arXiv subjects

Weiyi Sun

Publications and source records attributed to Weiyi Sun.

At least 19 recordsLinked to original sources

arXiv2026

Prefix Executability: Evaluating Tool-Using Agents Beyond Final Success

Tool-use benchmarks often score agents by final task success, but terminal outcomes can obscure whether the intermediate tool-call trajectory was executable. We propose prefix executability as a trajectory-level evaluation protocol for tool-using agents. The protocol measures whether every prefix of a predicted tool-call sequence remains executable under tool schemas, state preconditions, and data dependencies, and summarizes reliability using Prefix Executability Curves (PEC), First-Error Depth (FED), Trajectory Executability Rate (TER), horizon-normalized AUC-PEC, and Recovery Rate (RR). We instantiate the protocol on stratified samples from BFCL-v3 multi-turn and ToolSandbox, and on the complete Airline/Retail base splits of $τ^3$-bench. We observe that outcome-only and prefix-level rankings diverge: agents whose execution is conditioned on an unverified plan often achieve higher point-estimate final success than non-plan-conditioned baselines while substantially lowering TER and AUC-PEC. These divergences are precisely the failure modes that prefix executability is designed to expose: they reveal whether apparent success rests on an executable trajectory, where invalidity first enters, and whether success follows recovery from an earlier failure. As a validation intervention, explicit pre-execution plan checking can improve executable-prefix survival while maintaining competitive terminal success. These results show that final success alone is insufficient for evaluating tool-use agents and that prefix-level executability provides an actionable reliability signal for comparing agents, diagnosing failures, and assessing validation mechanisms.

cs.AI↗

arXiv2026

Ask Less, Ask Better: Allocating User Attention in Proactive LLM Agents with Interruption Budgets

LLM agents can ask clarifying questions to complete tasks, but answering each question demands users' time and effort. Existing evaluations typically measure this burden by counting questions, treating necessary and avoidable questions equally. We propose an evaluation protocol that distinguishes how often agents ask from how well they ask. An interruption budget limits the number of questions, while a user-attention cost charges extra for questions that are hard to read or not worth asking, as rated by a human-validated LLM judge. Beyond question counts, this cost captures perceived effort: on conversation pairs with the same number of questions, human annotators agreed with the cost on which conversation required more effort 75% of the time. Across eight LLMs and four customer-service domains, limiting agents to one to three questions reduces user-attention cost by 36-73% while keeping 86-94% of the task reward they earn with unlimited questions. At the same budget, reviewing each question against the memory of users' earlier answers cuts attention cost by another 13% with little reward loss and yields more valuable questions. Models also differ sharply in how well they ask: at the same user-attention cost, the most efficient models gain more than twice as much reward from their questions as the others. These results show that asking less and asking better are distinct goals, and that evaluations of proactive agents should measure both.

cs.CL↗

arXiv2026

Even cycle decomposition thresholds for dense multipartite graphs

Let $r\geq 2$ be an integer. An $r$-partite graph $Γ$ with vertex partition $V_1,\ldots,V_r$ is $2$-balanced if there exists a positive integer $n$ such that $n\leq |V_i|\leq 2n$ for every $i\in[r]$. For an integer $\ell\geq3$, let $C_\ell$ denote the cycle of length $\ell$, and define $\hatδ(Γ)=\min\{d_Γ(v,V_i)/|V_i|:i\in[r],\ v\in V(Γ)\setminus V_i\}$. Let $\hatδ^r_{C_\ell}$ denote the $C_\ell$-decomposition threshold for $2$-balanced $r$-partite graphs, that is, the least nonnegative real number $δ$ such that, for every $\varepsilon>0$, there exists $n_0$ such that every $C_\ell$-divisible $2$-balanced $r$-partite graph $G$ with $\min_i|V_i|>n_0$ and $\hatδ(G)\geqδ+\varepsilon$ admits a $C_\ell$-decomposition. We prove that $\hatδ^{r}_{C_4}=\frac{2}{3}$ and $\hatδ^{r}_{C_{2k}}=\frac{1}{2}$ for every $r\geq2$ and every $k\geq3$.

math.CO↗

arXiv2026

MXAttention: Data-Free Optimal Scaling and Pre-Normalization Quantization for MXFP4 Attention

The quadratic cost of attention is a major bottleneck in diffusion-based video generation models. MXFP4 attention provides a promising path toward efficient inference, but direct MXFP4 quantization often degrades generation quality due to two numerical issues: the clipping-underflow trade-off from power-of-two scaling and the row-wise normalization error introduced in the softmax loop. We propose MXAttention, a data-free post-training quantization framework for MXFP4 attention. MXAttention introduces two components: Universal Optimal Scaling (UOS), which exploits the periodic structure of power-of-two microscaling to derive a distribution-independent optimal scaling boundary Qmax=7.25 without calibration or search, and Pre-Normalization Quantization (PNQ), which quantizes unnormalized softmax exponentials before row-wise summation to preserve normalization by construction. Experiments on Wan2.2 and HunyuanVideo show that MXAttention closes at least 95% of the VBench Imaging Quality gap between OCP MXFP4 and FP16, substantially improves frame-level similarity, and preserves FP16-level generation quality with less than 0.01 absolute degradation on all reported VBench metrics. MXAttention also achieves performance competitive with strong NVFP4-based baselines with negligible overhead when fused into the attention pipeline. The implementation is publicly available in MindIE-SD.

cs.LG↗

arXiv2026

MT-OSC: Path for LLMs that Get Lost in Multi-Turn Conversation

Large language models (LLMs) suffer significant performance degradation when user instructions and context are distributed over multiple conversational turns, yet multi-turn (MT) interactions dominate chat interfaces. The routine approach of appending full chat history to prompts rapidly exhausts context windows, leading to increased latency, higher computational costs, and diminishing returns as conversations extend. We introduce MT-OSC, a One-off Sequential Condensation framework that efficiently and automatically condenses chat history in the background without disrupting the user experience. MT-OSC employs a Condenser Agent that uses a few-shot inference-based Condenser and a lightweight Decider to selectively retain essential information, reducing token counts by up to 72% in 10-turn dialogues. Evaluated across 13 state-of-the-art LLMs and diverse multi-turn benchmarks, MT-OSC consistently narrows the multi-turn performance gap - yielding improved or preserved accuracy across datasets while remaining robust to distractors and irrelevant turns. Our results establish MT-OSC as a scalable solution for multi-turn chats, enabling richer context within constrained input spaces, reducing latency and operational cost, while balancing performance.

cs.CL↗

arXiv2026

GSM-SEM: Benchmark and Framework for Generating Semantically Variant Augmentations

Benchmarks like GSM8K are popular measures of mathematical reasoning, but leaderboard gains can overstate true capability due to memorization of fixed test sets. Most robustness variants apply surface-level perturbations (paraphrases, renamings, number swaps, distractors) that largely preserve the underlying facts, and static releases can themselves become memorization targets over time. We introduce GSM-SEM, a reusable and stochastic framework for generating semantically diverse benchmark variants with substantially higher semantic variance than prior approaches. GSM-SEM perturbs problem statements by modifying entities, attributes, and/or relationships, frequently altering underlying facts and requiring models to recompute solutions under new conditions, while constraining generation to preserve the original calculations/answer and approximate problem difficulty. GSM-SEM generates fresh variants on each run without requiring re-annotation, reducing reliance on static public benchmarks for evaluation and thereby lowering the bias of memorization. We apply GSM-SEM on GSM8K and two existing variation suites (GSM-Symbolic and GSM-Plus), producing GSM8K-SEM, GSM-Symbolic-SEM, and GSM-Plus-SEM. Evaluating 14 SOTA LLMs, we observe consistent performance drops with larger decline when semantic perturbations are coupled with symbolic/plus variations (average drop rate 28% in maximum strictness configuration of GSM-SEM). We publicly release the three SEM variants as fully human-validated datasets. Finally, to demonstrate applicability beyond GSM-style math problems, we apply GSM-SEM to additional benchmarks including BigBenchHard, LogicBench, and NLR-BIRD.

cs.CL↗

arXiv2026

Temperature Dependence of Gain and Time Resolution in LGAD Detectors

Low-Gain Avalanche Diodes (LGADs) provide moderate internal gain and time resolutions of a few tens of picoseconds, making them a key technology for ultrafast timing in high-energy physics and beyond. However, both their gain and timing characteristics vary strongly with reverse-bias voltage and temperature. This work establishes a compact analytical framework that describes multi-temperature LGAD gain and timing behavior through an equivalent representation of the gain layer. The non-uniform multiplication region is replaced by an equivalent rectangular gain layer, from which a first-order bias-compensation relation for constant gain is derived and validated. Using multi-temperature measurements of LGADs designed by IHEP and fabricated by IME, together with an independent HPK dataset, we show that the gain-voltage curve family can be reconstructed from a reference-temperature main curve, substantially reducing characterization effort. The same idea is then extended to timing by decomposing the total time resolution into jitter and intrinsic components and representing their temperature dependences as component-wise equivalent bias offsets. The resulting framework provides a function-level description of multi-temperature LGAD time-resolution curves and offers a practical tool for calibration, operation, and reduced-density characterization of LGAD-based ultrafast timing systems.

physics.ins-det↗

arXiv2026

JTPRO: A Joint Tool-Prompt Reflective Optimization Framework for Language Agents

Large language model (LLM) agents augmented with external tools often struggle as number of tools grow large and become domain-specific. In such settings, ambiguous tool descriptions and under-specified agent instructions frequently lead to tool mis-selection and incorrect slot/value instantiation. We hypothesize that this is due to two root causes: generic, one-size-fits-all prompts that ignore tool-specific nuances, and underspecified tool schemas that lack clear guidance on when and how to use each tool and how to format its parameters. We introduce Joint Tool-Prompt Reflective Optimization (JTPRO), a framework for improving tool-calling reliability in trace-supervised settings by iteratively using rollout-driven reflection to co-optimize global instructions and per-tool schema/argument descriptions for accurate tool selection and argument instantiation in large tool inventories. JTPRO is designed to preserve only tool-local cues needed for correct disambiguation and slot filling. We evaluate JTPRO across multi-tool benchmarks, which account for different number of tools using three metrics: Tool Selection Accuracy (TSA), Slot Filling Accuracy(SFA), and Overall Success Rate(OSR) (correct tool + correct slots + correct values). JTPRO consistently outperforms strong baselines, including CoT-style agents, and reflective prompt optimizers such as GEPA by 5%-20% (relative) on OSR. Ablations show that joint optimization of instructions and tool schemas is more effective and robust than optimizing either component in isolation.

cs.AI↗

arXiv2026

DiffuMask: Diffusion Language Model for Token-level Prompt Pruning

In-Context Learning and Chain-of-Thought prompting improve reasoning in large language models (LLMs). These typically come at the cost of longer, more expensive prompts that may contain redundant information. Prompt compression based on pruning offers a practical solution, yet existing methods rely on sequential token removal which is computationally intensive. We present DiffuMask, a diffusion-based framework integrating hierarchical shot-level and token-level pruning signals, that enables rapid and parallel prompt pruning via iterative mask prediction. DiffuMask substantially accelerates the compression process via masking multiple tokens in each denoising step. It offers tunable control over retained content, preserving essential reasoning context and achieving up to 80\% prompt length reduction. Meanwhile, it maintains or improves accuracy across in-domain, out-of-domain, and cross-model settings. Our results show that DiffuMask provides a generalizable and controllable framework for prompt compression, facilitating faster and more reliable in-context reasoning in LLMs.

cs.CL↗

arXiv2026

Barriers to Discrete Reasoning with Transformers: A Survey Across Depth, Exactness, and Bandwidth

Transformers have become the foundational architecture for a broad spectrum of sequence modeling applications, underpinning state-of-the-art systems in natural language processing, vision, and beyond. However, their theoretical limitations in discrete reasoning tasks, such as arithmetic, logical inference, and algorithmic composition, remain a critical open problem. In this survey, we synthesize recent studies from three theoretical perspectives: circuit complexity, approximation theory, and communication complexity, to clarify the structural and computational barriers that transformers face when performing symbolic computations. By connecting these established theoretical frameworks, we provide an accessible and unified account of why current transformer architectures struggle to implement exact discrete algorithms, even as they excel at pattern matching and interpolation. We review key definitions, seminal results, and illustrative examples, highlighting challenges such as depth constraints, difficulty approximating discontinuities, and bottlenecks in inter-token communication. Finally, we discuss implications for model design and suggest promising directions for overcoming these foundational limitations.

cs.CL↗

arXiv2025

Open Agent Specification (Agent Spec): A Unified Representation for AI Agents

The proliferation of agent frameworks has led to fragmentation in how agents are defined, executed, and evaluated. Existing systems differ in their abstractions, data flow semantics, and tool integrations, making it difficult to share or reproduce workflows. We introduce Open Agent Specification (Agent Spec), a declarative language that defines AI agents and agentic workflows in a way that is compatible across frameworks, promoting reusability, portability and interoperability of AI agents. Agent Spec defines a common set of components, control and data flow semantics, and schemas that allow an agent to be defined once and executed across different runtimes. Agent Spec also introduces a standardized Evaluation harness to assess agent behavior and agentic workflows across runtimes - analogous to how HELM and related harnesses standardized LLM evaluation - so that performance, robustness, and efficiency can be compared consistently across frameworks. We demonstrate this using four distinct runtimes (LangGraph, CrewAI, AutoGen, and WayFlow) evaluated over three different benchmarks (SimpleQA Verified, $τ^2$-Bench and BIRD-SQL). We provide accompanying toolsets: a Python SDK (PyAgentSpec), a reference runtime (WayFlow), and adapters for popular frameworks (e.g., LangGraph, AutoGen, CrewAI). Agent Spec bridges the gap between model-centric and agent-centric standardization & evaluation, laying the groundwork for reliable, reusable, and portable agentic systems.

cs.AI↗

arXiv2025

Can LLMs Narrate Tabular Data? An Evaluation Framework for Natural Language Representations of Text-to-SQL System Outputs

In modern industry systems like multi-turn chat agents, Text-to-SQL technology bridges natural language (NL) questions and database (DB) querying. The conversion of tabular DB results into NL representations (NLRs) enables the chat-based interaction. Currently, NLR generation is typically handled by large language models (LLMs), but information loss or errors in presenting tabular results in NL remains largely unexplored. This paper introduces a novel evaluation method - Combo-Eval - for judgment of LLM-generated NLRs that combines the benefits of multiple existing methods, optimizing evaluation fidelity and achieving a significant reduction in LLM calls by 25-61%. Accompanying our method is NLR-BIRD, the first dedicated dataset for NLR benchmarking. Through human evaluations, we demonstrate the superior alignment of Combo-Eval with human judgments, applicable across scenarios with and without ground truth references.

cs.CL↗

arXiv2025

Flexible delivery of broadband, 100-fs mid-infrared pulses in the water-absorption band using hollow-core photonic crystal fibre

High quality free-space and over-fibre transmission of mid-IR light is limited by factors such as material-related absorption, diffraction, light leakage and nonlinearity. Conventional vacuum apparatus can be utilized for high-quality laser-beam delivery to address these issues, the deployment of such apparatus would, however, increase the system complexity, being detrimental to their practical applications. Here we report the successful use of evacuated hollow-core photonic crystal fibre (PCF) to flexibly transmit ultrafast mid-IR pulses over several meters, while preserving exceptional spatial, spectral and temporal fidelity. The PCF was engineered to feature a low-loss transmission band within the water absorption range, and an evacuated 5-m length was used to transmit Watt-level, 100 fs pulses centred at around 2.8 microns. A comparison between free-space transmission and air-filled PCF highlights the superior performance of the evacuated hollow-core PCF, indicating its strong suitability for the flexible delivery of sub-ps laser pulses in the mid-IR.

physics.optics↗

arXiv2025

Three-stage dynamics of nonlinear pulse amplification in ultrafast mid-infrared fiber amplifier with anomalous dispersion

Nonlinear pulse amplification in optical fiber, with capability of breaking the gain-bandwidth limitation, is a key technique for high-energy, ultrafast pulse generation. In the longer wavelength region (including 1.55 μm, 2 μm and 2.8 μm) where the gain fiber has normally strong anomalous dispersion, the nonlinear amplification process over fiber exhibits more complicated dynamics than that of its 1-μm counterpart, and the underlying mechanism of the nonlinear pulse propagation process in high-gain anomalous fiber is still elusive so far. Here, we demonstrate an in-depth study on the nonlinear amplification process in high-gain ultrafast mid-infrared fiber, providing clear physical understanding on the debate of adiabatic soliton compression. We unveil that under the high-gain condition, the ultrafast pulse launched into the anomalous gain fiber experiences successively three distinct stages, named as the balance between linear and nonlinear chirp, high-order-soliton-like pulse compression and pulse splitting due to high-order effects. While a relatively-clean ultrafast pulse can be obtained immediately after the high-order-soliton-like compression stage, excessive gain fiber length could hardly enhance further the pulse peak power due to soliton splitting. Our findings can provide several critical guidelines for designing high-power ultrafast fiber amplifiers at near- and mid-infrared wavelengths.

physics.optics↗

arXiv2024

The Performance of AC-coupled Strip LGAD developed by IHEP

The AC-coupled Strip LGAD (Strip AC-LGAD) is a novel LGAD design that diminishes the density of readout electronics through the use of strip electrodes, enabling the simultaneous measurement of time and spatial information. The Institute of High Energy Physics has designed a long Strip AC-LGAD prototype with a strip electrode length of 5.7 mm and pitches of 150 $μm$, 200 $μm$, and 250 $μm$. Spatial and timing resolutions of the long Strip AC-LGAD are studied by pico-second laser test and beta source tests. The laser test demonstrates that spatial resolution improves as the pitch size decreases, with an optimal resolution achieved at 8.3 $μ$m. Furthermore, the Beta source test yields a timing resolution of 37.6 ps.

physics.ins-det↗

arXiv2024

Development of the strip LGAD detector with double-end readout for future colliders

The Low-Gain Avalanche Diode (LGAD) is a new silicon detector and holds wide application prospects in particle physics experiments due to its excellent timing resolution. The LGAD with a pixel size of 1.3 mm $\times$ 1.3 mm was used to construct a High Granularity Timing Detector (HGTD) in ATLAS experiments to solve the pile-up problem. Meanwhile, the Circular Electron Positron Collider (CEPC) also proposes detectors using the LGAD. However, pixel LGAD exhibits higher readout electronics density and cost, which somewhat limits the application of LGADs. To decrease the readout electronics density, the Institute of High Energy Physics (IHEP) of the Chinese Academy of Sciences has designed strip LGADs with larger areas. These strip LGADs are all 19 mm in length but with different widths of 1.0 mm, 0.5 mm, and 0.3 mm. This article provides a detailed introduction to the design parameters of these strip LGADs and tests their electrical characteristics, including leakage current, break-down voltage, depletion capacitance, etc. The timing resolution and signal-to-noise ratio of the three strip LGAD sensors were investigated using a beta source test system. The position resolution parallel to the strip direction was tested and analyzed for the first time using a pico-second laser test system. Tests have demonstrated that the timing resolution of strip LGADs can reach about 37.5 ps, and position resolution parallel to the strip direction is better than 1 mm.

physics.ins-det↗

arXiv2023

Performance of AC-LGAD strip sensor designed for the DarkSHINE experiment

AC-coupled Low Gain Avalanche Detector (AC-LGAD) is a new precise detector technology developed in recent years. Based on the standard Low Gain Avalanche Detector (LGAD) technology, AC-LGAD sensors can provide excellent timing performance and spatial resolution. This paper presents the design and performance of several prototype AC-LGAD strip sensors for the DarkSHINE tracking system, as well as the electrical characteristics and spatial resolution of the prototype sensors from two batches of wafers with different $n^+$ dose.The range of spatial resolutions of 6.5$\mathrm{μm}$ $\sim$ 8.2$\mathrm{μm}$ and 8.8$\mathrm{μm}$ $\sim$ 12.3$\mathrm{μm}$ are achieved by the AC-LGAD sensors with 100$μm$ pitch size.

physics.ins-det↗

arXiv2023

The performance of large-pitch AC-LGAD with different N+ dose

AC-Coupled LGAD (AC-LGAD) is a new 4D detector developed based on the Low Gain Avalanche Diode (LGAD) technology, which can accurately measure the time and spatial information of particles. The Institute of High Energy Physics (IHEP) designed a large-size AC-LGAD with a pitch of 2000~\SI{}{\micro\metre} and AC pad of 1000~\SI{}{\micro\metre}, and explored the effect of N+ layer dose on the spatial resolution and time resolution. The spatial resolution varied from 36~\SI{}{\micro\metre} to 16~\SI{}{\micro\metre} depending on N+ dose for a charge corresponding to about 12 minimum ionizing particles. The jitter component of the time resolution does not change significantly with different N+ doses, and it is about 15-17 ps measured by laser. The AC-LGAD with a low N+ dose has a large attenuation factor and better spatial resolution in the central region between pads. In these specific conditions, large signal attenuation factor and low noise level are beneficial to improve the spatial resolution of the AC-LGAD sensor.

physics.ins-det↗