Mixture of experts (MoE) models achieve state-of-the-art results in language modeling but suffer from inefficient hardware utilization due to imbalanced token routing and communication overhead. While prior work has focused on optimizing MoE training and decoder architectures, inference for encoder-based MoE models in a multi-GPU with expert parallelism setting remains underexplored. We introduce MoEShard, an inference system that achieves perfect load balancing through tensor sharding of MoE experts. Unlike existing approaches that rely on heuristic capacity factors or drop tokens, MoEShard evenly distributes computation across GPUs and ensures full token retention, maximizing utilization regardless of routing skewness. We achieve this through a strategic row- and column-wise decomposition of expert matrices. This reduces idle time and avoids bottlenecks caused by imbalanced expert assignments. Furthermore, MoEShard minimizes kernel launches by fusing decomposed expert computations, significantly improving throughput. We evaluate MoEShard against DeepSpeed on encoder-based architectures, demonstrating speedups of up to 6.4$\times$ in time to first token (TTFT). Our results show that tensor sharding, when properly applied to experts, is a viable and effective strategy for efficient MoE inference.
Index recommendation is essential for improving query performance in database management systems (DBMSs) through creating an optimal set of indexes under specific constraints. Traditional methods, such as heuristic and learning-based approaches, are effective but face challenges like lengthy recommendation time, resource-intensive training, and poor generalization across different workloads and database schemas. To address these issues, we propose LLMIdxAdvis, a resource-efficient index advisor that uses large language models (LLMs) without extensive fine-tuning. LLMIdxAdvis frames index recommendation as a sequence-to-sequence task, taking target workload, storage constraint, and corresponding database environment as input, and directly outputting recommended indexes. It constructs a high-quality demonstration pool offline, using GPT-4-Turbo to synthesize diverse SQL queries and applying integrated heuristic methods to collect both default and refined labels. During recommendation, these demonstrations are ranked to inject database expertise via in-context learning. Additionally, LLMIdxAdvis extracts workload features involving specific column statistical information to strengthen LLM's understanding, and introduces a novel inference scaling strategy combining vertical scaling (via ''Index-Guided Major Voting'' and Best-of-N) and horizontal scaling (through iterative ''self-optimization'' with database feedback) to enhance reliability. Experiments on 3 OLAP and 2 real-world benchmarks reveal that LLMIdxAdvis delivers competitive index recommendation with reduced runtime, and generalizes effectively across different workloads and database schemas.
Deep learning models have an intrinsic privacy issue as they memorize parts of their training data, creating a privacy leakage. Membership Inference Attacks (MIA) exploit it to obtain confidential information about the data used for training, aiming to steal information. They can be repurposed as a measurement of data integrity by inferring whether it was used to train a machine learning model. While state-of-the-art attacks achieve a significant privacy leakage, their requirements are not feasible enough, hindering their role as practical tools to assess the magnitude of the privacy risk. Moreover, the most appropriate evaluation metric of MIA, the True Positive Rate at low False Positive Rate lacks interpretability. We claim that the incorporation of Few-Shot Learning techniques to the MIA field and a proper qualitative and quantitative privacy evaluation measure should deal with these issues. In this context, our proposal is twofold. We propose a Few-Shot learning based MIA, coined as the FeS-MIA model, which eases the evaluation of the privacy breach of a deep learning model by significantly reducing the number of resources required for the purpose. Furthermore, we propose an interpretable quantitative and qualitative measure of privacy, referred to as Log-MIA measure. Jointly, these proposals provide new tools to assess the privacy leakage and to ease the evaluation of the training data integrity of deep learning models, that is, to analyze the privacy breach of a deep learning model. Experiments carried out with MIA over image classification and language modeling tasks and its comparison to the state-of-the-art show that our proposals excel at reporting the privacy leakage of a deep learning model with little extra information.
Previous studies have established that language models manifest stereotyped biases. Existing debiasing strategies, such as retraining a model with counterfactual data, representation projection, and prompting often fail to efficiently eliminate bias or directly alter the models' biased internal representations. To address these issues, we propose BiasEdit, an efficient model editing method to remove stereotypical bias from language models through lightweight networks that act as editors to generate parameter updates. BiasEdit employs a debiasing loss guiding editor networks to conduct local edits on partial parameters of a language model for debiasing while preserving the language modeling abilities during editing through a retention loss. Experiments on StereoSet and Crows-Pairs demonstrate the effectiveness, efficiency, and robustness of BiasEdit in eliminating bias compared to tangental debiasing baselines and little to no impact on the language models' general capabilities. In addition, we conduct bias tracing to probe bias in various modules and explore bias editing impacts on different components of language models.
Deep Learning (DL) has emerged as a powerful tool for vulnerability detection, often outperforming traditional solutions. However, developing effective DL models requires large amounts of real-world data, which can be difficult to obtain in sufficient quantities. To address this challenge, DiverseVul dataset has been curated as the largest dataset of vulnerable and non-vulnerable C/C++ functions extracted exclusively from real-world projects. Its goal is to provide high-quality, large-scale samples for training DL models. However, during our study several inconsistencies were identified in the raw dataset while applying pre-processing techniques, highlighting the need for a refined version. In this work, we present a refined version of DiverseVul dataset, which is used to fine-tune a large language model, LLaMA 3.2, for vulnerability detection. Experimental results show that the use of pre-processing techniques led to an improvement in performance, with the model achieving an F1-Score of 66%, a competitive result when compared to our baseline, which achieved a 47% F1-Score in software vulnerability detection.
Existing benchmarks are becoming saturated and struggle to separate model performances due to factors like data contamination and advancing LLM capabilities. This paper introduces EMDM (Enhanced Model Differentiation Metric), a novel weighted metric that revitalizes benchmarks by enhancing model separation. EMDM integrates final answer and Chain-of-Thought (CoT) reasoning correctness, assigning weights based on the complexity and reasoning depth required to solve a given sample in the evaluation data. Using a baseline LLM in two setups-Unguided, where the model has no prior exposure to test samples, and Guided, where the model has prior knowledge of the desired answer-EMDM distinguishes instances of varying difficulty. The CoT and answer correctness from these setups inform an optimization objective for weight assignment, resulting in a more nuanced evaluation of model performance. Compared to the exact match (EM) metric, which achieves 17% separation on ARC-Challenge, EMDM achieves 46%, demonstrating its effectiveness in differentiating models based on reasoning and knowledge requirements.
Large Language Model (LLM) services often face challenges in achieving low inference latency and meeting Service Level Objectives (SLOs) under dynamic request patterns. Speculative decoding, which exploits lightweight models for drafting and LLMs for verification, has emerged as a compelling technique to accelerate LLM inference. However, existing speculative decoding solutions often fail to adapt to varying workloads and system environments, resulting in performance variability and SLO violations. In this paper, we introduce SpecServe, an efficient LLM inference system that dynamically adjusts speculative strategies according to real-time request loads and system configurations. SpecServe proposes a theoretical model to understand and predict the efficiency of speculative decoding across diverse scenarios. Additionally, it implements intelligent drafting and verification algorithms to guarantee optimal performance while achieving high SLO attainment. Experimental results on real-world LLM traces demonstrate that SpecServe consistently meets SLOs and achieves substantial performance improvements, yielding 1.14$\times$-14.3$\times$ speedups over state-of-the-art speculative inference systems.
Despite recent efforts in understanding the compression impact on large language models (LLMs) in terms of their downstream task performance and trustworthiness on relatively simpler uni-modal benchmarks (for example, question answering, common sense reasoning), their detailed study on multi-modal Large Vision-Language Models (LVLMs) is yet to be unveiled. Towards mitigating this gap, we present LVLM-Compress-Bench, a framework to first thoroughly study the broad impact of compression on the generative performance of LVLMs with multi-modal input driven tasks. In specific, we consider two major classes of compression for autoregressive models, namely KV cache and weight compression, for the dynamically growing intermediate cache and static weights, respectively. We use four LVLM variants of the popular LLaVA framework to present our analysis via integrating various state-of-the-art KV and weight compression methods including uniform, outlier-reduced, and group quantization for the KV cache and weights. With this framework we demonstrate on ten different multi-modal datasets with different capabilities including recognition, knowledge, language generation, spatial awareness, visual reasoning, hallucination and visual illusion identification, toxicity, stereotypes and bias. In specific, our framework demonstrates the compression impact on both general and ethically critical metrics leveraging a combination of real world and synthetic datasets to encompass diverse societal intersectional attributes. Extensive experimental evaluations yield diverse and intriguing observations on the behavior of LVLMs at different quantization budget of KV and weights, in both maintaining and losing performance as compared to the baseline model with FP16 data format. Code will be open-sourced at https://github.com/opengear-project/LVLM-compress-bench.
We study the task of personalized federated fine-tuning with heterogeneous data in the context of language models, where clients collaboratively fine-tune a language model (e.g., BERT, GPT) without sharing their local data, achieving personalization simultaneously. While recent efforts have applied parameter-efficient fine-tuning techniques like low-rank adaptation (LoRA) in federated settings, they typically use single or multiple independent low-rank adapters with predefined maximal and minimal ranks, which may not be optimal for diverse data sources over clients. To address this issue, we propose PF2LoRA, a new personalized federated fine-tuning algorithm built on a novel \emph{automatic rank learning approach via two-level LoRA}. Given the pretrained language model whose weight is frozen, our algorithm aims to learn two levels of adaptation simultaneously: the first level aims to learn a common adapter for all clients, while the second level fosters individual client personalization. A key advantage of PF2LoRA is its ability to adaptively determine a suitable rank based on an individual client's data, rather than relying on a predefined rank that is agnostic to data heterogeneity. We present a synthetic example that highlights how PF2LoRA automatically learns the ground-truth rank for each client, tailoring the adaptation to match the properties of their individual data. Notably, this approach introduces minimal additional memory overhead, as the second-level adaptation comprises a small number of parameters compared to the first level. Our experiments on natural language understanding and generation tasks demonstrate that PF2LoRA significantly outperforms existing federated fine-tuning methods.
Improvements in language model capabilities are often attributed to increasing model size or training data, but in some cases smaller models trained on curated data or with different architectural decisions can outperform larger ones trained on more tokens. What accounts for this? To quantify the impact of these design choices, we meta-analyze 92 open-source pretrained models across a wide array of scales, including state-of-the-art open-weights models as well as less performant models and those with less conventional design decisions. We find that by incorporating features besides model size and number of training tokens, we can achieve a relative 3-28% increase in ability to predict downstream performance compared with using scale alone. Analysis of model design decisions reveal insights into data composition, such as the trade-off between language and code tasks at 15-25\% code, as well as the better performance of some architectural decisions such as choosing rotary over learned embeddings. Broadly, our framework lays a foundation for more systematic investigation of how model development choices shape final capabilities.
Information technology has profoundly altered the way humans interact with information. The vast amount of content created, shared, and disseminated online has made it increasingly difficult to access relevant information. Over the past two decades, recommender systems and search (collectively referred to as information retrieval systems) have evolved significantly to address these challenges. Recent advances in large language models (LLMs) have demonstrated capabilities that surpass human performance in various language-related tasks and exhibit general understanding, reasoning, and decision-making abilities. This paper explores the transformative potential of LLM agents in enhancing recommender and search systems. We discuss the motivations and roles of LLM agents, and establish a classification framework to elaborate on the existing research. We highlight the immense potential of LLM agents in addressing current challenges in recommendation and search, providing insights into future research directions. This paper is the first to systematically review and classify the research on LLM agents in these domains, offering a novel perspective on leveraging this advanced AI technology for information retrieval. To help understand the existing works, we list the existing papers on LLM agent based recommendation and search at this link: https://github.com/tsinghua-fib-lab/LLM-Agent-for-Recommendation-and-Search.
Autoregressive (AR) models for image generation typically adopt a two-stage paradigm of vector quantization and raster-scan ``next-token prediction", inspired by its great success in language modeling. However, due to the huge modality gap, image autoregressive models may require a systematic reevaluation from two perspectives: tokenizer format and regression direction. In this paper, we introduce the frequency progressive autoregressive (\textbf{FAR}) paradigm and instantiate FAR with the continuous tokenizer. Specifically, we identify spectral dependency as the desirable regression direction for FAR, wherein higher-frequency components build upon the lower one to progressively construct a complete image. This design seamlessly fits the causality requirement for autoregressive models and preserves the unique spatial locality of image data. Besides, we delve into the integration of FAR and the continuous tokenizer, introducing a series of techniques to address optimization challenges and improve the efficiency of training and inference processes. We demonstrate the efficacy of FAR through comprehensive experiments on the ImageNet dataset and verify its potential on text-to-image generation.
Video anomaly detection (VAD) is crucial for video analysis and surveillance in computer vision. However, existing VAD models rely on learned normal patterns, which makes them difficult to apply to diverse environments. Consequently, users should retrain models or develop separate AI models for new environments, which requires expertise in machine learning, high-performance hardware, and extensive data collection, limiting the practical usability of VAD. To address these challenges, this study proposes customizable video anomaly detection (C-VAD) technique and the AnyAnomaly model. C-VAD considers user-defined text as an abnormal event and detects frames containing a specified event in a video. We effectively implemented AnyAnomaly using a context-aware visual question answering without fine-tuning the large vision language model. To validate the effectiveness of the proposed model, we constructed C-VAD datasets and demonstrated the superiority of AnyAnomaly. Furthermore, our approach showed competitive performance on VAD benchmark datasets, achieving state-of-the-art results on the UBnormal dataset and outperforming other methods in generalization across all datasets. Our code is available online at github.com/SkiddieAhn/Paper-AnyAnomaly.
Large Language Model based multi-agent systems are revolutionizing autonomous communication and collaboration, yet they remain vulnerable to security threats like unauthorized access and data breaches. To address this, we introduce AgentSafe, a novel framework that enhances MAS security through hierarchical information management and memory protection. AgentSafe classifies information by security levels, restricting sensitive data access to authorized agents. AgentSafe incorporates two components: ThreatSieve, which secures communication by verifying information authority and preventing impersonation, and HierarCache, an adaptive memory management system that defends against unauthorized access and malicious poisoning, representing the first systematic defense for agent memory. Experiments across various LLMs show that AgentSafe significantly boosts system resilience, achieving defense success rates above 80% under adversarial conditions. Additionally, AgentSafe demonstrates scalability, maintaining robust performance as agent numbers and information complexity grow. Results underscore effectiveness of AgentSafe in securing MAS and its potential for real-world application.
Temporal progression is an integral part of knowledge accumulation and update. Web search is frequently adopted as grounding for agent knowledge, yet an improper configuration affects the quality of the agent's responses. Here, we assess the agent behavior using distinct date-controlled tools (DCTs) as stress test to measure the knowledge variability of large language model (LLM) agents. We demonstrate the temporal effects of an LLM agent as a writing assistant, which uses web search to complete scientific publication abstracts. We show that the temporality of search engine translates into tool-dependent agent performance but can be alleviated with base model choice and explicit reasoning instructions such as chain-of-thought prompting. Our results indicate that agent design and evaluations should take a dynamical view and implement measures to account for the temporal influence of external resources to ensure reliability.
Privacy-preserving medical decision support for kidney disease requires localized deployment of large language models (LLMs) while maintaining clinical reasoning capabilities. Current solutions face three challenges: 1) Cloud-based LLMs pose data security risks; 2) Local model deployment demands technical expertise; 3) General LLMs lack mechanisms to integrate medical knowledge. Retrieval-augmented systems also struggle with medical document processing and clinical usability. We developed KidneyTalk-open, a desktop system integrating three technical components: 1) No-code deployment of state-of-the-art (SOTA) open-source LLMs (such as DeepSeek-r1, Qwen2.5) via local inference engine; 2) Medical document processing pipeline combining context-aware chunking and intelligent filtering; 3) Adaptive Retrieval and Augmentation Pipeline (AddRep) employing agents collaboration for improving the recall rate of medical documents. A graphical interface was designed to enable clinicians to manage medical documents and conduct AI-powered consultations without technical expertise. Experimental validation on 1,455 challenging nephrology exam questions demonstrates AddRep's effectiveness: achieving 29.1% accuracy (+8.1% over baseline) with intelligent knowledge integration, while maintaining robustness through 4.9% rejection rate to suppress hallucinations. Comparative case studies with the mainstream products (AnythingLLM, Chatbox, GPT4ALL) demonstrate KidneyTalk-open's superior performance in real clinical query. KidneyTalk-open represents the first no-code medical LLM system enabling secure documentation-enhanced medical Q&A on desktop. Its designs establishes a new framework for privacy-sensitive clinical AI applications. The system significantly lowers technical barriers while improving evidence traceability, enabling more medical staff or patients to use SOTA open-source LLMs conveniently.
Efficient vision-language understanding of large Remote Sensing Images (RSIs) is meaningful but challenging. Current Large Vision-Language Models (LVLMs) typically employ limited pre-defined grids to process images, leading to information loss when handling gigapixel RSIs. Conversely, using unlimited grids significantly increases computational costs. To preserve image details while reducing computational complexity, we propose a text-guided token pruning method with Dynamic Image Pyramid (DIP) integration. Our method introduces: (i) a Region Focus Module (RFM) that leverages text-aware region localization capability to identify critical vision tokens, and (ii) a coarse-to-fine image tile selection and vision token pruning strategy based on DIP, which is guided by RFM outputs and avoids directly processing the entire large imagery. Additionally, existing benchmarks for evaluating LVLMs' perception ability on large RSI suffer from limited question diversity and constrained image sizes. We construct a new benchmark named LRS-VQA, which contains 7,333 QA pairs across 8 categories, with image length up to 27,328 pixels. Our method outperforms existing high-resolution strategies on four datasets using the same data. Moreover, compared to existing token reduction methods, our approach demonstrates higher efficiency under high-resolution settings. Dataset and code are in https://github.com/VisionXLab/LRS-VQA.
We propose V2Flow, a novel tokenizer that produces discrete visual tokens capable of high-fidelity reconstruction, while ensuring structural and latent distribution alignment with the vocabulary space of large language models (LLMs). Leveraging this tight visual-vocabulary coupling, V2Flow enables autoregressive visual generation on top of existing LLMs. Our approach formulates visual tokenization as a flow-matching problem, aiming to learn a mapping from a standard normal prior to the continuous image distribution, conditioned on token sequences embedded within the LLMs vocabulary space. The effectiveness of V2Flow stems from two core designs. First, we propose a Visual Vocabulary resampler, which compresses visual data into compact token sequences, with each represented as a soft categorical distribution over LLM's vocabulary. This allows seamless integration of visual tokens into existing LLMs for autoregressive visual generation. Second, we present a masked autoregressive Rectified-Flow decoder, employing a masked transformer encoder-decoder to refine visual tokens into contextually enriched embeddings. These embeddings then condition a dedicated velocity field for precise reconstruction. Additionally, an autoregressive rectified-flow sampling strategy is incorporated, ensuring flexible sequence lengths while preserving competitive reconstruction quality. Extensive experiments show that V2Flow outperforms mainstream VQ-based tokenizers and facilitates autoregressive visual generation on top of existing. https://github.com/zhangguiwei610/V2Flow
Existing multimodal UAV object detection methods often overlook the impact of semantic gaps between modalities, which makes it difficult to achieve accurate semantic and spatial alignments, limiting detection performance. To address this problem, we propose a Large Language Model (LLM) guided Progressive feature Alignment Network called LPANet, which leverages the semantic features extracted from a large language model to guide the progressive semantic and spatial alignment between modalities for multimodal UAV object detection. To employ the powerful semantic representation of LLM, we generate the fine-grained text descriptions of each object category by ChatGPT and then extract the semantic features using the large language model MPNet. Based on the semantic features, we guide the semantic and spatial alignments in a progressive manner as follows. First, we design the Semantic Alignment Module (SAM) to pull the semantic features and multimodal visual features of each object closer, alleviating the semantic differences of objects between modalities. Second, we design the Explicit Spatial alignment Module (ESM) by integrating the semantic relations into the estimation of feature-level offsets, alleviating the coarse spatial misalignment between modalities. Finally, we design the Implicit Spatial alignment Module (ISM), which leverages the cross-modal correlations to aggregate key features from neighboring regions to achieve implicit spatial alignment. Comprehensive experiments on two public multimodal UAV object detection datasets demonstrate that our approach outperforms state-of-the-art multimodal UAV object detectors.
Visual grounding seeks to localize the image region corresponding to a free-form text description. Recently, the strong multimodal capabilities of Large Vision-Language Models (LVLMs) have driven substantial improvements in visual grounding, though they inevitably require fine-tuning and additional model components to explicitly generate bounding boxes or segmentation masks. However, we discover that a few attention heads in frozen LVLMs demonstrate strong visual grounding capabilities. We refer to these heads, which consistently capture object locations related to text semantics, as localization heads. Using localization heads, we introduce a straightforward and effective training-free visual grounding framework that utilizes text-to-image attention maps from localization heads to identify the target objects. Surprisingly, only three out of thousands of attention heads are sufficient to achieve competitive localization performance compared to existing LVLM-based visual grounding methods that require fine-tuning. Our findings suggest that LVLMs can innately ground objects based on a deep comprehension of the text-image relationship, as they implicitly focus on relevant image regions to generate informative text outputs. All the source codes will be made available to the public.