Large Language Models (LLMs) have shown impressive zero-shot performance across a variety of Natural Language Processing tasks, including document re-ranking. However, their effectiveness degrades on unseen tasks and domains, largely due to shifts in vocabulary and word distributions. In this paper, we investigate Task Arithmetic, a technique that combines the weights of LLMs pre-trained on different tasks or domains via simple mathematical operations, such as addition or subtraction, to adapt retrieval models without requiring additional fine-tuning. Our method is able to synthesize diverse tasks and domain knowledge into a single model, enabling effective zero-shot adaptation in different retrieval contexts. Extensive experiments on publicly available scientific, biomedical, and multilingual datasets show that our method improves state-of-the-art re-ranking performance by up to 18% in NDCG@10 and 15% in P@10. In addition to these empirical gains, our analysis provides insights into the strengths and limitations of Task Arithmetic as a practical strategy for zero-shot learning and model adaptation. We make our code publicly available at https://github.com/DetectiveMB/Task-Arithmetic-for-ZS-IR.
We introduce OSVBench, a new benchmark for evaluating Large Language Models (LLMs) in generating complete specification code pertaining to operating system kernel verification tasks. The benchmark first defines the specification generation problem into a program synthesis problem within a confined scope of syntax and semantics by providing LLMs with the programming model. The LLMs are required to understand the provided verification assumption and the potential syntax and semantics space to search for, then generate the complete specification for the potentially buggy operating system code implementation under the guidance of the high-level functional description of the operating system. This benchmark is built upon a real-world operating system kernel, Hyperkernel, and consists of 245 complex specification generation tasks in total, each is a long context task of about 20k-30k tokens. Our comprehensive evaluation of 12 LLMs exhibits the limited performance of the current LLMs on the specification generation tasks for operating system verification. Significant disparities in their performance on the benchmark highlight differences in their ability to handle long-context code generation tasks. The evaluation toolkit and benchmark are available at https://github.com/lishangyu-hkust/OSVBench.
Large-scale numerical computations make increasing use of low-precision (LP) floating point formats and mixed precision arithmetic, which can be enhanced by the technique of stochastic rounding (SR), that is, rounding an intermediate high-precision value up or down randomly as a function of the value's distance to the two rounding candidates. Stochastic rounding requires, in addition to the high-precision input value, a source of random bits. As the provision of high-quality random bits is an additional computational cost, it is of interest to require as few bits as possible while maintaining the desirable properties of SR in a given computation, or computational domain. This paper examines a number of possible implementations of few-bit stochastic rounding (FBSR), and shows how several natural implementations can introduce sometimes significant bias into the rounding process, which are not present in the case of infinite-bit, infinite-precision examinations of these implementations. The paper explores the impact of these biases in machine learning examples, and hence opens another class of configuration parameters of which practitioners should be aware when developing or adopting low-precision floating point. Code is available at http://github.com/graphcore-research/arith25-stochastic-rounding.
Kernel mean embeddings -- integrals of a kernel with respect to a probability distribution -- are essential in Bayesian quadrature, but also widely used in other computational tools for numerical integration or for statistical inference based on the maximum mean discrepancy. These methods often require, or are enhanced by, the availability of a closed-form expression for the kernel mean embedding. However, deriving such expressions can be challenging, limiting the applicability of kernel-based techniques when practitioners do not have access to a closed-form embedding. This paper addresses this limitation by providing a comprehensive dictionary of known kernel mean embeddings, along with practical tools for deriving new embeddings from known ones. We also provide a Python library that includes minimal implementations of the embeddings.
High-dimensional partial differential equations (PDEs) pose significant computational challenges across fields ranging from quantum chemistry to economics and finance. Although scientific machine learning (SciML) techniques offer approximate solutions, they often suffer from bias and neglect crucial physical insights. Inspired by inference-time scaling strategies in language models, we propose Simulation-Calibrated Scientific Machine Learning (SCaSML), a physics-informed framework that dynamically refines and debiases the SCiML predictions during inference by enforcing the physical laws. SCaSML leverages derived new physical laws that quantifies systematic errors and employs Monte Carlo solvers based on the Feynman-Kac and Elworthy-Bismut-Li formulas to dynamically correct the prediction. Both numerical and theoretical analysis confirms enhanced convergence rates via compute-optimal inference methods. Our numerical experiments demonstrate that SCaSML reduces errors by 20-50% compared to the base surrogate model, establishing it as the first algorithm to refine approximated solutions to high-dimensional PDE during inference. Code of SCaSML is available at https://github.com/Francis-Fan-create/SCaSML.
The internet is undergoing a historical transformation from the "Internet of Websites" to the "Internet of AgentSites." While traditional Websites served as the foundation for information hosting and dissemination, a new frontier is emerging where AgentSites serve as the hubs of the internet, where each AgentSite hosts one or more AI agents that receive tasks, address them, and deliver actionable solutions, marking a significant shift in the digital landscape and representing the next generation of online ecosystems. Under this vision, AIOS, the AI Agent Operating System, serves as the server for the development, deployment and execution of AI agents, which is a fundamental infrastructure for the Internet of Agentsites. In this paper, we introduce AIOS Server, a runtime framework to host agents and enable global-scale collaboration among decentralized agents. AIOS Server provides a communication protocol leveraging the Model Context Protocol (MCP) and JSON-RPC to enable agent-agent or human-agent interactions. Each AIOS node operates as a server to host and execute agents, while supporting peer-to-peer coordination without reliance on centralized orchestration. Based on AIOS Server, we further present the world's first practically deployed Internet of Agentsites (AIOS-IoA), including AgentHub for agent registration and discovery and AgentChat for interactive communication, at https://planet.aios.foundation. The agent discovery mechanism based on Distributed Hash Tables (DHT) and a Gossip protocol serves as the search engine for the internet of agentsites. This work provides a practical foundation for building the Internet of Agentsites-a new paradigm where autonomous agents become first-class citizens of the web. The implementation is available at https://github.com/agiresearch/AIOS.Server and is integrated into the AIOS main branch at https://github.com/agiresearch/AIOS.
Large language models (LLMs) have achieved remarkable success in contextual knowledge understanding. In this paper, we show that these concentrated massive values consistently emerge in specific regions of attention queries (Q) and keys (K) while not having such patterns in values (V) in various modern transformer-based LLMs (Q, K, and V mean the representations output by the query, key, and value layers respectively). Through extensive experiments, we further demonstrate that these massive values play a critical role in interpreting contextual knowledge (knowledge obtained from the current context window) rather than in retrieving parametric knowledge stored within the model's parameters. Our further investigation of quantization strategies reveals that ignoring these massive values leads to a pronounced drop in performance on tasks requiring rich contextual understanding, aligning with our analysis. Finally, we trace the emergence of concentrated massive values and find that such concentration is caused by Rotary Positional Encoding (RoPE), which has appeared since the first layers. These findings shed new light on how Q and K operate in LLMs and offer practical insights for model design and optimization. The Code is Available at https://github.com/MingyuJ666/Rope_with_LLM.
LLM-based agents have demonstrated great potential in generating and managing code within complex codebases. In this paper, we introduce WebGen-Bench, a novel benchmark designed to measure an LLM-based agent's ability to create multi-file website codebases from scratch. It contains diverse instructions for website generation, created through the combined efforts of human annotators and GPT-4o. These instructions span three major categories and thirteen minor categories, encompassing nearly all important types of web applications. To assess the quality of the generated websites, we use GPT-4o to generate test cases targeting each functionality described in the instructions, and then manually filter, adjust, and organize them to ensure accuracy, resulting in 647 test cases. Each test case specifies an operation to be performed on the website and the expected result after the operation. To automate testing and improve reproducibility, we employ a powerful web-navigation agent to execute tests on the generated websites and determine whether the observed responses align with the expected results. We evaluate three high-performance code-agent frameworks, Bolt.diy, OpenHands, and Aider, using multiple proprietary and open-source LLMs as engines. The best-performing combination, Bolt.diy powered by DeepSeek-R1, achieves only 27.8\% accuracy on the test cases, highlighting the challenging nature of our benchmark. Additionally, we construct WebGen-Instruct, a training set consisting of 6,667 website-generation instructions. Training Qwen2.5-Coder-32B-Instruct on Bolt.diy trajectories generated from a subset of this training set achieves an accuracy of 38.2\%, surpassing the performance of the best proprietary model.
Current neural network-based methods to the problem of document summarisation struggle when applied to datasets containing large inputs. In this paper we propose a new approach to the challenge of content-selection when dealing with end-to-end summarisation of user reviews of accommodations. We show that by combining an extractive approach with externally pre-trained sentence level embeddings in an addition to an abstractive summarisation model we can outperform existing methods when this is applied to the task of summarising a large input dataset. We also prove that predicting sentence level embedding of a summary increases the quality of an end-to-end system for loosely aligned source to target corpora, than compared to commonly predicting probability distributions of sentence selection.
Recent advances in large language models have demonstrated that Supervised Fine-Tuning (SFT) with Chain-of-Thought (CoT) reasoning data distilled from large reasoning models (e.g., DeepSeek R1) can effectively transfer reasoning capabilities to non-reasoning models. However, models fine-tuned with this approach inherit the "overthinking" problem from teacher models, producing verbose and redundant reasoning chains during inference. To address this challenge, we propose Long-Short Chain-of-Thought Mixture Supervised Fine-Tuning (LS-Mixture SFT), which combines long CoT reasoning dataset with their short counterparts obtained through structure-preserved rewriting. Our experiments demonstrate that models trained using the LS-Mixture SFT method, compared to those trained with direct SFT, achieved an average accuracy improvement of 2.3% across various benchmarks while substantially reducing model response length by approximately 47.61%. This work offers an approach to endow non-reasoning models with reasoning capabilities through supervised fine-tuning while avoiding the inherent overthinking problems inherited from teacher models, thereby enabling efficient reasoning in the fine-tuned models.
Neural Operators (NOs) provide a powerful framework for computations involving physical laws that can be modelled by (integro-) partial differential equations (PDEs), directly learning maps between infinite-dimensional function spaces that bypass both the explicit equation identification and their subsequent numerical solving. Still, NOs have so far primarily been employed to explore the dynamical behavior as surrogates of brute-force temporal simulations/predictions. Their potential for systematic rigorous numerical system-level tasks, such as fixed-point, stability, and bifurcation analysis - crucial for predicting irreversible transitions in real-world phenomena - remains largely unexplored. Toward this aim, inspired by the Equation-Free multiscale framework, we propose and implement a framework that integrates (local) NOs with advanced iterative numerical methods in the Krylov subspace, so as to perform efficient system-level stability and bifurcation analysis of large-scale dynamical systems. Beyond fixed point, stability, and bifurcation analysis enabled by local in time NOs, we also demonstrate the usefulness of local in space as well as in space-time ("patch") NOs in accelerating the computer-aided analysis of spatiotemporal dynamics. We illustrate our framework via three nonlinear PDE benchmarks: the 1D Allen-Cahn equation, which undergoes multiple concatenated pitchfork bifurcations; the Liouville-Bratu-Gelfand PDE, which features a saddle-node tipping point; and the FitzHugh-Nagumo (FHN) model, consisting of two coupled PDEs that exhibit both Hopf and saddle-node bifurcations.
Sampling from flat modes in discrete spaces is a crucial yet underexplored problem. Flat modes represent robust solutions and have broad applications in combinatorial optimization and discrete generative modeling. However, existing sampling algorithms often overlook the mode volume and struggle to capture flat modes effectively. To address this limitation, we propose \emph{Entropic Discrete Langevin Proposal} (EDLP), which incorporates local entropy into the sampling process through a continuous auxiliary variable under a joint distribution. The local entropy term guides the discrete sampler toward flat modes with a small overhead. We provide non-asymptotic convergence guarantees for EDLP in locally log-concave discrete distributions. Empirically, our method consistently outperforms traditional approaches across tasks that require sampling from flat basins, including Bernoulli distribution, restricted Boltzmann machines, combinatorial optimization, and binary neural networks.
The Qwen series has emerged as a leading family of open-source Large Language Models (LLMs), demonstrating remarkable capabilities in natural language understanding tasks. With the recent release of Qwen3, which exhibits superior performance across diverse benchmarks, there is growing interest in deploying these models efficiently in resource-constrained environments. Low-bit quantization presents a promising solution, yet its impact on Qwen3's performance remains underexplored. This study conducts a systematic evaluation of Qwen3's robustness under various quantization settings, aiming to uncover both opportunities and challenges in compressing this state-of-the-art model. We rigorously assess 5 existing classic post-training quantization techniques applied to Qwen3, spanning bit-widths from 1 to 8 bits, and evaluate their effectiveness across multiple datasets. Our findings reveal that while Qwen3 maintains competitive performance at moderate bit-widths, it experiences notable degradation in linguistic tasks under ultra-low precision, underscoring the persistent hurdles in LLM compression. These results emphasize the need for further research to mitigate performance loss in extreme quantization scenarios. We anticipate that this empirical analysis will provide actionable insights for advancing quantization methods tailored to Qwen3 and future LLMs, ultimately enhancing their practicality without compromising accuracy. Our project is released on https://github.com/Efficient-ML/Qwen3-Quantization and https://huggingface.co/collections/Efficient-ML/qwen3-quantization-68164450decb1c868788cb2b.
This paper investigates $\sim_{\mathcal{L}_3}$-identifiability, a form of complete counterfactual identifiability within the Pearl Causal Hierarchy (PCH) framework, ensuring that all Structural Causal Models (SCMs) satisfying the given assumptions provide consistent answers to all causal questions. To simplify this problem, we introduce exogenous isomorphism and propose $\sim_{\mathrm{EI}}$-identifiability, reflecting the strength of model identifiability required for $\sim_{\mathcal{L}_3}$-identifiability. We explore sufficient assumptions for achieving $\sim_{\mathrm{EI}}$-identifiability in two special classes of SCMs: Bijective SCMs (BSCMs), based on counterfactual transport, and Triangular Monotonic SCMs (TM-SCMs), which extend $\sim_{\mathcal{L}_2}$-identifiability. Our results unify and generalize existing theories, providing theoretical guarantees for practical applications. Finally, we leverage neural TM-SCMs to address the consistency problem in counterfactual reasoning, with experiments validating both the effectiveness of our method and the correctness of the theory.
Graph Edit Distance (GED) is a widely used metric for measuring similarity between two graphs. Computing the optimal GED is NP-hard, leading to the development of various neural and non-neural heuristics. While neural methods have achieved improved approximation quality compared to non-neural approaches, they face significant challenges: (1) They require large amounts of ground truth data, which is itself NP-hard to compute. (2) They operate as black boxes, offering limited interpretability. (3) They lack cross-domain generalization, necessitating expensive retraining for each new dataset. We address these limitations with GRAIL, introducing a paradigm shift in this domain. Instead of training a neural model to predict GED, GRAIL employs a novel combination of large language models (LLMs) and automated prompt tuning to generate a program that is used to compute GED. This shift from predicting GED to generating programs imparts various advantages, including end-to-end interpretability and an autonomous self-evolutionary learning mechanism without ground-truth supervision. Extensive experiments on seven datasets confirm that GRAIL not only surpasses state-of-the-art GED approximation methods in prediction quality but also achieves robust cross-domain generalization across diverse graph distributions.
Solving non-convex resource allocation problems poses significant challenges in wireless communication systems, often beyond the capability of traditional optimization techniques. To address this issue, we propose LLM-OptiRA, the first framework that leverages large language models (LLMs) to automatically detect and transform non-convex components into solvable forms, enabling fully automated resolution of non-convex resource allocation problems in wireless communication systems. LLM-OptiRA not only simplifies problem-solving by reducing reliance on expert knowledge, but also integrates error correction and feasibility validation mechanisms to ensure robustness. Experimental results show that LLM-OptiRA achieves an execution rate of 96% and a success rate of 80% on GPT-4, significantly outperforming baseline approaches in complex optimization tasks across diverse scenarios.
Nanobodies, single-domain antibody fragments derived from camelid heavy-chain-only antibodies, exhibit unique advantages such as compact size, high stability, and strong binding affinity, making them valuable tools in therapeutics and diagnostics. While recent advances in pretrained protein and antibody language models (PPLMs and PALMs) have greatly enhanced biomolecular understanding, nanobody-specific modeling remains underexplored and lacks a unified benchmark. To address this gap, we introduce NbBench, the first comprehensive benchmark suite for nanobody representation learning. Spanning eight biologically meaningful tasks across nine curated datasets, NbBench encompasses structure annotation, binding prediction, and developability assessment. We systematically evaluate eleven representative models--including general-purpose protein LMs, antibody-specific LMs, and nanobody-specific LMs--in a frozen setting. Our analysis reveals that antibody language models excel in antigen-related tasks, while performance on regression tasks such as thermostability and affinity remains challenging across all models. Notably, no single model consistently outperforms others across all tasks. By standardizing datasets, task definitions, and evaluation protocols, NbBench offers a reproducible foundation for assessing and advancing nanobody modeling.
Recent progress in Meta-Black-Box-Optimization (MetaBBO) has demonstrated that using RL to learn a meta-level policy for dynamic algorithm configuration (DAC) over an optimization task distribution could significantly enhance the performance of the low-level BBO algorithm. However, the online learning paradigms in existing works makes the efficiency of MetaBBO problematic. To address this, we propose an offline learning-based MetaBBO framework in this paper, termed Q-Mamba, to attain both effectiveness and efficiency in MetaBBO. Specifically, we first transform DAC task into long-sequence decision process. This allows us further introduce an effective Q-function decomposition mechanism to reduce the learning difficulty within the intricate algorithm configuration space. Under this setting, we propose three novel designs to meta-learn DAC policy from offline data: we first propose a novel collection strategy for constructing offline DAC experiences dataset with balanced exploration and exploitation. We then establish a decomposition-based Q-loss that incorporates conservative Q-learning to promote stable offline learning from the offline dataset. To further improve the offline learning efficiency, we equip our work with a Mamba architecture which helps long-sequence learning effectiveness and efficiency by selective state model and hardware-aware parallel scan respectively. Through extensive benchmarking, we observe that Q-Mamba achieves competitive or even superior performance to prior online/offline baselines, while significantly improving the training efficiency of existing online baselines. We provide sourcecodes of Q-Mamba at https://github.com/MetaEvo/Q-Mamba.
Large language models (LLMs) have become integral to various real-world applications, leveraging massive, web-sourced datasets like Common Crawl, C4, and FineWeb for pretraining. While these datasets provide linguistic data essential for high-quality natural language generation, they often contain harmful content, such as hate speech, misinformation, and biased narratives. Training LLMs on such unfiltered data risks perpetuating toxic behaviors, spreading misinformation, and amplifying societal biases which can undermine trust in LLM-driven applications and raise ethical concerns about their use. This paper presents a large-scale analysis of inappropriate content across these datasets, offering a comprehensive taxonomy that categorizes harmful webpages into Topical and Toxic based on their intent. We also introduce a prompt evaluation dataset, a high-accuracy Topical and Toxic Prompt (TTP), and a transformer-based model (HarmFormer) for harmful content filtering. Additionally, we create a new multi-harm open-ended toxicity benchmark (HAVOC) and provide crucial insights into how models respond to adversarial toxic inputs. We share TTP, TTP-Eval, HAVOC and a sample of C4 inferenced on HarmFormer. Our work offers insights into ensuring safer LLM pretraining and serves as a resource for Responsible AI (RAI) compliance.
Individual treatment effect estimation has gained significant attention in recent data science literature. This work introduces the Double Neural Network (Double-NN) method to address this problem within the framework of extended fiducial inference (EFI). In the proposed method, deep neural networks are used to model the treatment and control effect functions, while an additional neural network is employed to estimate their parameters. The universal approximation capability of deep neural networks ensures the broad applicability of this method. Numerical results highlight the superior performance of the proposed Double-NN method compared to the conformal quantile regression (CQR) method in individual treatment effect estimation. From the perspective of statistical inference, this work advances the theory and methodology for statistical inference of large models. Specifically, it is theoretically proven that the proposed method permits the model size to increase with the sample size $n$ at a rate of $O(n^{\zeta})$ for some $0 \leq \zeta<1$, while still maintaining proper quantification of uncertainty in the model parameters. This result marks a significant improvement compared to the range $0\leq \zeta < \frac{1}{2}$ required by the classical central limit theorem. Furthermore, this work provides a rigorous framework for quantifying the uncertainty of deep neural networks under the neural scaling law, representing a substantial contribution to the statistical understanding of large-scale neural network models.