arXiv subjects

Haifeng Wang

Publications and source records attributed to Haifeng Wang.

At least 73 records · Page 4Linked to original sources

arXiv2025

SafeToolBench: Pioneering a Prospective Benchmark to Evaluating Tool Utilization Safety in LLMs

Large Language Models (LLMs) have exhibited great performance in autonomously calling various tools in external environments, leading to better problem solving and task automation capabilities. However, these external tools also amplify potential risks such as financial loss or privacy leakage with ambiguous or malicious user instructions. Compared to previous studies, which mainly assess the safety awareness of LLMs after obtaining the tool execution results (i.e., retrospective evaluation), this paper focuses on prospective ways to assess the safety of LLM tool utilization, aiming to avoid irreversible harm caused by directly executing tools. To this end, we propose SafeToolBench, the first benchmark to comprehensively assess tool utilization security in a prospective manner, covering malicious user instructions and diverse practical toolsets. Additionally, we propose a novel framework, SafeInstructTool, which aims to enhance LLMs' awareness of tool utilization security from three perspectives (i.e., \textit{User Instruction, Tool Itself, and Joint Instruction-Tool}), leading to nine detailed dimensions in total. We experiment with four LLMs using different methods, revealing that existing approaches fail to capture all risks in tool utilization. In contrast, our framework significantly enhances LLMs' self-awareness, enabling a more safe and trustworthy tool utilization.

cs.CR↗

arXiv2025

RepoDebug: Repository-Level Multi-Task and Multi-Language Debugging Evaluation of Large Language Models

Large Language Models (LLMs) have exhibited significant proficiency in code debugging, especially in automatic program repair, which may substantially reduce the time consumption of developers and enhance their efficiency. Significant advancements in debugging datasets have been made to promote the development of code debugging. However, these datasets primarily focus on assessing the LLM's function-level code repair capabilities, neglecting the more complex and realistic repository-level scenarios, which leads to an incomplete understanding of the LLM's challenges in repository-level debugging. While several repository-level datasets have been proposed, they often suffer from limitations such as limited diversity of tasks, languages, and error types. To mitigate this challenge, this paper introduces RepoDebug, a multi-task and multi-language repository-level code debugging dataset with 22 subtypes of errors that supports 8 commonly used programming languages and 3 debugging tasks. Furthermore, we conduct evaluation experiments on 10 LLMs, where Claude 3.5 Sonnect, the best-performing model, still cannot perform well in repository-level debugging.

cs.SE↗

arXiv2025

Myocardial T1 mapping at 5T using multi-inversion recovery real-time spoiled GRE

Objective: To develop an accurate myocardial T1 mapping technique at 5T using Look-Locker-based multiple inversion-recovery with the real-time spoiled gradient echo (GRE) acquisition. Approach: The proposed T1 mapping technique (mIR-rt) samples the recovery of inverted magnetization using the real-time GRE and the images captured during diastole are selected for T1 fitting. Multiple-inversion recoveries are employed to increase the sample size for accurate fitting. The T1 mapping method was validated using Bloch simulation, phantom studies, and in 16 healthy volunteers at 5T. Main results: In both simulation and phantom studies, the T1 values measured by mIR-rt closely approximate the reference T1 values, with errors less than 3%, while the conventional MOLLI sequence underestimates T1 values. The myocardial T1 values at 5T are 1553 +/- 52 ms, 1531 +/- 53 ms, and 1526 +/- 60 ms (mean +/- standard deviation) at the apex, middle, and base, respectively. The T1 values measured by MOLLI (1350 +/- 48 ms, 1349 +/- 47 ms, and 1354 +/- 45 ms at the apex, middle, and base) were significantly lower than those of mIR-rt with p<0.001 for all three layers. The mIR-rt sequence method used in our study provides high reproducibility, particularly in the middle slices, supporting its practical relevance for myocardial T1 mapping. Significance: The proposed method is feasible for myocardial T1 mapping at 5T and provides better accuracy than the conventional MOLLI sequence.

physics.med-ph↗

arXiv2025

Red alert: Millions of "homeless" publications in Scopus should be resettled

Scopus is increasingly regarded as a high-quality and reliable data source for research and evaluation of scientific and scholarly activity. However, a puzzling phenomenon has been discovered occasionally: millions of records with author affiliation information collected in Scopus are oddly labeled as "country-undefined" by Scopus which is rarely to be detected in its counterpart Web of Science. This huge number of "homeless" records in Scopus is unacceptable for a widely used high-quality bibliographic database. By using data from the past 124 years, this brief communication tries to probe these affiliated but country-undefined records in Scopus. Our analysis identifies four primary causes for these "homeless" records: incomplete author affiliation addresses, Scopus' inability to recognize different variants of country/territory names, misspelled country/territory names in author affiliation addresses, and Scopus' insufficiency in correctly split and identify the clean affiliation addresses. To address this pressing issue, we put forward several recommendations to relevant stakeholders, with the aim of resettling millions of "homeless" records in Scopus and reducing its potential impact on Scopus-based literature retrieval, analysis, and evaluation.

cs.DL↗

arXiv2025

Gaussian kernel-based motion measurement

The growing demand for structural health monitoring has driven increasing interest in high-precision motion measurement, as structural information derived from extracted motions can effectively reflect the current condition of the structure. Among various motion measurement techniques, vision-based methods stand out due to their low cost, easy installation, and large-scale measurement. However, when it comes to sub-pixel-level motion measurement, current vision-based methods either lack sufficient accuracy or require extensive manual parameter tuning (e.g., pyramid layers, target pixels, and filter parameters) to reach good precision. To address this issue, we developed a novel Gaussian kernel-based motion measurement method, which can extract the motion between different frames via tracking the location of Gaussian kernels. The motion consistency, which fits practical structural conditions, and a super-resolution constraint, are introduced to increase accuracy and robustness of our method. Numerical and experimental validations show that it can consistently reach high accuracy without customized parameter setup for different test samples.

cs.CV↗

arXiv2025

CCQ: Convolutional Code for Extreme Low-bit Quantization in LLMs

The rapid scaling of Large Language Models (LLMs) elevates inference costs and compounds substantial deployment barriers. While quantization to 8 or 4 bits mitigates this, sub-3-bit methods face severe accuracy, scalability, and efficiency degradation. We propose Convolutional Code Quantization (CCQ), an inference-optimized quantization approach compressing LLMs to 2.0-2.75 bits with minimal accuracy loss. Departing from error-prone scalar quantization or slow vector quantization, CCQ integrates a hardware-aware bit-shift encoding and decoding solution with Convolutional Code, Hybrid Encoding, and Code Cluster, jointly overcoming accuracy-speed bottlenecks. We construct a lookup-free encoding space, enabling a linear mapping between the codebook and weight vectors, thereby optimizing inference performance. Meanwhile, by drawing on the concept of data mapping from vector quantization, we minimize the performance degradation of the model under extremely low-bit conditions. Experiments demonstrate that CCQ achieves outstanding performance on LLMs across various benchmarks. We compress DeepSeek-V3 (671B total parameters) to 184GB and ERNIE-4.5-300B-A47B to 89GB, enabling single-GPU deployment of ERNIE 4.5 and eliminating inter-card communication. The 2-bit ERNIE-4.5-300B-A47B model and inference engine have been open-sourced.

cs.LG↗

arXiv2025

BeamLoRA: Beam-Constraint Low-Rank Adaptation

Due to the demand for efficient fine-tuning of large language models, Low-Rank Adaptation (LoRA) has been widely adopted as one of the most effective parameter-efficient fine-tuning methods. Nevertheless, while LoRA improves efficiency, there remains room for improvement in accuracy. Herein, we adopt a novel perspective to assess the characteristics of LoRA ranks. The results reveal that different ranks within the LoRA modules not only exhibit varying levels of importance but also evolve dynamically throughout the fine-tuning process, which may limit the performance of LoRA. Based on these findings, we propose BeamLoRA, which conceptualizes each LoRA module as a beam where each rank naturally corresponds to a potential sub-solution, and the fine-tuning process becomes a search for the optimal sub-solution combination. BeamLoRA dynamically eliminates underperforming sub-solutions while expanding the parameter space for promising ones, enhancing performance with a fixed rank. Extensive experiments across three base models and 12 datasets spanning math reasoning, code generation, and commonsense reasoning demonstrate that BeamLoRA consistently enhances the performance of LoRA, surpassing the other baseline methods.

cs.CL↗

arXiv2025

Curiosity-Driven Reinforcement Learning from Human Feedback

Reinforcement learning from human feedback (RLHF) has proven effective in aligning large language models (LLMs) with human preferences, but often at the cost of reduced output diversity. This trade-off between diversity and alignment quality remains a significant challenge. Drawing inspiration from curiosity-driven exploration in reinforcement learning, we introduce curiosity-driven RLHF (CD-RLHF), a framework that incorporates intrinsic rewards for novel states, alongside traditional sparse extrinsic rewards, to optimize both output diversity and alignment quality. We demonstrate the effectiveness of CD-RLHF through extensive experiments on a range of tasks, including text summarization and instruction following. Our approach achieves significant gains in diversity on multiple diversity-oriented metrics while maintaining alignment with human preferences comparable to standard RLHF. We make our code publicly available at https://github.com/ernie-research/CD-RLHF.

cs.CL↗

arXiv2025

TransBench: Breaking Barriers for Transferable Graphical User Interface Agents in Dynamic Digital Environments

Graphical User Interface (GUI) agents, which autonomously operate on digital interfaces through natural language instructions, hold transformative potential for accessibility, automation, and user experience. A critical aspect of their functionality is grounding - the ability to map linguistic intents to visual and structural interface elements. However, existing GUI agents often struggle to adapt to the dynamic and interconnected nature of real-world digital environments, where tasks frequently span multiple platforms and applications while also being impacted by version updates. To address this, we introduce TransBench, the first benchmark designed to systematically evaluate and enhance the transferability of GUI agents across three key dimensions: cross-version transferability (adapting to version updates), cross-platform transferability (generalizing across platforms like iOS, Android, and Web), and cross-application transferability (handling tasks spanning functionally distinct apps). TransBench includes 15 app categories with diverse functionalities, capturing essential pages across versions and platforms to enable robust evaluation. Our experiments demonstrate significant improvements in grounding accuracy, showcasing the practical utility of GUI agents in dynamic, real-world environments. Our code and data will be publicly available at GitHub.

cs.HC↗

arXiv2025

HomeBench: Evaluating LLMs in Smart Homes with Valid and Invalid Instructions Across Single and Multiple Devices

Large language models (LLMs) have the potential to revolutionize smart home assistants by enhancing their ability to accurately understand user needs and respond appropriately, which is extremely beneficial for building a smarter home environment. While recent studies have explored integrating LLMs into smart home systems, they primarily focus on handling straightforward, valid single-device operation instructions. However, real-world scenarios are far more complex and often involve users issuing invalid instructions or controlling multiple devices simultaneously. These have two main challenges: LLMs must accurately identify and rectify errors in user instructions and execute multiple user instructions perfectly. To address these challenges and advance the development of LLM-based smart home assistants, we introduce HomeBench, the first smart home dataset with valid and invalid instructions across single and multiple devices in this paper. We have experimental results on 13 distinct LLMs; e.g., GPT-4o achieves only a 0.0% success rate in the scenario of invalid multi-device instructions, revealing that the existing state-of-the-art LLMs still cannot perform well in this situation even with the help of in-context learning, retrieval-augmented generation, and fine-tuning. Our code and dataset are publicly available at https://github.com/BITHLP/HomeBench.

cs.CL↗

arXiv2025

ToolSpectrum : Towards Personalized Tool Utilization for Large Language Models

While integrating external tools into large language models (LLMs) enhances their ability to access real-time information and domain-specific services, existing approaches focus narrowly on functional tool selection following user instructions, overlooking the context-aware personalization in tool selection. This oversight leads to suboptimal user satisfaction and inefficient tool utilization, particularly when overlapping toolsets require nuanced selection based on contextual factors. To bridge this gap, we introduce ToolSpectrum, a benchmark designed to evaluate LLMs' capabilities in personalized tool utilization. Specifically, we formalize two key dimensions of personalization, user profile and environmental factors, and analyze their individual and synergistic impacts on tool utilization. Through extensive experiments on ToolSpectrum, we demonstrate that personalized tool utilization significantly improves user experience across diverse scenarios. However, even state-of-the-art LLMs exhibit the limited ability to reason jointly about user profiles and environmental factors, often prioritizing one dimension at the expense of the other. Our findings underscore the necessity of context-aware personalization in tool-augmented LLMs and reveal critical limitations for current models. Our data and code are available at https://github.com/Chengziha0/ToolSpectrum.

cs.CL↗

arXiv2025

Unveiling Knowledge Utilization Mechanisms in LLM-based Retrieval-Augmented Generation

Considering the inherent limitations of parametric knowledge in large language models (LLMs), retrieval-augmented generation (RAG) is widely employed to expand their knowledge scope. Since RAG has shown promise in knowledge-intensive tasks like open-domain question answering, its broader application to complex tasks and intelligent assistants has further advanced its utility. Despite this progress, the underlying knowledge utilization mechanisms of LLM-based RAG remain underexplored. In this paper, we present a systematic investigation of the intrinsic mechanisms by which LLMs integrate internal (parametric) and external (retrieved) knowledge in RAG scenarios. Specially, we employ knowledge stream analysis at the macroscopic level, and investigate the function of individual modules at the microscopic level. Drawing on knowledge streaming analyses, we decompose the knowledge utilization process into four distinct stages within LLM layers: knowledge refinement, knowledge elicitation, knowledge expression, and knowledge contestation. We further demonstrate that the relevance of passages guides the streaming of knowledge through these stages. At the module level, we introduce a new method, knowledge activation probability entropy (KAPE) for neuron identification associated with either internal or external knowledge. By selectively deactivating these neurons, we achieve targeted shifts in the LLM's reliance on one knowledge source over the other. Moreover, we discern complementary roles for multi-head attention and multi-layer perceptron layers during knowledge formation. These insights offer a foundation for improving interpretability and reliability in retrieval-augmented LLMs, paving the way for more robust and transparent generative solutions in knowledge-intensive domains.

cs.CL↗

arXiv2025

BEVWorld: A Multimodal World Simulator for Autonomous Driving via Scene-Level BEV Latents

World models have attracted increasing attention in autonomous driving for their ability to forecast potential future scenarios. In this paper, we propose BEVWorld, a novel framework that transforms multimodal sensor inputs into a unified and compact Bird's Eye View (BEV) latent space for holistic environment modeling. The proposed world model consists of two main components: a multi-modal tokenizer and a latent BEV sequence diffusion model. The multi-modal tokenizer first encodes heterogeneous sensory data, and its decoder reconstructs the latent BEV tokens into LiDAR and surround-view image observations via ray-casting rendering in a self-supervised manner. This enables joint modeling and bidirectional encoding-decoding of panoramic imagery and point cloud data within a shared spatial representation. On top of this, the latent BEV sequence diffusion model performs temporally consistent forecasting of future scenes, conditioned on high-level action tokens, enabling scene-level reasoning over time. Extensive experiments demonstrate the effectiveness of BEVWorld on autonomous driving benchmarks, showcasing its capability in realistic future scene generation and its benefits for downstream tasks such as perception and motion prediction.

cs.CV↗

arXiv2025

FlashMask: Efficient and Rich Mask Extension of FlashAttention

The computational and memory demands of vanilla attention scale quadratically with the sequence length $N$, posing significant challenges for processing long sequences in Transformer models. FlashAttention alleviates these challenges by eliminating the $O(N^2)$ memory dependency and reducing attention latency through IO-aware memory optimizations. However, its native support for certain attention mask types is limited, and it does not inherently accommodate more complex masking requirements. Previous approaches resort to using dense masks with $O(N^2)$ memory complexity, leading to inefficiencies. In this paper, we propose FlashMask, an extension of FlashAttention that introduces a column-wise sparse representation of attention masks. This approach efficiently represents a wide range of mask types and facilitates the development of optimized kernel implementations. By adopting this novel representation, FlashMask achieves linear memory complexity $O(N)$, suitable for modeling long-context sequences. Moreover, this representation enables kernel optimizations that eliminate unnecessary computations by leveraging sparsity in the attention mask, without sacrificing computational accuracy, resulting in higher computational efficiency. We evaluate FlashMask's performance in fine-tuning and alignment training of LLMs such as SFT, LoRA, DPO, and RM. FlashMask achieves significant throughput improvements, with end-to-end speedups ranging from 1.65x to 3.22x compared to existing FlashAttention dense method. Additionally, our kernel-level comparisons demonstrate that FlashMask surpasses the latest counterpart, FlexAttention, by 12.1% to 60.7% in terms of kernel TFLOPs/s, achieving 37.8% to 62.3% of the theoretical maximum FLOPs/s on the A100 GPU. The code is open-sourced on PaddlePaddle and integrated into PaddleNLP, supporting models with over 100 billion parameters for contexts up to 128K tokens.

cs.LG↗

arXiv2025

Inner Thinking Transformer: Leveraging Dynamic Depth Scaling to Foster Adaptive Internal Thinking

Large language models (LLMs) face inherent performance bottlenecks under parameter constraints, particularly in processing critical tokens that demand complex reasoning. Empirical analysis reveals challenging tokens induce abrupt gradient spikes across layers, exposing architectural stress points in standard Transformers. Building on this insight, we propose Inner Thinking Transformer (ITT), which reimagines layer computations as implicit thinking steps. ITT dynamically allocates computation through Adaptive Token Routing, iteratively refines representations via Residual Thinking Connections, and distinguishes reasoning phases using Thinking Step Encoding. ITT enables deeper processing of critical tokens without parameter expansion. Evaluations across 162M-466M parameter models show ITT achieves 96.5\% performance of a 466M Transformer using only 162M parameters, reduces training data by 43.2\%, and outperforms Transformer/Loop variants in 11 benchmarks. By enabling elastic computation allocation during inference, ITT balances performance and efficiency through architecture-aware optimization of implicit thinking pathways.

cs.CL↗

arXiv2025

A unified convergence analysis framework of the energy-stable ETDRK3 schemes for the No-slope-selection thin film model

This paper establishes a unified framework for the space-time convergence analysis of the energy-stable third-order accurate exponential time differencing Runge-Kutta schemes. By employing Fourier pseudo-spectral discretization in space and the inner product technique, we derive a rigorous Fourier eigenvalue analysis, which provides a detailed optimal convergence rate and error estimate. The primary challenge is addressing the complex nonlinear terms in the NSS equation. Fortunately, this challenge could be resolved through careful eigenvalue bound estimates for various operators.

math.NA↗

arXiv2025

K-space Diffusion Model Based MR Reconstruction Method for Simultaneous Multislice Imaging

Simultaneous Multi-Slice(SMS) is a magnetic resonance imaging (MRI) technique which excites several slices concurrently using multiband radiofrequency pulses to reduce scanning time. However, due to its variable data structure and difficulty in acquisition, it is challenging to integrate SMS data as training data into deep learning frameworks.This study proposed a novel k-space diffusion model of SMS reconstruction that does not utilize SMS data for training. Instead, it incorporates Slice GRAPPA during the sampling process to reconstruct SMS data from different acquisition modes.Our results demonstrated that this method outperforms traditional SMS reconstruction methods and can achieve higher acceleration factors without in-plane aliasing.

eess.IV↗

arXiv2025

Robust altermagnetism and compensated ferrimagnetism in MnPX$_3$-based (X = S or Se) heterostructures

The recent research interests in the non-relativistic spin splitting of electronic band structures have led to the exploration of altermagnets and other compensated magnets. Here, we show that various types of non-relativistic spin splitting can be robustly induced by constructing Van der Waals heterostructures consisting of materials with intra-plane anti-ferromagnetic orders and suitable substrates. Using MnPX$_3$ (X = S or Se) as an example, which has a Néel magnetic order, we demonstrate that altermagnetic spin splitting can arise in the AA-stacking MnPX$_3$/MPX$_3$ (M = Cd, Mg, or Zn) heterostructures. For the AB-stacking heterostructures that are semiconducting, ferrimagnetic-type spin splitting emerges, and the fully compensated magnetization is protected by the Luttinger theorem. By combining with a Van der Waals ferroelectric substrate like CuInP$_2$S$_6$, MnPX$_3$-based heterostructures can show tunable spin splitting and spin-related properties that depend on the electronic band structures and ferroelectric polarization, which can be non-volatilely reversed by applying an out-of-plane electric field. Our study provides a route to induce tunable non-relativistic spin splitting in experimentally synthesizable two-dimensional magnets.

cond-mat.mtrl-sci↗