Collaborating with materials science experts, we curated a novel dataset from recent journal publications, featuring real-world goals, constraints, and methods for designing real-world applications. Using this dataset, we test LLM-based agents that generate hypotheses for achieving given goals under specific constraints. To assess the relevance and quality of these hypotheses, we propose a novel scalable evaluation metric that emulates the process a materials scientist would use to evaluate a hypothesis critically. Our curated dataset, proposed method, and evaluation framework aim to advance future research in accelerating materials discovery and design with LLMs.
Reasoning abilities of LLMs have been a key focus in recent years. One challenging reasoning domain with interesting nuances is legal reasoning, which requires careful application of rules, and precedents while balancing deductive and analogical reasoning, and conflicts between rules. Although there have been a few works on using LLMs for legal reasoning, their focus has been on overall accuracy. In this paper, we dig deeper to do a step-by-step analysis and figure out where they commit errors. We use the college-level Multiple Choice Question-Answering (MCQA) task from the \textit{Civil Procedure} dataset and propose a new error taxonomy derived from initial manual analysis of reasoning chains with respect to several LLMs, including two objective measures: soundness and correctness scores. We then develop an LLM-based automated evaluation framework to identify reasoning errors and evaluate the performance of LLMs. The computation of soundness and correctness on the dataset using the auto-evaluator framework reveals several interesting insights. Furthermore, we show that incorporating the error taxonomy as feedback in popular prompting techniques marginally increases LLM performance. Our work will also serve as an evaluation framework that can be used in detailed error analysis of reasoning chains for logic-intensive complex tasks.
Speech-to-text technologies have been shown to improve text input efficiency and potentially lower the barriers to writing. Recent LLM-assisted dictation tools aim to support writing with speech by bridging the gaps between speaking and traditional writing. This case study reports on the real-world writing experiences of twelve academic or creative writers using one such tool, Rambler, to write various pieces such as blog posts, diaries, screenplays, notes, or fictional stories, etc. Through a ten-day diary study, we identified the participants' in-context writing strategies using Rambler, such as how they expanded from an outline or organized their loose thoughts for different writing goals. The interviews uncovered the psychological and productivity affordances of writing with speech, pointing to future directions of designing for this writing modality and the utilization of AI support.
Large language models (LLMs) have shown high agreement with human raters across a variety of tasks, demonstrating potential to ease the challenges of human data collection. In computational social science (CSS), researchers are increasingly leveraging LLM annotations to complement slow and expensive human annotations. Still, guidelines for collecting and using LLM annotations, without compromising the validity of downstream conclusions, remain limited. We introduce Confidence-Driven Inference: a method that combines LLM annotations and LLM confidence indicators to strategically select which human annotations should be collected, with the goal of producing accurate statistical estimates and provably valid confidence intervals while reducing the number of human annotations needed. Our approach comes with safeguards against LLM annotations of poor quality, guaranteeing that the conclusions will be both valid and no less accurate than if we only relied on human annotations. We demonstrate the effectiveness of Confidence-Driven Inference over baselines in statistical estimation tasks across three CSS settings--text politeness, stance, and bias--reducing the needed number of human annotations by over 25% in each. Although we use CSS settings for demonstration, Confidence-Driven Inference can be used to estimate most standard quantities across a broad range of NLP problems.
Large language models (LLMs) have significantly advanced human language understanding and generation, with pretraining data quality and organization being crucial to their performance. Multi-stage pretraining is a promising approach, but existing methods often lack quantitative criteria for data partitioning and instead rely on intuitive heuristics. In this paper, we propose the novel Four-quadRAnt Multi-stage prEtraining Strategy (FRAMES), guided by the established principle of organizing the pretraining process into four stages to achieve significant loss reductions four times. This principle is grounded in two key findings: first, training on high Perplexity (PPL) data followed by low PPL data, and second, training on low PPL difference (PD) data followed by high PD data, both causing the loss to drop significantly twice and performance enhancements. By partitioning data into four quadrants and strategically organizing them, FRAMES achieves a remarkable 16.8% average improvement over random sampling across MMLU and CMMLU, effectively boosting LLM performance.
Large Language Models (LLMs) have shown remarkable progress, but their real-world application necessitates reliable calibration. This study conducts a comprehensive analysis of calibration degradation of LLMs across four dimensions: models, calibration metrics, tasks, and confidence extraction methods. Initial analysis showed that the relationship between alignment and calibration is not always a trade-off, but under stricter analysis conditions, we found the alignment process consistently harms calibration. This highlights the need for (1) a careful approach when measuring model confidences and calibration errors and (2) future research into algorithms that can help LLMs to achieve both instruction-following and calibration without sacrificing either.
In this paper, we investigate whether current state-of-the-art large language models (LLMs) are effective as AI tutors and whether they demonstrate pedagogical abilities necessary for good AI tutoring in educational dialogues. Previous efforts towards evaluation have been limited to subjective protocols and benchmarks. To bridge this gap, we propose a unified evaluation taxonomy with eight pedagogical dimensions based on key learning sciences principles, which is designed to assess the pedagogical value of LLM-powered AI tutor responses grounded in student mistakes or confusions in the mathematical domain. We release MRBench - a new evaluation benchmark containing 192 conversations and 1,596 responses from seven state-of-the-art LLM-based and human tutors, providing gold annotations for eight pedagogical dimensions. We assess reliability of the popular Prometheus2 and Llama-3.1-8B LLMs as evaluators and analyze each tutor's pedagogical abilities, highlighting which LLMs are good tutors and which ones are more suitable as question-answering systems. We believe that the presented taxonomy, benchmark, and human-annotated labels will streamline the evaluation process and help track the progress in AI tutors' development.
We address the challenging problem of fine-grained text-driven human motion generation. Existing works generate imprecise motions that fail to accurately capture relationships specified in text due to: (1) lack of effective text parsing for detailed semantic cues regarding body parts, (2) not fully modeling linguistic structures between words to comprehend text comprehensively. To tackle these limitations, we propose a novel fine-grained framework Fg-T2M++ that consists of: (1) an LLMs semantic parsing module to extract body part descriptions and semantics from text, (2) a hyperbolic text representation module to encode relational information between text units by embedding the syntactic dependency graph into hyperbolic space, and (3) a multi-modal fusion module to hierarchically fuse text and motion features. Extensive experiments on HumanML3D and KIT-ML datasets demonstrate that Fg-T2M++ outperforms SOTA methods, validating its ability to accurately generate motions adhering to comprehensive text semantics.
Large Language Models (LLMs) have been recently adopted in interactive systems requiring communication. As the false belief in a model can harm the usability of such systems, LLMs should not have cognitive biases that humans have. Psychologists especially focus on the False Consensus Effect (FCE), a cognitive bias where individuals overestimate the extent to which others share their beliefs or behaviors, because FCE can distract smooth communication by posing false beliefs. However, previous studies have less examined FCE in LLMs thoroughly, which needs more consideration of confounding biases, general situations, and prompt changes. Therefore, in this paper, we conduct two studies to examine the FCE phenomenon in LLMs. In Study 1, we investigate whether LLMs have FCE. In Study 2, we explore how various prompting styles affect the demonstration of FCE. As a result of these studies, we identified that popular LLMs have FCE. Also, the result specifies the conditions when FCE becomes more or less prevalent compared to normal usage.
Large language models (LLMs) provide effective solutions in various application scenarios, with the support of retrieval-augmented generation (RAG). However, developers face challenges in integrating LLM and RAG into software systems, due to lacking interface specifications, various requirements from software context, and complicated system management. In this paper, we have conducted a comprehensive study of 100 open-source applications that incorporate LLMs with RAG support, and identified 18 defect patterns. Our study reveals that 77% of these applications contain more than three types of integration defects that degrade software functionality, efficiency, and security. Guided by our study, we propose systematic guidelines for resolving these defects in software life cycle. We also construct an open-source defect library Hydrangea.
Although Deep Reinforcement Learning (DRL) and Large Language Models (LLMs) each show promise in addressing decision-making challenges in autonomous driving, DRL often suffers from high sample complexity, while LLMs have difficulty ensuring real-time decision making. To address these limitations, we propose TeLL-Drive, a hybrid framework that integrates an Teacher LLM to guide an attention-based Student DRL policy. By incorporating risk metrics, historical scenario retrieval, and domain heuristics into context-rich prompts, the LLM produces high-level driving strategies through chain-of-thought reasoning. A self-attention mechanism then fuses these strategies with the DRL agent's exploration, accelerating policy convergence and boosting robustness across diverse driving conditions. Our experimental results, evaluated across multiple traffic scenarios, show that TeLL-Drive outperforms existing baseline methods, including other LLM-based approaches, in terms of success rates, average returns, and real-time feasibility. Ablation studies underscore the importance of each model component, especially the synergy between the attention mechanism and LLM-driven guidance. These findings suggest that TeLL-Drive significantly enhances both the adaptability and safety of autonomous driving systems, while offering a more efficient and scalable approach for policy learning. Full validation results are available on our website.
While recent efforts have begun integrating large language models (LLMs) into foreign language education (FLE), they often rely on traditional approaches to learning tasks without fully embracing educational methodologies, thus lacking adaptability to language learning. To address this gap, we argue that LLMs have the potential to serve as effective tutors in FLE. Specifically, LLMs can play three critical roles: (1) as data enhancers, improving the creation of learning materials or serving as student simulations; (2) as task predictors, serving as learner assessment or optimizing learning pathway; and (3) as agents, enabling personalized and inclusive education. We encourage interdisciplinary research to explore these roles, fostering innovation while addressing challenges and risks, ultimately advancing FLE through the thoughtful integration of LLMs.
Developing intelligent agents for long-term cooperation in dynamic open-world scenarios is a major challenge in multi-agent systems. Traditional Multi-agent Reinforcement Learning (MARL) frameworks like centralized training decentralized execution (CTDE) struggle with scalability and flexibility. They require centralized long-term planning, which is difficult without custom reward functions, and face challenges in processing multi-modal data. CTDE approaches also assume fixed cooperation strategies, making them impractical in dynamic environments where agents need to adapt and plan independently. To address decentralized multi-agent cooperation, we propose Decentralized Adaptive Knowledge Graph Memory and Structured Communication System (DAMCS) in a novel Multi-agent Crafter environment. Our generative agents, powered by Large Language Models (LLMs), are more scalable than traditional MARL agents by leveraging external knowledge and language for long-term planning and reasoning. Instead of fully sharing information from all past experiences, DAMCS introduces a multi-modal memory system organized as a hierarchical knowledge graph and a structured communication protocol to optimize agent cooperation. This allows agents to reason from past interactions and share relevant information efficiently. Experiments on novel multi-agent open-world tasks show that DAMCS outperforms both MARL and LLM baselines in task efficiency and collaboration. Compared to single-agent scenarios, the two-agent scenario achieves the same goal with 63% fewer steps, and the six-agent scenario with 74% fewer steps, highlighting the importance of adaptive memory and structured communication in achieving long-term goals. We publicly release our project at: https://happyeureka.github.io/damcs.
Large Language Models (LLMs) commonly rely on explicit refusal prefixes for safety, making them vulnerable to prefix injection attacks. We introduce HumorReject, a novel data-driven approach that reimagines LLM safety by decoupling it from refusal prefixes through humor as an indirect refusal strategy. Rather than explicitly rejecting harmful instructions, HumorReject responds with contextually appropriate humor that naturally defuses potentially dangerous requests. Our approach effectively addresses common "over-defense" issues while demonstrating superior robustness against various attack vectors. Our findings suggest that improvements in training data design can be as important as the alignment algorithm itself in achieving effective LLM safety.
The rise of blockchain technologies has greatly accelerated the development and deployment of smart contracts. However, their inherent vulnerabilities and susceptibility to bugs have led to significant financial losses, underscoring the challenges in securing smart contracts. While traditional auditing methods are crucial, they often fall short in addressing the increasing complexity and volume of smart contracts. Recent advancements in Large Language Models (LLMs) offer promising solutions for enhancing software auditing by automatically identifying security vulnerabilities. Despite their potential, the practical application of these models is hindered by substantial computational demands. This paper investigates the feasibility of using smaller, fine-tuned models to achieve comparable or even superior results in smart contract auditing. We introduce the FTSmartAudit framework, which is designed to develop cost-effective, specialized models for smart contract auditing through the fine-tuning of LLMs. Our contributions include: (1) a single-task learning framework that streamlines data preparation, training, evaluation, and continuous learning; (2) a robust dataset generation method utilizing domain-special knowledge distillation to produce high-quality datasets from advanced models like GPT-4o; (3) an adaptive learning strategy to maintain model accuracy and robustness; (4) the proven effectiveness of fine-tuned models in detecting specific vulnerabilities and complex logical errors; and (5) a framework that can be extended to other domains requiring LLM solutions. Our experimental results demonstrate that smaller models can surpass state-of-the-art commercial models and tools in detecting vulnerabilities in smart contracts.
Effective query-item relevance modeling is pivotal for enhancing user experience and safeguarding user satisfaction in e-commerce search systems. Recently, benefiting from the vast inherent knowledge, Large Language Model (LLM) approach demonstrates strong performance and long-tail generalization ability compared with previous neural-based specialized relevance learning methods. Though promising, current LLM-based methods encounter the following inadequacies in practice: First, the massive parameters and computational demands make it difficult to be deployed online. Second, distilling LLM models to online models is a feasible direction, but the LLM relevance modeling is a black box, and its rich intrinsic knowledge is difficult to extract and apply online. To improve the interpretability of LLM and boost the performance of online relevance models via LLM, we propose an Explainable LLM-driven Multi-dimensional Distillation framework for e-commerce relevance learning, which comprises two core components: (1) An Explainable LLM for relevance modeling (ELLM-rele), which decomposes the relevance learning into intermediate steps and models relevance learning as a Chain-of-Thought (CoT) reasoning, thereby enhancing both interpretability and performance of LLM. (2) A Multi-dimensional Knowledge Distillation (MKD) architecture that transfers the knowledge of ELLM-rele to current deployable interaction-based and representation-based student models from both the relevance score distribution and CoT reasoning aspects. Through distilling the probabilistic and CoT reasoning knowledge, MKD improves both the semantic interaction and long-tail generalization abilities of student models. Extensive offline evaluations and online experiments on Taobao search ad scene demonstrate that our proposed framework significantly enhances e-commerce relevance learning performance and user experience.
The rapid proliferation of large language models has driven the need for efficient GPU training clusters. However, ensuring high-performance training in these clusters is challenging due to the complexity of software-hardware interactions and the frequent occurrence of training anomalies. Since existing diagnostic tools are narrowly tailored to specific issues, there are gaps in their ability to address anomalies spanning the entire training stack. In response, we introduce XPUTimer, a real-time diagnostic framework designed for distributed LLM training at scale. XPUTimer first integrates a lightweight tracing daemon to monitor key code segments with minimal overhead. Additionally, it features a diagnostic engine that employs novel intra-kernel tracing and holistic aggregated metrics to efficiently identify and resolve anomalies. Deployment of XPUTimer across 6,000 GPUs over eight months demonstrated significant improvements across the training stack, validating its effectiveness in real-world scenarios.
Large language models (LLMs) have achieved remarkable success in natural language processing tasks, but their high computational and memory demands pose challenges for deployment on resource-constrained devices. Binarization, as an efficient compression method that reduces model weights to just 1 bit, significantly lowers both computational and memory requirements. Despite this, the binarized LLM still contains redundancy, which can be further compressed. Semi-structured pruning provides a promising approach to achieve this, which offers a better trade-off between model performance and hardware efficiency. However, simply combining binarization with semi-structured pruning can lead to a significant performance drop. To address this issue, we propose a Progressive Binarization with Semi-Structured Pruning (PBS$^2$P) method for LLM compression. We first propose a Stepwise semi-structured Pruning with Binarization Optimization (SPBO). Our optimization strategy significantly reduces the total error caused by pruning and binarization, even below that of the no-pruning scenario. Furthermore, we design a Coarse-to-Fine Search (CFS) method to select pruning elements more effectively. Extensive experiments demonstrate that PBS$^2$P achieves superior accuracy across various LLM families and evaluation metrics, noticeably outperforming state-of-the-art (SOTA) binary PTQ methods. The code and models will be available at https://github.com/XIANGLONGYAN/PBS2P.
Although LLMs have achieved significant success, their reliance on large volumes of human-annotated data has limited their potential for further scaling. In this situation, utilizing self-generated synthetic data has become crucial for fine-tuning LLMs without extensive human annotation. However, current methods often fail to ensure consistent improvements across iterations, with performance stagnating after only minimal updates. To overcome these challenges, we introduce Dynamic Noise Preference Optimization (DNPO). DNPO employs a dynamic sample labeling mechanism to construct preference pairs for training and introduces controlled, trainable noise into the preference optimization process. Our approach effectively prevents stagnation and enables continuous improvement. In experiments with Zephyr-7B, DNPO consistently outperforms existing methods, showing an average performance boost of 2.6% across multiple benchmarks. Additionally, DNPO shows a significant improvement in model-generated data quality, with a 29.4% win-loss rate gap compared to the baseline in GPT-4 evaluations. This highlights its effectiveness in enhancing model performance through iterative refinement.
This paper focuses on the task of hallucination detection, which aims to determine the truthfulness of LLM-generated statements. To address this problem, a popular class of methods utilize the LLM's self-consistencies in its beliefs in a set of logically related augmented statements generated by the LLM, which does not require external knowledge databases and can work with both white-box and black-box LLMs. However, in many existing approaches, the augmented statements tend to be very monotone and unstructured, which makes it difficult to integrate meaningful information from the LLM beliefs in these statements. Also, many methods work with the binarized version of the LLM's belief, instead of the continuous version, which significantly loses information. To overcome these limitations, in this paper, we propose Belief Tree Propagation (BTProp), a probabilistic framework for LLM hallucination detection. BTProp introduces a belief tree of logically related statements by recursively decomposing a parent statement into child statements with three decomposition strategies, and builds a hidden Markov tree model to integrate the LLM's belief scores in these statements in a principled way. Experiment results show that our method improves baselines by 3%-9% (evaluated by AUROC and AUC-PR) on multiple hallucination detection benchmarks. Code is available at https://github.com/UCSB-NLP-Chang/BTProp.
Recent latent-space monitoring techniques have shown promise as defenses against LLM attacks. These defenses act as scanners that seek to detect harmful activations before they lead to undesirable actions. This prompts the question: Can models execute harmful behavior via inconspicuous latent states? Here, we study such obfuscated activations. We show that state-of-the-art latent-space defenses -- including sparse autoencoders, representation probing, and latent OOD detection -- are all vulnerable to obfuscated activations. For example, against probes trained to classify harmfulness, our attacks can often reduce recall from 100% to 0% while retaining a 90% jailbreaking rate. However, obfuscation has limits: we find that on a complex task (writing SQL code), obfuscation reduces model performance. Together, our results demonstrate that neural activations are highly malleable: we can reshape activation patterns in a variety of ways, often while preserving a network's behavior. This poses a fundamental challenge to latent-space defenses.
As large language models (LLMs) advance, their potential applications have grown significantly. However, it remains difficult to evaluate LLM behavior on user-defined tasks and craft effective pipelines to do so. Many users struggle with where to start, often referred to as the "blank page problem." ChainBuddy, an AI workflow generation assistant built into the ChainForge platform, aims to tackle this issue. From a single prompt or chat, ChainBuddy generates a starter evaluative LLM pipeline in ChainForge aligned to the user's requirements. ChainBuddy offers a straightforward and user-friendly way to plan and evaluate LLM behavior and make the process less daunting and more accessible across a wide range of possible tasks and use cases. We report a within-subjects user study comparing ChainBuddy to the baseline interface. We find that when using AI assistance, participants reported a less demanding workload, felt more confident, and produced higher quality pipelines evaluating LLM behavior. However, we also uncover a mismatch between subjective and objective ratings of performance: participants rated their successfulness similarly across conditions, while independent experts rated participant workflows significantly higher with AI assistance. Drawing connections to the Dunning-Kruger effect, we draw design implications for the future of workflow generation assistants to mitigate the risk of over-reliance.
The efficient deployment of large language models (LLMs) in online settings requires optimizing inference performance under stringent latency constraints, particularly the time-to-first-token (TTFT) and time-per-output-token (TPOT). This paper focuses on the query scheduling problem for LLM inference with prefix reuse, a technique that leverages shared prefixes across queries to reduce computational overhead. Our work reveals previously unknown limitations of the existing first-come-first-serve (FCFS) and longest-prefix-match (LPM) scheduling strategies with respect to satisfying latency constraints. We present a formal theoretical framework for LLM query scheduling under RadixAttention, a prefix reuse mechanism that stores and reuses intermediate representations in a radix tree structure. Our analysis establishes the NP-hardness of the scheduling problem with prefix reuse under TTFT constraints and proposes a novel scheduling algorithm, $k$-LPM, which generalizes existing methods by balancing prefix reuse and fairness in query processing. Theoretical guarantees demonstrate that $k$-LPM achieves improved TTFT performance under realistic traffic patterns captured by a data generative model. Empirical evaluations in a realistic serving setting validates our findings, showing significant reductions in P99 TTFT compared to baseline methods.
Retrieval augmented generation (RAG) is frequently used to mitigate hallucinations and provide up-to-date knowledge for large language models (LLMs). However, given that document retrieval is an imprecise task and sometimes results in erroneous or even harmful content being presented in context, this raises the question of how LLMs handle retrieved information: If the provided content is incorrect, does the model know to ignore it, or does it recapitulate the error? Conversely, when the model's initial response is incorrect, does it always know to use the retrieved information to correct itself, or does it insist on its wrong prior response? To answer this, we curate a dataset of over 1200 questions across six domains (e.g., drug dosages, Olympic records, locations) along with content relevant to answering each question. We further apply precise perturbations to the answers in the content that range from subtle to blatant errors. We benchmark six top-performing LLMs, including GPT-4o, on this dataset and find that LLMs are susceptible to adopting incorrect retrieved content, overriding their own correct prior knowledge over 60% of the time. However, the more unrealistic the retrieved content is (i.e. more deviated from truth), the less likely the model is to adopt it. Also, the less confident a model is in its initial response (via measuring token probabilities), the more likely it is to adopt the information in the retrieved content. We exploit this finding and demonstrate simple methods for improving model accuracy where there is conflicting retrieved content. Our results highlight a difficult task and benchmark for LLMs -- namely, their ability to correctly discern when it is wrong in light of correct retrieved content and to reject cases when the provided content is incorrect.
We introduce Agentic Reasoning, a framework that enhances large language model (LLM) reasoning by integrating external tool-using agents. Unlike conventional LLM-based reasoning approaches, which rely solely on internal inference, Agentic Reasoning dynamically engages web search, code execution, and structured reasoning-context memory to solve complex problems requiring deep research and multi-step logical deduction. Our framework introduces the Mind Map agent, which constructs a structured knowledge graph to track logical relationships, improving deductive reasoning. Additionally, the integration of web-search and coding agents enables real-time retrieval and computational analysis, enhancing reasoning accuracy and decision-making. Evaluations on PhD-level scientific reasoning (GPQA) and domain-specific deep research tasks demonstrate that our approach significantly outperforms existing models, including leading retrieval-augmented generation (RAG) systems and closed-source LLMs. Moreover, our results indicate that agentic reasoning improves expert-level knowledge synthesis, test-time scalability, and structured problem-solving. The code is at: https://github.com/theworldofagents/Agentic-Reasoning.
Graphs are ubiquitous structures found in numerous real-world applications, such as drug discovery, recommender systems, and social network analysis. To model graph-structured data, graph neural networks (GNNs) have become a popular tool. However, existing GNN architectures encounter challenges in cross-graph learning where multiple graphs have different feature spaces. To address this, recent approaches introduce text-attributed graphs (TAGs), where each node is associated with a textual description, which can be projected into a unified feature space using textual encoders. While promising, this method relies heavily on the availability of text-attributed graph data, which is difficult to obtain in practice. To bridge this gap, we propose a novel method named Topology-Aware Node description Synthesis (TANS), leveraging large language models (LLMs) to convert existing graphs into text-attributed graphs. The key idea is to integrate topological information into LLMs to explain how graph topology influences node semantics. We evaluate our TANS on text-rich, text-limited, and text-free graphs, demonstrating its applicability. Notably, on text-free graphs, our method significantly outperforms existing approaches that manually design node features, showcasing the potential of LLMs for preprocessing graph-structured data in the absence of textual information. The code and data are available at https://github.com/Zehong-Wang/TANS.
Post-training quantization (PTQ) is a promising approach to reducing the storage and computational requirements of large language models (LLMs) without additional training cost. Recent PTQ studies have primarily focused on quantizing only weights to sub-8-bits while maintaining activations at 8-bits or higher. Accurate sub-8-bit quantization for both weights and activations without relying on quantization-aware training remains a significant challenge. We propose a novel quantization method called block clustered quantization (BCQ) wherein each operand tensor is decomposed into blocks (a block is a group of contiguous scalars), blocks are clustered based on their statistics, and a dedicated optimal quantization codebook is designed for each cluster. As a specific embodiment of this approach, we propose a PTQ algorithm called Locally-Optimal BCQ (LO-BCQ) that iterates between the steps of block clustering and codebook design to greedily minimize the quantization mean squared error. When weight and activation scalars are encoded to W4A4 format (with 0.5-bits of overhead for storing scaling factors and codebook selectors), we advance the current state-of-the-art by demonstrating <1% loss in inference accuracy across several LLMs and downstream tasks.
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.
Large language models (LLMs) can learn vast amounts of knowledge from diverse domains during pre-training. However, long-tail knowledge from specialized domains is often scarce and underrepresented, rarely appearing in the models' memorization. Prior work has shown that in-context learning (ICL) with retriever augmentation can help LLMs better capture long-tail knowledge, reducing their reliance on pre-trained data. Despite these advances, we observe that LLM predictions for long-tail questions remain uncertain to variations in retrieved samples. To take advantage of the uncertainty in ICL for guiding LLM predictions toward correct answers on long-tail samples, we propose a reinforcement learning-based dynamic uncertainty ranking method for ICL that accounts for the varying impact of each retrieved sample on LLM predictions. Our approach prioritizes more informative and stable samples while demoting misleading ones, updating rankings based on the feedback from the LLM w.r.t. each retrieved sample. To enhance training efficiency and reduce query costs, we introduce a learnable dynamic ranking threshold, adjusted when the model encounters negative prediction shifts. Experimental results on various question-answering datasets from different domains show that our method outperforms the best baseline by $2.76\%$, with a notable $5.96\%$ boost in accuracy on long-tail questions that elude zero-shot inference.
Books, while often rich in cultural insights, can also mirror societal biases of their eras - biases that Large Language Models (LLMs) may learn and perpetuate during training. We introduce a novel method to trace and quantify these biases using fine-tuned LLMs. We develop BookPAGE, a corpus comprising 593 fictional books across seven decades (1950-2019), to track bias evolution. By fine-tuning LLMs on books from each decade and using targeted prompts, we examine shifts in biases related to gender, sexual orientation, race, and religion. Our findings indicate that LLMs trained on decade-specific books manifest biases reflective of their times, with both gradual trends and notable shifts. For example, model responses showed a progressive increase in the portrayal of women in leadership roles (from 8% to 22%) from the 1950s to 2010s, with a significant uptick in the 1990s (from 4% to 12%), possibly aligning with third-wave feminism. Same-sex relationship references increased markedly from the 1980s to 2000s (from 0% to 10%), mirroring growing LGBTQ+ visibility. Concerningly, negative portrayals of Islam rose sharply in the 2000s (26% to 38%), likely reflecting post-9/11 sentiments. Importantly, we demonstrate that these biases stem mainly from the books' content and not the models' architecture or initial training. Our study offers a new perspective on societal bias trends by bridging AI, literary studies, and social science research.
Large Language Models have proved surprisingly effective at solving a wide range of tasks from just a handful of examples. However, their lack of reliability and modularity limits their capacity to tackle large problems that require many steps of reasoning. In response, researchers have proposed advanced pipelines that leverage domain-specific knowledge to chain smaller prompts, provide intermediate feedback and improve performance through search. However, the current complexity of writing, tuning, maintaining and improving such pipelines has limited their sophistication. We propose oracular programming, a foundational paradigm for building LLM-enabled applications that lets domain experts express high-level problem-solving strategies as programs with unresolved choice points. These choice points are resolved at runtime by LLMs, which generalize from user-provided examples of correct and incorrect decisions. An oracular program is composed of three orthogonal components: a strategy that consists in a nondeterministic program with choice points that can be reified into a search tree, a policy that specifies how to navigate this tree with the help of LLM oracles, and a set of demonstrations that describe successful and unsuccessful search tree navigation scenarios across diverse problem instances. Each component is expressed in a dedicated programming language and can be independently improved or substituted. We address the key programming language design challenges of modularly composing oracular programs and enforcing consistency between their components as they evolve.
Large language models (LLMs) have become ubiquitous, thus it is important to understand their risks and limitations. Smaller LLMs can be deployed where compute resources are constrained, such as edge devices, but with different propensity to generate harmful output. Mitigation of LLM harm typically depends on annotating the harmfulness of LLM output, which is expensive to collect from humans. This work studies two questions: How do smaller LLMs rank regarding generation of harmful content? How well can larger LLMs annotate harmfulness? We prompt three small LLMs to elicit harmful content of various types, such as discriminatory language, offensive content, privacy invasion, or negative influence, and collect human rankings of their outputs. Then, we evaluate three state-of-the-art large LLMs on their ability to annotate the harmfulness of these responses. We find that the smaller models differ with respect to harmfulness. We also find that large LLMs show low to moderate agreement with humans. These findings underline the need for further work on harm mitigation in LLMs.
We present an outcome-driven fine-tuning framework that enhances the forecasting capabilities of large language models (LLMs) without relying on human-curated reasoning samples. Our method leverages model self-play to generate pairs of diverse reasoning trajectories and probabilistic forecasts for a set of diverse questions that resolve after the models' knowledge cutoff date. We then rank pairs of these reasoning traces by their distance to the actual outcomes before fine-tuning the model via Direct Preference Optimization (DPO). On a separate test set, our approach increases prediction accuracy of Phi-4 14B and DeepSeek-R1 14B by between 7--10\% over a base model and a DPO fine-tuned control model with randomized labels, bringing them on par with forecasting capabilities of much larger frontier models like GPT-4o.
Long-context large language models (LLMs) have recently shown strong performance in information retrieval and long-document QA. However, to tackle the most challenging intellectual problems, LLMs must reason effectively in long and complex contexts (e.g., frontier mathematical research). Studying how LLMs handle increasing reasoning complexity and context length is essential, yet existing benchmarks lack a solid basis for quantitative evaluation. Inspired by the abstraction of GSM-8K problems as computational graphs, and the ability to introduce noise by adding unnecessary nodes and edges, we develop a grade school math problem generator capable of producing arithmetic problems with infinite difficulty and context length under fine-grained control. Using our newly synthesized GSM-Infinite benchmark, we comprehensively evaluate existing LLMs. We find a consistent sigmoid decline in reasoning performance as complexity increases, along with a systematic inference scaling trend: exponentially increasing inference computation yields only linear performance gains. These findings underscore the fundamental limitations of current long-context LLMs and the key challenges in scaling reasoning capabilities. Our GSM-Infinite benchmark provides a scalable and controllable testbed for systematically studying and advancing LLM reasoning in long and complex contexts.
The rapid advancement of large language models (LLMs) has increased the need for guardrail models to ensure responsible use, particularly in detecting unsafe and illegal content. While substantial safety data exist in English, multilingual guardrail modeling remains underexplored due to the scarcity of open-source safety data in other languages. To address this gap, we propose a novel two-player Reinforcement Learning (RL) framework, where a generator and a guardrail model co-evolve adversarially to produce high-quality synthetic data for multilingual guardrail training. We theoretically formalize this interaction as a two-player game, proving convergence to a Nash equilibrium. Empirical evaluations show that our model \ours outperforms state-of-the-art models, achieving nearly 10% improvement over LlamaGuard3 (8B) on English benchmarks while being 4.5x faster at inference with a significantly smaller model (0.5B). We achieve substantial advancements in multilingual safety tasks, particularly in addressing the imbalance for lower-resource languages in a collected real dataset. Ablation studies emphasize the critical role of synthetic data generation in bridging the imbalance in open-source data between English and other languages. These findings establish a scalable and efficient approach to synthetic data generation, paving the way for improved multilingual guardrail models to enhance LLM safety. Code, model, and data will be open-sourced at https://github.com/yihedeng9/DuoGuard.
Large language models (LLMs) demonstrate impressive capabilities across many tasks yet risk reproducing copyrighted content verbatim, raising legal and ethical concerns. Although methods like differential privacy or neuron editing can reduce memorization, they typically require costly retraining or direct access to model weights and may degrade performance. To address these challenges, we propose TokenSwap, a lightweight, post-hoc approach that replaces the probabilities of grammar-related tokens with those from a small auxiliary model (e.g., DistilGPT-2). We run extensive experiments on commercial grade models such as Pythia-6.9b and LLaMA-3-8b and demonstrate that our method effectively reduces well-known cases of memorized generation by upto 10x with little to no impact on downstream tasks. Our approach offers a uniquely accessible and effective solution to users of real-world systems.
This work studies the problem of large language model (LLM) unlearning, aiming to remove unwanted data influences (e.g., copyrighted or harmful content) while preserving model utility. Despite the increasing demand for unlearning, a technically-grounded optimization framework is lacking. Gradient ascent (GA)-type methods, though widely used, are suboptimal as they reverse the learning process without controlling optimization divergence (i.e., deviation from the pre-trained state), leading to risks of over-forgetting and potential model collapse. Negative preference optimization (NPO) has been proposed to address this issue and is considered one of the state-of-the-art LLM unlearning approaches. In this work, we revisit NPO and identify another critical issue: reference model bias. This bias arises from using the reference model (i.e., the model prior to unlearning) to evaluate the unlearning success, which can compromise NPO's effectiveness. Specifically, it leads to (a) uneven allocation of optimization power across forget data with varying difficulty levels and (b) ineffective gradient weight smoothing during the early stages of unlearning optimization. To overcome these challenges, we propose a simple yet effective unlearning optimization framework, called SimNPO, showing that `simplicity' in removing the reliance on a reference model (through the lens of simple preference optimization) benefits unlearning. We provide deeper insights into SimNPO's advantages through an analysis based on mixtures of Markov chains. Extensive experiments further validate SimNPO's efficacy on benchmarks like TOFU and MUSE, as well as its robustness against relearning attacks. Codes are available at https://github.com/OPTML-Group/Unlearn-Simple.
The Parry-Lord oral-formulaic theory was a breakthrough in understanding how oral narrative poetry is learned, composed, and transmitted by illiterate bards. In this paper, we provide an annotated reading of the mechanism underlying this theory from the lens of large language models (LLMs) and generative artificial intelligence (AI). We point out the the similarities and differences between oral composition and LLM generation, and comment on the implications to society and AI policy.
Understanding the internal mechanisms of large language models (LLMs) is integral to enhancing their reliability, interpretability, and inference processes. We present Constituent-Aware Pooling (CAP), a methodology designed to analyse how LLMs process compositional linguistic structures. Grounded in principles of compositionality, mechanistic interpretability, and information theory, CAP systematically intervenes in model activations through constituent-based pooling at various model levels. Our experiments on inverse definition modelling, hypernym and synonym prediction reveal critical insights into transformers' limitations in handling compositional abstractions. No specific layer integrates tokens into unified semantic representations based on their constituent parts. We observe fragmented information processing, which intensifies with model size, suggesting that larger models struggle more with these interventions and exhibit greater information dispersion. This fragmentation likely stems from transformers' training objectives and architectural design, preventing systematic and cohesive representations. Our findings highlight fundamental limitations in current transformer architectures regarding compositional semantics processing and model interpretability, underscoring the critical need for novel approaches in LLM design to address these challenges.
Large Language Models (LLMs) have revolutionized natural language processing (NLP), excelling in tasks like text generation and summarization. However, their increasing adoption in mission-critical applications raises concerns about hardware-based threats, particularly bit-flip attacks (BFAs). BFAs, enabled by fault injection methods such as Rowhammer, target model parameters in memory, compromising both integrity and performance. Identifying critical parameters for BFAs in the vast parameter space of LLMs poses significant challenges. While prior research suggests transformer-based architectures are inherently more robust to BFAs compared to traditional deep neural networks, we challenge this assumption. For the first time, we demonstrate that as few as three bit-flips can cause catastrophic performance degradation in an LLM with billions of parameters. Current BFA techniques are inadequate for exploiting this vulnerability due to the difficulty of efficiently identifying critical parameters within the immense parameter space. To address this, we propose AttentionBreaker, a novel framework tailored for LLMs that enables efficient traversal of the parameter space to identify critical parameters. Additionally, we introduce GenBFA, an evolutionary optimization strategy designed to refine the search further, isolating the most critical bits for an efficient and effective attack. Empirical results reveal the profound vulnerability of LLMs to AttentionBreaker. For example, merely three bit-flips (4.129 x 10^-9% of total parameters) in the LLaMA3-8B-Instruct 8-bit quantized (W8) model result in a complete performance collapse: accuracy on MMLU tasks drops from 67.3% to 0%, and Wikitext perplexity skyrockets from 12.6 to 4.72 x 10^5. These findings underscore the effectiveness of AttentionBreaker in uncovering and exploiting critical vulnerabilities within LLM architectures.
One approach to reducing the massive costs of large language models (LLMs) is the use of quantized or sparse representations for training or deployment. While post-training compression methods are very popular, the question of obtaining even more accurate compressed models by directly training over such representations, i.e., Quantization-Aware Training (QAT), is still open: for example, a recent study (arXiv:2411.04330v2) put the "optimal" bit-width at which models can be trained using QAT, while staying accuracy-competitive with standard FP16/BF16 precision, at 8-bits weights and activations. We advance this state-of-the-art via a new method called QuEST, which is Pareto-competitive with FP16, i.e., it provides better accuracy at lower model size, while training models with weights and activations in 4-bits or less. Moreover, QuEST allows stable training with 1-bit weights and activations. QuEST achieves this by improving two key aspects of QAT methods: (1) accurate and fast quantization of the (continuous) distributions of weights and activations via Hadamard normalization and MSE-optimal fitting; (2) a new trust gradient estimator based on the idea of explicitly minimizing the error between the noisy gradient computed over quantized states and the "true" (but unknown) full-precision gradient. Experiments on Llama-type architectures show that QuEST induces stable scaling laws across the entire range of hardware-supported precisions, and can be extended to sparse representations. We provide GPU kernel support showing that models produced by QuEST can be executed efficiently. Our code is available at https://github.com/IST-DASLab/QuEST.
Creating interactive scenes often involves complex programming tasks. Although large language models (LLMs) like ChatGPT can generate code from natural language, their output is often error-prone, particularly when scripting interactions among multiple elements. The linear conversational structure limits the editing of individual elements, and lacking graphical and precise control complicates visual integration. To address these issues, we integrate an element-level modularization technique that processes textual descriptions for individual elements through separate LLM modules, with a central module managing interactions among elements. This modular approach allows for refining each element independently. We design a graphical user interface, MoGraphGPT , which combines modular LLMs with enhanced graphical control to generate codes for 2D interactive scenes. It enables direct integration of graphical information and offers quick, precise control through automatically generated sliders. Our comparative evaluation against an AI coding tool, Cursor Composer, as the baseline system and a usability study show MoGraphGPT significantly improves easiness, controllability, and refinement in creating complex 2D interactive scenes with multiple visual elements in a coding-free manner.
Uncertainty quantification (UQ) methods for Large Language Models (LLMs) encompasses a variety of approaches, with two major types being particularly prominent: information-based, which focus on model confidence expressed as token probabilities, and consistency-based, which assess the semantic relationship between multiple outputs generated using repeated sampling. Several recent methods have combined these two approaches and shown impressive performance in various applications. However, they sometimes fail to outperform much simpler baseline methods. Our investigation reveals distinctive characteristics of LLMs as probabilistic models, which help to explain why these UQ methods underperform in certain tasks. Based on these findings, we propose a new way of synthesizing model confidence and output consistency that leads to a family of efficient and robust UQ methods. We evaluate our approach across a variety of tasks such as question answering, abstractive summarization, and machine translation, demonstrating sizable improvements over state-of-the-art UQ approaches.
Fingerprinting refers to the process of identifying underlying Machine Learning (ML) models of AI Systemts, such as Large Language Models (LLMs), by analyzing their unique characteristics or patterns, much like a human fingerprint. The fingerprinting of Large Language Models (LLMs) has become essential for ensuring the security and transparency of AI-integrated applications. While existing methods primarily rely on access to direct interactions with the application to infer model identity, they often fail in real-world scenarios involving multi-agent systems, frequent model updates, and restricted access to model internals. In this paper, we introduce a novel fingerprinting framework designed to address these challenges by integrating static and dynamic fingerprinting techniques. Our approach identifies architectural features and behavioral traits, enabling accurate and robust fingerprinting of LLMs in dynamic environments. We also highlight new threat scenarios where traditional fingerprinting methods are ineffective, bridging the gap between theoretical techniques and practical application. To validate our framework, we present an extensive evaluation setup that simulates real-world conditions and demonstrate the effectiveness of our methods in identifying and monitoring LLMs in Gen-AI applications. Our results highlight the framework's adaptability to diverse and evolving deployment contexts.
Game-based interventions are widely used to combat misinformation online by employing the "inoculation approach". However, most current interventions are designed as single-player games, presenting players with limited predefined choices. Such restrictions reduce replayability and may lead to an overly simplistic understanding of the processes of misinformation phenomenon and the debunking. This study seeks to address these issues, and empower people to better understand the opinion influencing and misinformation debunking processes. We did this by creating a Player versus Player (PvP) game where participants attempt to either generate or debunk misinformation to convince LLM-represented public opinion. Using a within-subjects mixed-methods study design (N=47), we found that this game significantly raised participants' media literacy and improved their ability to identify misinformation. Our qualitative exploration revealed how participants' use of debunking and content creation strategies deepened their understanding of the nature of disinformation. We demonstrate how LLMs can be integrated into PvP games to foster greater understanding of contrasting viewpoints and highlight social challenges.
Tool-calling has changed Large Language Model (LLM) applications by integrating external tools, significantly enhancing their functionality across diverse tasks. However, this integration also introduces new security vulnerabilities, particularly in the tool scheduling mechanisms of LLM, which have not been extensively studied. To fill this gap, we present ToolCommander, a novel framework designed to exploit vulnerabilities in LLM tool-calling systems through adversarial tool injection. Our framework employs a well-designed two-stage attack strategy. Firstly, it injects malicious tools to collect user queries, then dynamically updates the injected tools based on the stolen information to enhance subsequent attacks. These stages enable ToolCommander to execute privacy theft, launch denial-of-service attacks, and even manipulate business competition by triggering unscheduled tool-calling. Notably, the ASR reaches 91.67% for privacy theft and hits 100% for denial-of-service and unscheduled tool calling in certain cases. Our work demonstrates that these vulnerabilities can lead to severe consequences beyond simple misuse of tool-calling systems, underscoring the urgent need for robust defensive strategies to secure LLM Tool-calling systems.
Identifying user intents in information-seeking dialogs is crucial for a system to meet user's information needs. Intent prediction (IP) is challenging and demands sufficient dialogs with human-labeled intents for training. However, manually annotating intents is resource-intensive. While large language models (LLMs) have been shown to be effective in generating synthetic data, there is no study on using LLMs to generate intent-aware information-seeking dialogs. In this paper, we focus on leveraging LLMs for zero-shot generation of large-scale, open-domain, and intent-aware information-seeking dialogs. We propose SOLID, which has novel self-seeding and multi-intent self-instructing schemes. The former improves the generation quality by using the LLM's own knowledge scope to initiate dialog generation; the latter prompts the LLM to generate utterances sequentially, and mitigates the need for manual prompt design by asking the LLM to autonomously adapt its prompt instruction when generating complex multi-intent utterances. Furthermore, we propose SOLID-RL, which is further trained to generate a dialog in one step on the data generated by SOLID. We propose a length-based quality estimation mechanism to assign varying weights to SOLID-generated dialogs based on their quality during the training process of SOLID-RL. We use SOLID and SOLID-RL to generate more than 300k intent-aware dialogs, surpassing the size of existing datasets. Experiments show that IP methods trained on dialogs generated by SOLID and SOLID-RL achieve better IP quality than ones trained on human-generated dialogs.
The substantial computational and memory demands of Large Language Models (LLMs) hinder their deployment. Block Floating Point (BFP) has proven effective in accelerating linear operations, a cornerstone of LLM workloads. However, as sequence lengths grow, nonlinear operations, such as Attention, increasingly become performance bottlenecks due to their quadratic computational complexity. These nonlinear operations are predominantly executed using inefficient floating-point formats, which renders the system challenging to optimize software efficiency and hardware overhead. In this paper, we delve into the limitations and potential of applying BFP to nonlinear operations. Given our findings, we introduce a hardware-software co-design framework (DB-Attn), including: (i) DBFP, an advanced BFP version, overcomes nonlinear operation challenges with a pivot-focus strategy for diverse data and an adaptive grouping strategy for flexible exponent sharing. (ii) DH-LUT, a novel lookup table algorithm dedicated to accelerating nonlinear operations with DBFP format. (iii) An RTL-level DBFP-based engine is implemented to support DB-Attn, applicable to FPGA and ASIC. Results show that DB-Attn provides significant performance improvements with negligible accuracy loss, achieving 74% GPU speedup on Softmax of LLaMA and 10x low overhead performance improvement over SOTA designs.
The grammatical knowledge of language models (LMs) is often measured using a benchmark of linguistic minimal pairs, where the LMs are presented with a pair of acceptable and unacceptable sentences and required to judge which is more acceptable. Conventional approaches directly compare sentence probabilities assigned by LMs, but recent large language models (LLMs) are trained to perform tasks via prompting, and thus, the raw probabilities they assign may not fully reflect their grammatical knowledge. In this study, we attempt to derive more accurate acceptability judgments from LLMs using prompts and templates. Through extensive experiments in English and Chinese, we compare nine judgment methods and find two of them, a probability readout method -- in-template LP and a prompt-based method -- Yes/No probability computing, achieve higher accuracy than the conventional ones. Our analysis reveals that these methods excel in different linguistic phenomena, suggesting they access different aspects of LLMs' knowledge. We also find that ensembling the two methods outperforms single methods. Consequently, we recommend these techniques, either individually or ensembled, as more effective alternatives to conventional approaches for assessing grammatical knowledge in LLMs.
Scaling Large Language Model (LLM) training relies on multi-dimensional parallelism, where High-Bandwidth Domains (HBDs) are critical for communication-intensive parallelism like Tensor Parallelism (TP) and Expert Parallelism (EP). However, existing HBD architectures face fundamental limitations in scalability, cost, and fault resiliency: switch-centric HBDs (e.g., NVL-72) incur prohibitive scaling costs, while GPU-centric HBDs (e.g., TPUv3/Dojo) suffer from severe fault propagation. Switch-GPU hybrid HBDs such as TPUv4 takes a middle-ground approach by leveraging Optical Circuit Switches, but the fault explosion radius remains large at the cube level (e.g., 64 TPUs). We propose InfinitePOD, a novel transceiver-centric HBD architecture that unifies connectivity and dynamic switching at the transceiver level using Optical Circuit Switching (OCS). By embedding OCS within each transceiver, InfinitePOD achieves reconfigurable point-to-multipoint connectivity, allowing the topology to adapt into variable-size rings. This design provides: i) datacenter-wide scalability without cost explosion; ii) fault resilience by isolating failures to a single node, and iii) full bandwidth utilization for fault-free GPUs. Key innovations include a Silicon Photonic (SiPh) based low-cost OCS transceiver (OCSTrx), a reconfigurable k-hop ring topology co-designed with intra-/inter-node communication, and an HBD-DCN orchestration algorithm maximizing GPU utilization while minimizing cross-ToR datacenter network traffic. The evaluation demonstrates that InfinitePOD achieves 31% of the cost of NVL-72, near-zero GPU waste ratio (over one order of magnitude lower than NVL-72 and TPUv4), near-zero cross-ToR traffic when node fault ratios under 7%, and improves Model FLOPs Utilization by 3.37x compared to NVIDIA DGX (8 GPUs per Node).
The Bourne-Again Shell (Bash) command-line interface for Linux systems has complex syntax and requires extensive specialized knowledge. Using the natural language to Bash command (NL2SH) translation capabilities of large language models (LLMs) for command composition circumvents these issues. However, the NL2SH performance of LLMs is difficult to assess due to inaccurate test data and unreliable heuristics for determining the functional equivalence of Bash commands. We present a manually verified test dataset of 600 instruction-command pairs and a training dataset of 40,939 pairs, increasing the size of previous datasets by 441% and 135%, respectively. Further, we present a novel functional equivalence heuristic that combines command execution with LLM evaluation of command outputs. Our heuristic can determine the functional equivalence of two Bash commands with 95% confidence, a 16% increase over previous heuristics. Evaluation of popular LLMs using our test dataset and heuristic demonstrates that parsing, in-context learning, in-weight learning, and constrained decoding can improve NL2SH accuracy by up to 32%. Our findings emphasize the importance of dataset quality, execution-based evaluation and translation method for advancing NL2SH translation. Our code is available at https://github.com/westenfelder/NL2SH
Proteins play a pivotal role in living organisms, yet understanding their functions presents significant challenges, including the limited flexibility of classification-based methods, the inability to effectively leverage spatial structural information, and the lack of systematic evaluation metrics for protein Q&A systems. To address these limitations, we propose Prot2Chat, a novel framework that integrates multimodal protein representations with natural language through a unified module, enabling large language model (LLM)-driven answer generation. Our model incorporates a modified ProteinMPNN encoder, which encodes protein sequence and structural information in a unified manner, a protein-text adapter with cross-attention mechanisms, and a LLaMA3 decoder. To optimize training efficiency, we freeze the encoder and employ LoRA techniques for the decoder. We conducted experiments on two datasets, both automated metrics and expert evaluations demonstrate the superior performance of our model. Furthermore, zero-shot prediction results highlight its strong generalization capabilities. This framework offers a promising solution for bridging protein domain knowledge with natural language understanding, paving the way for transformative advancements in protein-related research.
Large Language Models (LLMs) are increasingly used to assess news credibility, yet little is known about how they make these judgments. While prior research has examined political bias in LLM outputs or their potential for automated fact-checking, their internal evaluation processes remain largely unexamined. Understanding how LLMs assess credibility provides insights into AI behavior and how credibility is structured and applied in large-scale language models. This study benchmarks the reliability and political classifications of state-of-the-art LLMs - Gemini 1.5 Flash (Google), GPT-4o mini (OpenAI), and LLaMA 3.1 (Meta) - against structured, expert-driven rating systems such as NewsGuard and Media Bias Fact Check. Beyond assessing classification performance, we analyze the linguistic markers that shape LLM decisions, identifying which words and concepts drive their evaluations. We uncover patterns in how LLMs associate credibility with specific linguistic features by examining keyword frequency, contextual determinants, and rank distributions. Beyond static classification, we introduce a framework in which LLMs refine their credibility assessments by retrieving external information, querying other models, and adapting their responses. This allows us to investigate whether their assessments reflect structured reasoning or rely primarily on prior learned associations.
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.
Understanding and Mitigating the Bias Inheritance in LLM-based Data Augmentation on Downstream Tasks
Generating synthetic datasets via large language models (LLMs) themselves has emerged as a promising approach to improve LLM performance. However, LLMs inherently reflect biases present in their training data, leading to a critical challenge: when these models generate synthetic data for training, they may propagate and amplify their inherent biases that can significantly impact model fairness and robustness on downstream tasks--a phenomenon we term bias inheritance. This work presents the first systematic investigation in understanding, analyzing, and mitigating bias inheritance. We study this problem by fine-tuning LLMs with a combined dataset consisting of original and LLM-augmented data, where bias ratio represents the proportion of augmented data. Through systematic experiments across 10 classification and generation tasks, we analyze how 6 different types of biases manifest at varying bias ratios. Our results reveal that bias inheritance has nuanced effects on downstream tasks, influencing both classification tasks and generation tasks differently. Then, our analysis identifies three key misalignment factors: misalignment of values, group data, and data distributions. Based on these insights, we propose three mitigation strategies: token-based, mask-based, and loss-based approaches. Experiments demonstrate that these strategies also work differently on various tasks and bias, indicating the substantial challenges to fully mitigate bias inheritance. We hope this work can provide valuable insights to the research of LLM data augmentation.
Large language models (LLMs) achieve impressive performance by scaling model parameters, but this comes with significant inference overhead. Feed-forward networks (FFNs), which dominate LLM parameters, exhibit high activation sparsity in hidden neurons. To exploit this, researchers have proposed using a mixture-of-experts (MoE) architecture, where only a subset of parameters is activated. However, existing approaches often require extensive training data and resources, limiting their practicality. We propose CMoE (Carved MoE), a novel framework to efficiently carve MoE models from dense models. CMoE achieves remarkable performance through efficient expert grouping and lightweight adaptation. First, neurons are grouped into shared and routed experts based on activation rates. Next, we construct a routing mechanism without training from scratch, incorporating a differentiable routing process and load balancing. Using modest data, CMoE produces a well-designed, usable MoE from a 7B dense model within five minutes. With lightweight fine-tuning, it achieves high-performance recovery in under an hour. We make our code publicly available at https://github.com/JarvisPei/CMoE.
Groundbreaking advancements in text-to-image generation have recently been achieved with the emergence of diffusion models. These models exhibit a remarkable ability to generate highly artistic and intricately detailed images based on textual prompts. However, obtaining desired generation outcomes often necessitates repetitive trials of manipulating text prompts just like casting spells on a magic mirror, and the reason behind that is the limited capability of semantic understanding inherent in current image generation models. Specifically, existing diffusion models encode the text prompt input with a pre-trained encoder structure, which is usually trained on a limited number of image-caption pairs. The state-of-the-art large language models (LLMs) based on the decoder-only structure have shown a powerful semantic understanding capability as their architectures are more suitable for training on very large-scale unlabeled data. In this work, we propose to enhance text-to-image diffusion models by borrowing the strength of semantic understanding from large language models, and devise a simple yet effective adapter to allow the diffusion models to be compatible with the decoder-only structure. Meanwhile, we also provide a supporting theoretical analysis with various architectures (e.g., encoder-only, encoder-decoder, and decoder-only), and conduct extensive empirical evaluations to verify its effectiveness. The experimental results show that the enhanced models with our adapter module are superior to the stat-of-the-art models in terms of text-to-image generation quality and reliability.
Model merging aggregates Large Language Models (LLMs) finetuned on different tasks into a stronger one. However, parameter conflicts between models leads to performance degradation in averaging. While model routing addresses this issue by selecting individual models during inference, it imposes excessive storage and compute costs, and fails to leverage the common knowledge from different models. In this work, we observe that different layers exhibit varying levels of parameter conflicts. Building on this insight, we average layers with minimal parameter conflicts and use a novel task-level expert routing for layers with significant conflicts. To further reduce storage costs, inspired by task arithmetic sparsity, we decouple multiple fine-tuned experts into a dense expert and several sparse experts. Considering the out-of-distribution samples, we select and merge appropriate experts based on the task uncertainty of the input data. We conduct extensive experiments on both LLaMA and Qwen with varying parameter scales, and evaluate on real-world reasoning tasks. Results demonstrate that our method consistently achieves significant performance improvements while requiring less system cost compared to existing methods.
Alzheimer's Disease (AD) is an irreversible neurodegenerative disease affecting 50 million people worldwide. Low-cost, accurate identification of key markers of AD is crucial for timely diagnosis and intervention. Language impairment is one of the earliest signs of cognitive decline, which can be used to discriminate AD patients from normal control individuals. Patient-interviewer dialogues may be used to detect such impairments, but they are often mixed with ambiguous, noisy, and irrelevant information, making the AD detection task difficult. Moreover, the limited availability of AD speech samples and variability in their speech styles pose significant challenges in developing robust speech-based AD detection models. To address these challenges, we propose DECT, a novel speech-based domain-specific approach leveraging large language models (LLMs) for fine-grained linguistic analysis and label-switched label-preserved data generation. Our study presents four novelties: We harness the summarizing capabilities of LLMs to identify and distill key Cognitive-Linguistic information from noisy speech transcripts, effectively filtering irrelevant information. We leverage the inherent linguistic knowledge of LLMs to extract linguistic markers from unstructured and heterogeneous audio transcripts. We exploit the compositional ability of LLMs to generate AD speech transcripts consisting of diverse linguistic patterns to overcome the speech data scarcity challenge and enhance the robustness of AD detection models. We use the augmented AD textual speech transcript dataset and a more fine-grained representation of AD textual speech transcript data to fine-tune the AD detection model. The results have shown that DECT demonstrates superior model performance with an 11% improvement in AD detection accuracy on the datasets from DementiaBank compared to the baselines.
Effective asynchronous planning, or the ability to efficiently reason and plan over states and actions that must happen in parallel or sequentially, is essential for agents that must account for time delays, reason over diverse long-horizon tasks, and collaborate with other agents. While large language model (LLM) agents show promise in high-level task planning, current benchmarks focus primarily on short-horizon tasks and do not evaluate such asynchronous planning capabilities. We introduce Robotouille, a challenging benchmark environment designed to test LLM agents' ability to handle long-horizon asynchronous scenarios. Our synchronous and asynchronous datasets capture increasingly complex planning challenges that go beyond existing benchmarks, requiring agents to manage overlapping tasks and interruptions. Our results show that ReAct (gpt4-o) achieves 47% on synchronous tasks but only 11% on asynchronous tasks, highlighting significant room for improvement. We further analyze failure modes, demonstrating the need for LLM agents to better incorporate long-horizon feedback and self-audit their reasoning during task execution. Code is available at https://github.com/portal-cornell/robotouille.
Large Language Models (LLMs) have achieved significantly advanced capabilities in understanding and generating human language text, which have gained increasing popularity over recent years. Apart from their state-of-the-art natural language processing (NLP) performance, considering their widespread usage in many industries, including medicine, finance, education, etc., security concerns over their usage grow simultaneously. In recent years, the evolution of backdoor attacks has progressed with the advancement of defense mechanisms against them and more well-developed features in the LLMs. In this paper, we adapt the general taxonomy for classifying machine learning attacks on one of the subdivisions - training-time white-box backdoor attacks. Besides systematically classifying attack methods, we also consider the corresponding defense methods against backdoor attacks. By providing an extensive summary of existing works, we hope this survey can serve as a guideline for inspiring future research that further extends the attack scenarios and creates a stronger defense against them for more robust LLMs.
Traditional autonomous driving systems often struggle with reasoning in complex, unexpected scenarios due to limited comprehension of spatial relationships. In response, this study introduces a Large Language Model (LLM)-based Autonomous Driving (AD) assistance system that integrates a vision adapter and an LLM reasoning module to enhance visual understanding and decision-making. The vision adapter, combining YOLOv4 and Vision Transformer (ViT), extracts comprehensive visual features, while GPT-4 enables human-like spatial reasoning and response generation. Experimental evaluations with 45 experienced drivers revealed that the system closely mirrors human performance in describing situations and moderately aligns with human decisions in generating appropriate responses.
Despite extensive safety alignment efforts, large language models (LLMs) remain vulnerable to jailbreak attacks that elicit harmful behavior. While existing studies predominantly focus on attack methods that require technical expertise, two critical questions remain underexplored: (1) Are jailbroken responses truly useful in enabling average users to carry out harmful actions? (2) Do safety vulnerabilities exist in more common, simple human-LLM interactions? In this paper, we demonstrate that LLM responses most effectively facilitate harmful actions when they are both actionable and informative--two attributes easily elicited in multi-step, multilingual interactions. Using this insight, we propose HarmScore, a jailbreak metric that measures how effectively an LLM response enables harmful actions, and Speak Easy, a simple multi-step, multilingual attack framework. Notably, by incorporating Speak Easy into direct request and jailbreak baselines, we see an average absolute increase of 0.319 in Attack Success Rate and 0.426 in HarmScore in both open-source and proprietary LLMs across four safety benchmarks. Our work reveals a critical yet often overlooked vulnerability: Malicious users can easily exploit common interaction patterns for harmful intentions.
Large Language Models (LLMs) have shown significant capability across various tasks, with their real-world effectiveness often driven by prompt design. While recent research has focused on optimizing prompt content, the role of prompt formatting, a critical but often overlooked dimension, has received limited systematic investigation. In this paper, we introduce Content-Format Integrated Prompt Optimization (CFPO), an innovative methodology that jointly optimizes both prompt content and formatting through an iterative refinement process. CFPO leverages natural language mutations to explore content variations and employs a dynamic format exploration strategy that systematically evaluates diverse format options. Our extensive evaluations across multiple tasks and open-source LLMs demonstrate that CFPO demonstrates measurable performance improvements compared to content-only optimization methods. This highlights the importance of integrated content-format optimization and offers a practical, model-agnostic approach to enhancing LLM performance. Code will be available at https://github.com/HenryLau7/CFPO.
We explore the feasibility and effectiveness of using LLM-driven autonomous systems for Assumed Breach penetration testing in enterprise networks. We introduce a novel prototype that, driven by Large Language Models (LLMs), can compromise accounts within a real-life Active Directory testbed. Our research provides a comprehensive evaluation of the prototype's capabilities, and highlights both strengths and limitations while executing attack. The evaluation uses a realistic simulation environment (Game of Active Directory, GOAD) to capture intricate interactions, stochastic outcomes, and timing dependencies that characterize live network scenarios. The study concludes that autonomous LLMs are able to conduct Assumed Breach simulations, potentially democratizing access to penetration testing for organizations facing budgetary constraints. The prototype's source code, traces, and analyzed logs are released as open-source to enhance collective cybersecurity and facilitate future research in LLM-driven cybersecurity automation.
Jailbreak attacks against large language models (LLMs) aim to induce harmful behaviors in LLMs through carefully crafted adversarial prompts. To mitigate attacks, one way is to perform adversarial training (AT)-based alignment, i.e., training LLMs on some of the most adversarial prompts to help them learn how to behave safely under attacks. During AT, the length of adversarial prompts plays a critical role in the robustness of aligned LLMs. This paper focuses on adversarial suffix jailbreak attacks and unveils that to defend against a jailbreak attack with an adversarial suffix of length $\Theta(M)$, it is enough to align LLMs on prompts with adversarial suffixes of length $\Theta(\sqrt{M})$. Theoretically, we analyze the adversarial in-context learning of linear transformers on linear regression tasks and prove a robust generalization bound for trained transformers. The bound depends on the term $\Theta(\sqrt{M_{\text{test}}}/M_{\text{train}})$, where $M_{\text{train}}$ and $M_{\text{test}}$ are the number of adversarially perturbed in-context samples during training and testing. Empirically, we conduct AT on popular open-source LLMs and evaluate their robustness against jailbreak attacks of different adversarial suffix lengths. Results confirm a positive correlation between the attack success rate and the ratio of the square root of the adversarial suffix during jailbreaking to the length during AT. Our findings show that it is practical to defend "long-length" jailbreak attacks via efficient "short-length" AT. The code is available at https://github.com/fshp971/adv-icl.
Recent studies have highlighted the presence of cultural biases in Large Language Models (LLMs), yet often lack a robust methodology to dissect these phenomena comprehensively. Our work aims to bridge this gap by delving into the Food domain, a universally relevant yet culturally diverse aspect of human life. We introduce FmLAMA, a multilingual dataset centered on food-related cultural facts and variations in food practices. We analyze LLMs across various architectures and configurations, evaluating their performance in both monolingual and multilingual settings. By leveraging templates in six different languages, we investigate how LLMs interact with language-specific and cultural knowledge. Our findings reveal that (1) LLMs demonstrate a pronounced bias towards food knowledge prevalent in the United States; (2) Incorporating relevant cultural context significantly improves LLMs' ability to access cultural knowledge; (3) The efficacy of LLMs in capturing cultural nuances is highly dependent on the interplay between the probing language, the specific model architecture, and the cultural context in question. This research underscores the complexity of integrating cultural understanding into LLMs and emphasizes the importance of culturally diverse datasets to mitigate biases and enhance model performance across different cultural domains.
As large language models (LLMs) become integral to diverse applications, ensuring their reliability under varying input conditions is crucial. One key issue affecting this reliability is order sensitivity, wherein slight variations in input arrangement can lead to inconsistent or biased outputs. Although recent advances have reduced this sensitivity, the problem remains unresolved. This paper investigates the extent of order sensitivity in closed-source LLMs by conducting experiments across multiple tasks, including paraphrasing, relevance judgment, and multiple-choice questions. Our results show that input order significantly affects performance across tasks, with shuffled inputs leading to measurable declines in output accuracy. Few-shot prompting demonstrates mixed effectiveness and offers partial mitigation, however, fails to fully resolve the problem. These findings highlight persistent risks, particularly in high-stakes applications, and point to the need for more robust LLMs or improved input-handling techniques in future development.
This work presents a custom approach to developing a domain specific knowledge assistant for sustainability reporting using the International Financial Reporting Standards (IFRS). In this domain, there is no publicly available question-answer dataset, which has impeded the development of a high-quality chatbot to support companies with IFRS reporting. The two key contributions of this project therefore are: (1) A high-quality synthetic question-answer (QA) dataset based on IFRS sustainability standards, created using a novel generation and evaluation pipeline leveraging Large Language Models (LLMs). This comprises 1,063 diverse QA pairs that address a wide spectrum of potential user queries in sustainability reporting. Various LLM-based techniques are employed to create the dataset, including chain-of-thought reasoning and few-shot prompting. A custom evaluation framework is developed to assess question and answer quality across multiple dimensions, including faithfulness, relevance, and domain specificity. The dataset averages a score range of 8.16 out of 10 on these metrics. (2) Two architectures for question-answering in the sustainability reporting domain - a RAG pipeline and a fully LLM-based pipeline. The architectures are developed by experimenting, fine-tuning, and training on the QA dataset. The final pipelines feature an LLM fine-tuned on domain specific data and an industry classification component to improve the handling of complex queries. The RAG architecture achieves an accuracy of 85.32% on single-industry and 72.15% on cross-industry multiple-choice questions, outperforming the baseline approach by 4.67 and 19.21 percentage points, respectively. The LLM-based pipeline achieves an accuracy of 93.45% on single-industry and 80.30% on cross-industry multiple-choice questions, an improvement of 12.80 and 27.36 percentage points over the baseline, respectively.