arXiv subjects

Ziyi Xu

Publications and source records attributed to Ziyi Xu.

At least 19 recordsLinked to original sources

arXiv2026

HiSparse: Scaling Sparse-Attention Decoding with Hierarchical KV Cache Management

Top-k sparse attention makes long-context LLM decoding cheap to compute: each step reads only a few thousand selected KV entries rather than the full context. Serving systems, however, typically keep the entire KV cache in GPU HBM so that every position stays selectable, so a request's memory bill still grows with its full context length--decoding hits a capacity wall long before it runs out of compute, and a context whose KV cache exceeds HBM cannot be served at all. We present HiSparse, an exact, indexer-agnostic hierarchical KV cache for sparse-attention serving. HiSparse keeps each request's full KV history in host memory and bounds its decode footprint with a small, fixed-size GPU cache; a fused CUDA kernel resolves each layer's selections--hit detection, LRU replacement, and host-to-device fetches--inside the decode CUDA graph; and, for models that share selections across layers, exact layer-wise prefetching hides roughly half of the remaining miss overhead. Because only KV placement changes, model outputs are unchanged. HiSparse is merged into upstream SGLang and evaluated across three sparse-attention families (DSA, NSA, and Quest) on H200, B200, and GH200 platforms: it improves peak generation throughput by up to 4.7x on long-context workloads while preserving comparable per-token latency and reducing time-to-first-token at high load--and a no-IO oracle shows the resolution mechanism itself adds no measurable per-token cost, leaving host-device IO as the only price of bounded residency.

cs.DC↗

arXiv2026

Timing and Spectral Analysis of the 2024 Outburst of 2S 1553$-$542 with NuSTAR and NICER

We report a timing and spectral study of the 2024 outburst of the Be/X-ray binary pulsar 2S~1553$-$542 using \textit{NuSTAR} and \textit{NICER} observations. From the \textit{NuSTAR} light curve we measure a pulse period of $9.285022\pm0.000001$~s. The energy-resolved pulse profiles are dominated by a single peak and show a wing-like structure most clearly in the $12$--$22$~keV band. The pulsed fraction remains above 60\% and increases with energy. The phase-averaged \textit{NuSTAR} spectrum is described by an absorbed blackbody plus cutoff power-law continuum, together with an iron emission line and a cyclotron absorption feature. Using the \texttt{cyclabs} model, we obtain a cyclotron energy of $E_{\rm cyc}\simeq24.1$~keV, corresponding to a magnetic field strength of $B\sim3\times10^{12}$~G. Phase-resolved spectroscopy shows that the continuum and cyclotron-line parameters vary with pulse phase, and that the line becomes poorly constrained around the pulse-wing phase. We also searched the short \textit{NICER} GTIs for transient mHz variability using wavelet analysis and a CEEMDAN-based Hilbert--Huang transform. Localized excesses near $\sim10$~mHz and $\sim20$~mHz are found, but the short exposures, COI effects, red-noise fluctuations, and the lack of a well-constrained Fourier peak limit their significance. We therefore treat them as candidate mHz variability rather than firm mHz QPO detections.

astro-ph.HE↗

arXiv2026

DynaFlow: Transparent and Flexible Intra-Device Parallelism via Programmable Operator Scheduling

Intra-device parallelism addresses resource under-utilization in ML inference and training by overlapping the execution of operators with different resource usage. However, its wide adoption is hindered by a fundamental conflict with the static, sequential programming model of existing frameworks. Integrating these strategies requires invasive, model-specific code overhauls, representing an intractable engineering cost. This is further amplified by the high sensitivity of strategies to execution contexts (e.g., workload, model architecture, hardware), forcing developers to implement and maintain multiple specialized solutions. To address this, we propose DynaFlow, a framework that enables the transparent and flexible integration of intra-device parallelism by decoupling the logical model definition from the physical execution schedule. DynaFlow introduces a flexible frontend with annotations for graph partitioning and a programmable interface for defining custom intra-device parallelism strategies. Its efficient backend manages complex control/data-flow asynchronously, uses custom memory management to eliminate copy overheads, and preserves compatibility with optimizations like CUDA Graphs and TorchInductor. We demonstrate that DynaFlow can integrate representative parallelism strategies into 6 state-of-the-art ML systems with minimal code changes, achieving up to a 1.29x throughput improvement. DynaFlow is publicly available at https://github.com/uw-syfi/DynaFlow.

cs.DC↗

arXiv2026

Ergodic Imitation for Adaptive Exploration around Demonstrations

In robotics, a common challenge in imitation learning is the mismatch between training and deployment conditions, caused, for example, by environmental changes or imperfect observation and control. When a robot follows a nominal trajectory under such mismatch, it may become stuck and fail to complete the task. This calls for adaptive online exploration strategies that remain grounded in demonstrations. To this end, we propose an adaptive ergodic imitation approach that constructs a target distribution from the geometry of the retrieved demonstrations and uses it to generate trajectories that adaptively interpolate between tracking and exploration. Our method extends ergodic control beyond its traditional role in area-coverage and search by incorporating demonstrations into a retrieval-based receding-horizon framework for adaptive imitation.

cs.RO↗

arXiv2026

Pythia: Exploiting Workflow Predictability for Efficient Agent-Native LLM Serving

As LLM applications grow more complex, developers are increasingly adopting multi-agent architectures to decompose workflows into specialized, collaborative components, introducing structure that constrains agent behavior and exposes useful semantic predictability. Unlike traditional LLM serving, which operates under highly dynamic and uncertain conditions, this structured topology enables opportunities to reduce runtime uncertainty$\unicode{x2015}$yet existing systems fail to exploit it, treating agentic workloads as generic traffic and incurring significant inefficiencies. Our analysis of production traces from an agent-serving platform and an internal coding assistant reveals key bottlenecks, including low prefix cache hit rates, severe resource contention from long-context requests, and substantial queuing delays due to suboptimal scaling. To address these challenges, we propose Pythia, a multi-agent serving system that captures workflow semantics through a simple interface at the serving layer, unlocking new optimization opportunities and substantially improving throughput and job completion time over state-of-the-art baselines.

cs.MA↗

arXiv2026

Thermal Casimir Force Imaging of Nonequilibrium Hot Electrons

The thermal Casimir effect, arising from fluctuating electromagnetic fields of thermally agitated charges, induces thermosensitive forces and presents a novel approach to detecting nanoscale hot electrons, elusive yet ubiquitous in modern miniaturized transistors. However, detecting thermal Casimir forces at the nanoscale remains extremely challenging due to background forces such as electrostatic force and quantum Casimir force. In this study, we present the first non-contact force measurement of hot electrons based on the thermal Casimir effect. Using an atomic force microscope (AFM) with a dual-resonant tip, we achieve thermosensitive force detection of nonequilibrium hot electrons while effectively suppressing background thermo-insensitive forces, thereby distinguishing them from cold electrons. In silicon nanoconstriction devices, the measured thermal Casimir pressure reaches approximately 3 bar at a separation of 5 nm at an electron temperature of about 10^3 K. Our work introduces a novel methodology for hot electron nanothermometry and provides critical insights into the thermo-mechanical properties of post-Moore nanoelectronics.

cond-mat.mes-hall↗

arXiv2026

Towards High-Goodput LLM Serving with Prefill-decode Multiplexing

Large Language Model (LLM) serving must meet stringent Service Level Objectives (SLOs) for both the prefill and decode phases. Some existing solutions disaggregate the two phases, causing potential resource idleness or compute redundancy. Others split the prefill phase into chunks and fuse it with decode iteration, creating a dilemma between SLO compliance and high utilization. To address these issues, an efficient serving system should dynamically adapt compute allocation, decouple compute from memory management, and execute prefill and decode independently. We present MuxWise, an LLM serving framework that adopts a new paradigm, intra-GPU prefill-decode multiplexing, to meet these requirements. To fully exploit the paradigm, MuxWise integrates a bubble-less multiplex engine, a contention-tolerant estimator, and an SLO-aware dispatcher. Evaluation shows that MuxWise improves peak throughput under SLO guarantees by an average of 2.20x (up to 3.06x) over state-of-the-art baselines.

cs.OS↗

arXiv2026

XGrammar-2: Dynamic and Efficient Structured Generation Engine for Agentic LLMs

Modern LLM agents increasingly rely on dynamic structured generation, such as tool calling and response protocols. Unlike traditional structured generation with static structures, these workloads vary both across requests and within a request, posing new challenges to existing engines. We present XGrammar-2, a structured generation engine for dynamic agentic workloads. Our design is based on two key ideas: first-class support for tag-triggered structure switching, and fine-grained reuse across requests with different output structures. Concretely, XGrammar-2 introduces TagDispatch for dynamic structural dispatching and Cross-Grammar Cache for substructure-level cache reuse across grammars. It further improves efficiency with an Earley-based adaptive token mask cache, just-in-time compilation, and repetition state compression. Experiments show that XGrammar-2 achieves over 6x faster compilation than prior structured generation engines, and incurs near-zero end-to-end overhead in modern LLM serving systems.

cs.AI↗

arXiv2025

FailSafe: High-performance Resilient Serving

Tensor parallelism (TP) enables large language models (LLMs) to scale inference efficiently across multiple GPUs, but its tight coupling makes systems fragile: a single GPU failure can halt execution, trigger costly KVCache recomputation, and introduce long-term compute and memory imbalance. We present FailSafe, a fault-tolerant TP serving system that sustains high performance under irregular GPU availability. FailSafe introduces three techniques to balance computation and memory across GPUs: (1) Cyclic KVCache Placement for uniform memory utilization, (2) Hybrid Attention combining tensor- and data-parallel attention to eliminate stragglers, and (3) Fine-Grained Load-Aware Routing to dynamically balance requests. It further employs proactive KVCache backup and on-demand weight recovery to avoid expensive recomputation and redundant data transfers. We implement these techniques in a lightweight serving engine compatible with existing LLM infrastructures. Evaluated on an 8xH100 DGX system with real-world fault traces and representative workloads, FailSafe achieves up to 2x higher throughput and two orders of magnitude lower recovery latency compared to standard fault handling approaches. Even with up to three GPU failures, FailSafe sustains high throughput and balanced utilization, demonstrating robust and efficient LLM serving under dynamic and unreliable hardware conditions.

cs.DC↗

arXiv2025

LiDAR-VGGT: Cross-Modal Coarse-to-Fine Fusion for Globally Consistent and Metric-Scale Dense Mapping

Reconstructing large-scale colored point clouds is an important task in robotics, supporting perception, navigation, and scene understanding. Despite advances in LiDAR inertial visual odometry (LIVO), its performance remains highly sensitive to extrinsic calibration. Meanwhile, 3D vision foundation models, such as VGGT, suffer from limited scalability in large environments and inherently lack metric scale. To overcome these limitations, we propose LiDAR-VGGT, a novel framework that tightly couples LiDAR inertial odometry with the state-of-the-art VGGT model through a two-stage coarse- to-fine fusion pipeline: First, a pre-fusion module with robust initialization refinement efficiently estimates VGGT poses and point clouds with coarse metric scale within each session. Then, a post-fusion module enhances cross-modal 3D similarity transformation, using bounding-box-based regularization to reduce scale distortions caused by inconsistent FOVs between LiDAR and camera sensors. Extensive experiments across multiple datasets demonstrate that LiDAR-VGGT achieves dense, globally consistent colored point clouds and outperforms both VGGT-based methods and LIVO baselines. The implementation of our proposed novel color point cloud evaluation toolkit will be released as open source.

cs.RO↗

arXiv2025

High pressure lattice dynamics study of few layer-$α$-In$_2$Se$_3$

Few-layer $α$-In$_2$Se$_3$ has been studied under pressure using Raman spectroscopy in a diamond anvil cell up to 60 GPa (at room temperature). A combination of AFM and Raman was used to estimate the thickness of the specimens. While few-layer $α$-In$_2$Se$_3$ shows identical structural evolution with the one of the bulk powder-like form of $α$-In$_2$Se$_3$ ( $α$ $\rightarrow$ $β^{'}$ $\rightarrow$ IV ), an abrupt $β^{'}$ $\rightarrow$ IV phase transition (at 45 GPa) was observed, in contrast with the case of the bulk specimen where the two phases coexist over a wide pressure range. This is attributed to the difference in specimens morphology, $i.e.$ single crystal and powder in the case of few-layer and bulk $α$-In$_2$Se$_3$, respectively. This study documents the significance of specimens morphology on the observed pressure-induced phase transitions. The methodology developed in this study for performing high-pressure Raman measurements can be applied to other nanodimensional layered materials.

cond-mat.mtrl-sci↗

arXiv2025

Query-Centric Diffusion Policy for Generalizable Robotic Assembly

The robotic assembly task poses a key challenge in building generalist robots due to the intrinsic complexity of part interactions and the sensitivity to noise perturbations in contact-rich settings. The assembly agent is typically designed in a hierarchical manner: high-level multi-part reasoning and low-level precise control. However, implementing such a hierarchical policy is challenging in practice due to the mismatch between high-level skill queries and low-level execution. To address this, we propose the Query-centric Diffusion Policy (QDP), a hierarchical framework that bridges high-level planning and low-level control by utilizing queries comprising objects, contact points, and skill information. QDP introduces a query-centric mechanism that identifies task-relevant components and uses them to guide low-level policies, leveraging point cloud observations to improve the policy's robustness. We conduct comprehensive experiments on the FurnitureBench in both simulation and real-world settings, demonstrating improved performance in skill precision and long-horizon success rate. In the challenging insertion and screwing tasks, QDP improves the skill-wise success rate by over 50% compared to baselines without structured queries.

cs.RO↗

arXiv2025

Near-Field Beam Training Through Beam Diverging

This paper investigates beam training techniques for near-field (NF) extremely large-scale antenna arrays (ELAAs). Existing NF beam training methods predominantly rely on beam focusing, where the base station (BS) transmits highly spatially selective beams to locate the user equipment (UE). However, these beam-focusing-based schemes suffer from both high beam sweeping overhead and limited accuracy in the NF, primarily due to the narrow beams' high susceptibility to misalignment. To address this, we propose a novel NF beam training paradigm using diverging beams. Specifically, we introduce the beam diverging effect and exploit it for low-overhead, high-accuracy beam training. First, we design a diverging codeword to induce the beam diverging effect with a single radio frequency (RF) chain. Next, we develop a diverging polar-domain codebook (DPC) along with a hierarchical method that enables angular-domain localization of the UE with only 2 log_2(N) pilots, where N denotes the number of antennas. Finally, we enhance beam training performance through two additional techniques: a DPC angular range reduction strategy to improve the effectiveness of beam diverging, and a pilot set expansion method to increase overall beam training accuracy. Numerical results show that our algorithm achieves near-optimal accuracy with a small pilot overhead, outperforming existing methods.

cs.IT↗

arXiv2025

3D Near-Field Beam Training for Uniform Planar Arrays through Beam Diverging

In future 6G communication systems, large-scale antenna arrays promise enhanced signal strength and spatial resolution, but they also increase the complexity of beam training. Moreover, as antenna counts grow and carrier wavelengths shrink, the channel model transits from far-field (FF) planar waves to near-field (NF) spherical waves, further complicating the beam training process. This paper focuses on millimeter-wave (mmWave) systems equipped with large-scale uniform planar arrays (UPAs), which produce 3D beam patterns and introduce additional challenges for NF beam training. Existing methods primarily rely on either FF steering or NF focusing codewords, both of which are highly sensitive to mismatches in user equipment (UE) location, leading to high sensitivity to even slight mismatch and excessive training overhead. In contrast, we introduce a novel beam training approach leveraging the beam-diverging effect, which enables adjustable wide-beam coverage using only a single radio frequency (RF) chain. Specifically, we first analyze the spatial characteristics of this effect in UPA systems and leverage them to construct hierarchical codebooks for coarse UE localization. Then, we develop a 3D sampling mechanism to build an NF refinement codebook for precise beam training. Numerical results demonstrate that the proposed algorithm achieves superior beam training performance while maintaining low training overhead.

cs.IT↗

arXiv2025

Strata: Hierarchical Context Caching for Long Context Language Model Serving

Large Language Models (LLMs) with expanding context windows face significant performance hurdles. While caching key-value (KV) states is critical for avoiding redundant computation, the storage footprint of long-context caches quickly exceeds GPU memory capacity, forcing production systems to adopt hierarchical caching across memory hierarchies. However, transferring large cached contexts back to the GPU introduces severe performance bottlenecks: fragmented I/O from paged layouts prevents full bandwidth utilization, and existing schedulers fail to account for cache-loading delays, leaving systems loading-bound rather than compute-bound. We present Strata, a hierarchical context caching framework designed for efficient long context LLM serving. Strata introduces GPU-assisted I/O to combat KV cache fragmentation, decoupling GPU and CPU memory layouts and employs cache-aware request scheduling to balance compute with I/O latency and overlapping unavoidable stalls with complementary tasks. Built on SGLang and deployed in production, Strata achieves up to 5x lower Time-To-First-Token (TTFT) compared to vLLM + LMCache and 3.75x speedup over NVIDIA TensorRT-LLM on long-context benchmarks, without degrading short-context performance.

cs.DC↗

arXiv2025

AnchorCrafter: Animate Cyber-Anchors Selling Your Products via Human-Object Interacting Video Generation

The generation of anchor-style product promotion videos presents promising opportunities in e-commerce, advertising, and consumer engagement. Despite advancements in pose-guided human video generation, creating product promotion videos remains challenging. In addressing this challenge, we identify the integration of human-object interactions (HOI) into pose-guided human video generation as a core issue. To this end, we introduce AnchorCrafter, a novel diffusion-based system designed to generate 2D videos featuring a target human and a customized object, achieving high visual fidelity and controllable interactions. Specifically, we propose two key innovations: the HOI-appearance perception, which enhances object appearance recognition from arbitrary multi-view perspectives and disentangles object and human appearance, and the HOI-motion injection, which enables complex human-object interactions by overcoming challenges in object trajectory conditioning and inter-occlusion management. Extensive experiments show that our system improves object appearance preservation by 7.5\% and doubles the object localization accuracy compared to existing state-of-the-art approaches. It also outperforms existing approaches in maintaining human motion consistency and high-quality video generation. Project page including data, code, and Huggingface demo: https://github.com/cangcz/AnchorCrafter.

cs.CV↗

arXiv2025

LEMON-Mapping: Loop-Enhanced Large-Scale Multi-Session Point Cloud Merging and Optimization for Globally Consistent Mapping

Multi-robot collaboration is becoming increasingly critical and presents significant challenges in modern robotics, especially for building a globally consistent, accurate map. Traditional multi-robot pose graph optimization (PGO) methods ensure basic global consistency but ignore the geometric structure of the map, and only use loop closures as constraints between pose nodes, leading to divergence and blurring in overlapping regions. To address this issue, we propose LEMON-Mapping, a loop-enhanced framework for large-scale, multi-session point cloud fusion and optimization. We re-examine the role of loops for multi-robot mapping and introduce three key innovations. First, we develop a robust loop processing mechanism that rejects outliers and a loop recall strategy to recover mistakenly removed but valid loops. Second, we introduce spatial bundle adjustment for multi-robot maps, reducing divergence and eliminating blurring in overlaps. Third, we design a PGO-based approach that leverages refined bundle adjustment constraints to propagate local accuracy to the entire map. We validate LEMON-Mapping on several public datasets and a self-collected dataset. The experimental results show superior mapping accuracy and global consistency of our framework compared to traditional merging methods. Scalability experiments also demonstrate its strong capability to handle scenarios involving numerous robots.

cs.RO↗

arXiv2025

XGrammar: Flexible and Efficient Structured Generation Engine for Large Language Models

The applications of LLM Agents are becoming increasingly complex and diverse, leading to a high demand for structured outputs that can be parsed into code, structured function calls, and embodied agent commands. These developments bring significant demands for structured generation in LLM inference. Context-free grammar is a flexible approach to enable structured generation via constrained decoding. However, executing context-free grammar requires going through several stack states over all tokens in vocabulary during runtime, bringing non-negligible overhead for structured generation. In this paper, we propose XGrammar, a flexible and efficient structure generation engine for large language models. XGrammar accelerates context-free grammar execution by dividing the vocabulary into context-independent tokens that can be prechecked and context-dependent tokens that need to be interpreted during runtime. We further build transformations to expand the grammar context and reduce the number of context-independent tokens. Additionally, we build an efficient persistent stack to accelerate the context-dependent token checks. Finally, we co-design the grammar engine with LLM inference engine to overlap grammar computation with GPU executions. Evaluation results show that XGrammar can achieve up to 100x speedup over existing solutions. Combined with an LLM inference engine, it can generate near-zero overhead structure generation in end-to-end low-LLM serving.

cs.CL↗