SearcharxivSearch

arXiv subjects

Yongchao He

Publications and source records attributed to Yongchao He.

16 recordsLinked to original sources

FluxMoE: Decoupling Expert Residency for High-Performance MoE Serving

Mixture-of-Experts (MoE) models have become mainstream for scaling language models to hundreds of billions of expert parameters. Despite sparse expert activation, existing inference engines keep all experts GPU-resident, crowding out the key-value cache in large-batch, long-output offline workloads. We present FluxMoE, which decouples experts from physical GPU residency and adapts their footprint to available memory through a new \emph{expert paging} abstraction. FluxMoE combines PagedTensor for transparent remapping, a bandwidth-balanced hierarchy spanning losslessly compressed GPU memory and host DRAM, and a budget-aware residency planner. Unlike CPU-GPU co-inference and whole-layer offloading, FluxMoE streams weights on demand while keeping expert computation on GPUs. We implement FluxMoE atop vLLM and evaluate it on three MoE models. For GLM-4.5 on 8$\times$H20 GPUs, FluxMoE delivers up to 7.2$\times$ vLLM's throughput and 79.0\% lower average Time-Per-Output-Token (TPOT), without measurable model-quality loss using lossless compression. For Mixtral-8$\times$7B-Instruct on 2$\times$L40S GPUs, where weight-resident vLLM cannot fit, FluxMoE delivers 4.3$\times$ KTransformers's throughput and 29.1\% lower average TPOT.

cs.LG

InstantInfer: Enabling Fast LLM Cold Start with Communicating Finite Automata

Cold starts in large language model (LLM) inference services significantly affect user experience, yet they remain inefficient due to sequential initialization and a massive number of fine-grained I/O requests issued by complex software components. Although refactoring the program can yield advantages such as concurrent execution and I/O merging, this approach is error-prone and carries correctness risks when dealing with massive, heterogeneous components. We propose the Communicating Finite Automata (CFA) abstraction to systematically analyze cross-component optimization opportunities, and design a programming framework to enable CFA-based component program refactoring. This framework preserves the original sequential program structure while enabling safe concurrent component execution. We prove the correctness of the program refactoring. We apply the CFA abstraction and framework to refactor process tree creation, tensor loading, and model switching in vLLM, forming a new cold-start system named InstantInfer. Extensive experiments demonstrate that InstantInfer substantially accelerates LLM cold starts (achieving up to 7.2 times speedup) and exhibits robustness across diverse GPUs, workloads, and scales.

cs.DC

Cross-Field Channel Parameter Estimation and Channel Characterization at THz Bands in Indoor Scenarios

The terahertz (THz) frequency band offers the potential for ultra-high data rate transmission in future wireless communication systems. To extend the transmission distance and enhance spectral efficiency, the deployment of large-scale antenna arrays emerges as a promising solution in the THz band. This paper targets the critical challenge of cross-field (hybrid near-field/far-field) channel parameter estimation and channel characterization in such configurations. We first establish a 260-380 GHz virtual uniform linear array (ULA) measurement framework in an indoor scenario, capturing high-resolution channel transfer functions (CTFs) that reveal spatial non-stationarity and cross-field wavefront characteristics. Building upon these empirical observations, we propose a cross-field space-alternating generalized expectation-maximization (SAGE) algorithm that discriminatively estimates near-field and far-field multipath components (MPCs) via Bayesian phase-curvature classification, while explicitly tracking spatial birth-death phenomena through visibility region estimation. Analysis of the measurement data validates the algorithm's effectiveness in resolving cross-field MPCs and quantifies that near-field MPCs account for over 90% of total MPCs at 2 m transmission distance (380 GHz). We observe that spatial non-stationarity intensifies as the carrier frequency increases and the transmission distance decreases. These findings offer quantitative guidelines for channel modeling and system design in wireless THz communication systems.

eess.SP

A Hybrid Near-field Indoor Channel Model for THz Bands Based on Surface Scattering Characteristics

Terahertz (THz) communication and extremely large-scale MIMO (XL-MIMO) are essential for achieving ultra-high data rates in future 6G systems. However, at sub-millimeter wavelengths, typical indoor materials exhibit significant roughness that invalidates conventional ideal smooth surface assumptions, while massive array apertures introduce pronounced near-field effects and spatial non-stationarity. To address these challenges, this paper proposes a hybrid near-field channel model utilizing surface scattering characteristics based on distinct measurement campaigns. First, based on typical indoor materials scattering measurements across the 260-400 GHz band, an improved Beckmann-Kirchhoff (B-K) model is developed to accurately characterize surface roughness and diffuse scattering behavior. The model independently analyzes single-bounce (SB) and multi-bounce (MB) clusters by applying deterministic rough surface scattering theory and geometry-statistical approach, respectively. Then, using near-field spatial non-stationarity measurements from a 630-element virtual array in the 330-360 GHz band, a Dual-Gaussian Mixture Model (DMM) and a Negative Binomial (NB) distribution are adopted to describe the lengths and the number of spatial visibility regions (VRs), respectively. Additionally, a Weibull distribution is employed to model the intra-region power fluctuations. Finally, comprehensive XL-MIMO channel evaluations within the same band demonstrate that the proposed model aligns closely with measured results in terms of the spatial cross-correlation function (SCCF), frequency cross-correlation function (FCF), and channel capacity. By reproducing the spatial sparsity of THz band, the proposed model overcomes the limitation of conventional standard models, such as 3GPP 38.901 and WINNER II, in significantly overestimating channel capacity.

eess.SP

Channel Measurements and Characterization with Phase Drift Compensation for Outdoor 330-360 GHz MIMO Communications

In this paper, an outdoor channel measurement campaign at 330-360 GHz employing a 128 * 4 virtual antenna array (VAA)-based multiple-input multiple-output (MIMO) configuration is conducted. The transmitter (Tx) and receiver (Rx) location pairs are classified into line-of-sight (LoS) and obstructed-LoS (OLoS) scenarios to enable a detailed investigation of outdoor terahertz (THz) band channel characteristics. During the measurement process, the stationarity of the outdoor environment is carefully verified, and a linear phase drift (PD) effect is identified. Then, we propose a PD-aware Space-Alternating Generalized Expectation-Maximization (SAGE) algorithm, which significantly improves both delay resolution and channel parameter estimation accuracy. Based on the processed measurement data, we characterize key channel properties, including the power delay profile, path loss, shadow fading, delay spread, angular spread, Rician K-factor, as well as their cumulative distribution functions and correlation characteristics. In addition, near-field effects and MIMO-specific properties, including the spatial non-stationarity and the cluster birth-death property, are analyzed.

eess.SP

A Measurement-Based Parameterization of Physics Reflection Models for Terahertz Communication

The accurate modeling of reflection coefficients is pivotal for developing reliable channel models in emerging terahertz (THz) communications. This study establishes a 300$\sim$400 GHz channel measurement platform to measure the reflection coefficients of various materials. Based on the analysis of measured data, we propose the single-layer interference with an extended-parameterized Lorentz/Drude (SLI-EPLD) reflection coefficient model. In this model, a sub-band modeling strategy is adopted to characterize the variation of reflection coefficients with frequency, while a parameterized mapping approach is employed to ensure the stability of model parameters. Furthermore, the weighted sub-band fitting for trend regression (WF-TREND) algorithm is introduced to achieve precise sub-band parameter fitting. Validation results demonstrate superior performance to existing models across multiple materials. The reflection coefficient model established in this work serves as a critical foundation for channel modeling in 300$\sim$400 GHz for high-THz communication.

eess.SP

CascadeInfer: Length-Aware Scheduling of LLM Serving with Low Latency and Load Balancing

Efficiently harnessing GPU compute is critical to improving user experience and reducing operational costs in large language model (LLM) services. However, current inference engine schedulers overlook the attention backend's sensitivity to request-length heterogeneity within a batch. As state-of-the-art models now support context windows exceeding 128K tokens, this once-tolerable inefficiency has escalated into a primary system bottleneck, causing severe performance degradation through GPU underutilization and increased latency. We present CascadeInfer, a runtime system that dynamically reschedules requests across multiple instances serving the same LLM to mitigate per-instance length heterogeneity. CascadeInfer partitions these instances into length-specialized groups, each handling requests within a designated length range, naturally forming a pipeline as requests flow through them. CascadeInfer devises a dynamic programming algorithm to efficiently find the stage partition with the best QoE, employs runtime range refinement together with decentralized load (re)balance both across and within groups, achieving a balanced and efficient multi-instance service. Our evaluation shows that, under the same configuration, CascadeInfer reduces end-to-end latency by up to 67% and tail latency by up to 69%, while improving overall system throughput by up to 2.89 times compared to the state-of-the-art multi-instance scheduling systems.

cs.DC

A Unified Sparse Attention via Multi-Granularity Compression

Efficient long-context understanding and reasoning are increasingly vital for large language model (LLM) applications such as multi-turn dialogue and program analysis. However, the core self-attention mechanism scales quadratically with sequence length, creating a fundamental computational bottleneck. Existing sparse attention methods alleviate this issue but face trade-offs: training-based methods are costly and cannot be directly applied as acceleration plugins for other models, while inference-time methods often compromise efficiency or cross-modal generality. To address these limitations, we present UniSparse, a unified mechanism that introduces the notion of composite tokens--compact representations that aggregate multi-granularity contextual information. Building on this abstraction, UniSparse dynamically constructs sparse attention through multi-granularity compression and block-level selection, enabling efficient and hardware-friendly execution on GPU. Across multiple modalities and tasks ranging from synthetic benchmarks to real-world applications, UniSparse consistently surpasses state-of-the-art sparse attention methods (e.g., MInference, XAttention, FlexPrefill) in both accuracy and efficiency, achieving $\ge$ 99% of full-attention accuracy and up to 2.61$\times$ faster attention computation than FlashAttention.

cs.CL

SIMPLE: Disaggregating Sampling from GPU Inference into a Decision Plane for Faster Distributed LLM Serving

As large language models (LLMs) scale out with tensor parallelism (TP) and pipeline parallelism (PP) and production stacks have aggressively optimized the data plane (attention/GEMM and KV cache), sampling, the decision plane that turns logits into tokens, becomes a new bottleneck. This creates a structural holdout: sampling neither expands with TP nor balances across PP stages, so its share of iteration time grows as GPUs get faster and it caps pipeline frequency at the last stage. We present SIMPLE, a stage-agnostic, sequence-parallel, overlappable decision plane that disaggregates sampling into a CPU-side service and shrinks its runtime footprint back to a minor, hidden role. SIMPLE combines: (1) sequence-parallel sampling, which shards work along the batch dimension and removes vocabulary-axis collectives; (2) a CPU-based algorithm with column-wise penalties and truncation-first filtering to realize single-pass, linear-time kernels; and (3) speculative hot-vocab sampling (SHVS), which samples on a small hot set with rejection-correctness and uses a simple sizing model to choose the hot-vocab size that maximizes throughput. In evaluation, SIMPLE improves end-to-end throughput by up to 96% and reduces P95 latency by 20-65%. Crucially, SIMPLE requires no user-side code changes and composes with existing data-plane optimizations, unlocking scaling benefits that compound with future GPU generations.

cs.DC

Theoretical and Empirical Study of Spatial Power Focusing Effect for Sparse Arrays at Terahertz Band

This work investigates the spatial power focusing effect for large-scale sparse arrays at terahertz (THz) band, combining theoretical analysis with experimental validation. Specifically, based on a Green's function channel model, we analyze the power distribution along the $z$-axis, deriving a closed-form expression to characterize the focusing effect. Furthermore, the factors influencing the focusing effect, including phase noise and positional deviations, are theoretically analyzed and numerically simulated. Finally, a 300 GHz measurement platform based on a vector network analyzer (VNA) is constructed for experimental validation. The measurement results demonstrate close consistence with theoretical simulation results, confirming the spatial power focusing effect for sparse arrays.

eess.SP

HeteroSpec: Leveraging Contextual Heterogeneity for Efficient Speculative Decoding

Autoregressive decoding inherently limits the inference throughput of Large Language Model (LLM) due to its sequential dependency. Speculative decoding mitigates this by verifying multiple predicted tokens in parallel, but its efficiency remains constrained by what we identify as verification heterogeneity -- the uneven difficulty of verifying different speculative candidates. In practice, a small subset of high-confidence predictions accounts for most successful verifications, yet existing methods treat all candidates uniformly, leading to redundant computation. We present HeteroSpec, a heterogeneity-adaptive speculative decoding framework that allocates verification effort in proportion to candidate uncertainty. HeteroSpec estimates verification complexity using a lightweight entropy-based quantifier, partitions candidates via a data-driven stratification policy, and dynamically tunes speculative depth and pruning thresholds through coordinated optimization. Across five benchmarks and four LLMs, HeteroSpec delivers an average 4.24$\times$ decoding speedup over state-of-the-art methods such as EAGLE-3, while preserving exact output distributions. Crucially, HeteroSpec requires no model retraining and remains compatible with other inference optimizations, making it a practical direction for improving speculative decoding efficiency.

cs.CL

MegatronApp: Efficient and Comprehensive Management on Distributed LLM Training

The rapid escalation in the parameter count of large language models (LLMs) has transformed model training from a single-node endeavor into a highly intricate, cross-node activity. While frameworks such as Megatron-LM successfully integrate tensor (TP), pipeline (PP), and data (DP) parallelism to enable trillion-parameter training, they simultaneously expose practitioners to unprecedented systems-level challenges in performance optimization, diagnosis, and interpretability. MegatronApp is an open-source toolchain expressly designed to meet these challenges. It introduces four orthogonal, yet seamlessly composable modules--MegaScan, MegaFBD, MegaDPP, and MegaScope--that collectively elevate the reliability, efficiency, and transparency of production-scale training. This paper presents the motivation, architecture, and distinctive contributions of each module, and elucidates how their synergistic integration augments the Megatron-LM ecosystem.

cs.DC

SiPipe: Bridging the CPU-GPU Utilization Gap for Efficient Pipeline-Parallel LLM Inference

As inference workloads for large language models (LLMs) scale to meet growing user demand, pipeline parallelism (PP) has become a widely adopted strategy for multi-GPU deployment, particularly in cross-node setups, to improve key-value (KV) cache capacity and inference throughput. However, PP suffers from inherent inefficiencies caused by three types of execution bubbles-load-imbalance, intra-stage, and inter-stage-which limit pipeline saturation. We present SiPipe, a heterogeneous pipeline design that improves throughput by leveraging underutilized CPU resources to offload auxiliary computation and communication. SiPipe incorporates three key techniques-CPU sampling, a token-safe execution model, and structure-aware transmission-to mitigate pipeline bubbles and improve execution efficiency. Across diverse LLMs, SiPipe achieves up to 2.1 times higher throughput, 43% lower per-token latency, and up to 23% higher average GPU utilization compared to the state-of-the-art vLLM under the same PP configuration, demonstrating its generality across LLMs and deployment scenarios.

cs.DC

Reconfigurable Intelligent Surface-Aided Dual-Function Radar and Communication Systems With MU-MIMO Communication

In this paper, we investigate an reconfigurable intelligent surface (RIS)-aided integrated sensing and communication (ISAC) system. Our objective is to maximize the achievable sum rate of the multi-antenna communication users through the joint active and passive beamforming. {Specifically}, the weighted minimum mean-square error (WMMSE) method is { first} used to reformulate the original problem into an equivalent one. Then, we utilize an alternating optimization (AO) { algorithm} to decouple the optimization variables and decompose this challenging problem into two subproblems. Given reflecting coefficients, a penalty-based algorithm is utilized to deal with the non-convex radar signal-to-noise ratio (SNR) constraints. For the given beamforming matrix of the BS, we apply majorization-minimization (MM) to transform the problem into a quadratic constraint quadratic programming (QCQP) problem, which is ultimately solved using a semidefinite relaxation (SDR)-based algorithm. Simulation results illustrate the advantage of deploying RIS in the considered multi-user MIMO (MU-MIMO) ISAC systems.

eess.SP

Joint Beamforming Design for Double Active RIS-assisted Radar-Communication Coexistence Systems

Integrated sensing and communication (ISAC) technology has been considered as one of the key candidate technologies in the next-generation wireless communication systems. However, when radar and communication equipment coexist in the same system, i.e. radar-communication coexistence (RCC), the interference from communication systems to radar can be large and cannot be ignored. Recently, reconfigurable intelligent surface (RIS) has been introduced into RCC systems to reduce the interference. However, the "multiplicative fading" effect introduced by passive RIS limits its performance. To tackle this issue, we consider a double active RIS-assisted RCC system, which focuses on the design of the radar's beamforming vector and the active RISs' reflecting coefficient matrices, to maximize the achievable data rate of the communication system. The considered system needs to meet the radar detection constraint and the power budgets at the radar and the RISs. Since the problem is non-convex, we propose an algorithm based on the penalty dual decomposition (PDD) framework. Specifically, we initially introduce auxiliary variables to reformulate the coupled variables into equation constraints and incorporate these constraints into the objective function through the PDD framework. Then, we decouple the equivalent problem into several subproblems by invoking the block coordinate descent (BCD) method. Furthermore, we employ the Lagrange dual method to alternately optimize these subproblems. Simulation results verify the effectiveness of the proposed algorithm. Furthermore, the results also show that under the same power budget, deploying double active RISs in RCC systems can achieve higher data rate than those with single active RIS and double passive RISs.

eess.SP

Secure Wireless Communication in Active RIS-Assisted DFRC System

This work considers a dual-functional radar and communication (DFRC) system with an active reconfigurable intelligent surface (RIS) and a potential eavesdropper. Our purpose is to maximize the secrecy rate (SR) of the system by jointly designing the beamforming matrix at the DFRC base station (BS) and the reflecting coefficients at the active RIS, subject to the signal-to-interference-plus-noise-ratio (SINR) constraint of the radar echo and the power consumption constraints at the DFRC-BS and active RIS. An alternating optimization (AO) algorithm based on semi-definite relaxation (SDR) and majorizationminimization (MM) is applied to solve the SR-maximization problem by alternately optimizing the beamforming matrix and the reflecting coefficients. Specifically, we first apply the SDR and successive convex approximation (SCA) methods to transform the two subproblems into more tractable forms, then the MM method is applied to derive a concave surrogate function and iteratively solve the subproblems. Finally, simulation results indicate that the active RIS can better confront the impact of "multiplicative fading" and outperforms traditional passive RIS in terms of both secure data rate and radar sensing performance.

eess.SP