arXiv subjects

Wenfeng Wang

Publications and source records attributed to Wenfeng Wang.

9 recordsLinked to original sources

arXiv2026

MoE-Prism: Disentangling Monolithic Experts for Elastic MoE Services via Model-System Co-Designs

Mixture-of-Experts (MoE) scales model capacity through sparse activation, and is becoming an important architecture for large language models (LLMs). However, existing MoE serving systems typically execute all requests under a fixed routing configuration, limiting their ability to exploit heterogeneous computation requirements across requests. Routing top-$k$, which determines the number of routed experts activated per token, directly controls routed-expert computation and provides a natural mechanism for request-level compute elasticity. Realizing this capability, however, requires finer-grained routing units and efficient runtime execution for heterogeneous routing budgets. We present \textsc{MoE-Prism}, a model and system support framework for request-level compute elasticity in MoE serving. \textsc{MoE-Prism}decomposes monolithic experts into fine-grained sub-experts to expose denser routing operating points and provides a $k$-aware serving runtime that effectively serves heterogeneous routing budgets under both throughput-oriented and latency-sensitive workloads. We implement \textsc{MoE-Prism} on top of vLLM and evaluate it on three representative MoE models. \textsc{MoE-Prism} expands the number of available routing operating points by $4\times$, improves offline inference throughput by up to 33.9\%, and reduces online serving TTFT under heterogeneous workloads. These results demonstrate practical elastic MoE serving with request-level routing targets.

cs.CL↗

arXiv2026

Implement Kubernetes Pod-Level Remote Attestation for Confidential Workloads on dstack

The rise of LLM-as-a-Service and other confidential cloud workloads demands cryptographic proof that user data is processed in a trusted, untampered environment. Existing solutions, notably Confidential Containers (CoCo), enforce a strict "one Pod per VM" model that attests only the Guest OS stack, leaving container-level identity unverified and incurring prohibitive per-VM resource overhead. We present dstack-capsule, a Kubernetes platform that enables Pod-level remote attestation on Intel TDX by allowing multiple Pods to share a single Confidential VM while each retains independent, hardware-backed proof of identity. Our key insight is a two-layer attestation architecture: static platform measurements are frozen in RTMR[3] via an irreversible privilege fuse, while dynamic Pod identities (pod_uid, pod_spec_hash, workload_id) are embedded in the TDX Quote's report_data field and signed by hardware on every request. dstack-capsule introduces (1) a Pod-level attestation protocol binding Pod spec digests to hardware-signed Quotes; (2) a privilege fuse mechanism that atomically transitions a node from setup mode to secure mode; (3) a multi-layer sandbox spanning storage, runtime, admission, API, and network isolation layers; and (4) a complete open-source implementation based on Kubernetes 1.32, Intel TDX, and Sysbox. We evaluate the security properties, attestation correctness, and performance characteristics of dstack-capsule, demonstrating that it achieves Pod-granularity verification without the resource overhead of per-VM isolation.

cs.CR↗

arXiv2026

PCR: A Prefetch-Enhanced Cache Reuse System for Low-Latency RAG Serving

Retrieval-Augmented Generation (RAG) systems enhance the performance of large language models (LLMs) by incorporating supplementary retrieved documents, enabling more accurate and context-aware responses. However, integrating these external documents often results in very long input sequences, which significantly increases computation costs during the prefill stage, where key-value (KV) representations for all input tokens are generated. This latency bottleneck becomes especially pronounced under high-throughput serving scenarios. KV-cache reuse offers a promising solution by storing previously computed KV states for shared input prefixes, thereby avoiding redundant computation across requests that contain overlapping context. Yet, the effectiveness of cache reuse is often limited by three practical challenges: low cache hit rates due to naive eviction policies, high CPU-GPU data transfer overhead, and slow SSD I/O when caches spill to storage. To address these issues, we propose PCR, a system designed to maximize KV-cache reuse efficiency through intelligent prefetching and pipelined data movement. Specifically, PCR introduces three key techniques: (1) a prefix-tree caching structure with a look-ahead LRU replacement policy that uses pending requests in the scheduler queue to improve cache hit ratios; (2) layer-wise overlapping that pipelines KV-cache loading and GPU computation across CUDA streams to hide communication latency; and (3) queue-based prefetching that proactively loads relevant KV caches from SSD into DRAM before they are needed. Extensive experiments show that PCR outperforms existing KV-cache reuse methods, achieving up to a 2.47x speedup in terms of average TTFT.

cs.DC↗

arXiv2025

MoE-SpeQ: Speculative Quantized Decoding with Proactive Expert Prefetching and Offloading for Mixture-of-Experts

The immense memory requirements of state-of-the-art Mixture-of-Experts (MoE) models present a significant challenge for inference, often exceeding the capacity of a single accelerator. While offloading experts to host memory is a common solution, it introduces a severe I/O bottleneck over the PCIe bus, as the data-dependent nature of expert selection places these synchronous transfers directly on the critical path of execution, crippling performance. This paper argues that the I/O bottleneck can be overcome by trading a small amount of cheap, on-device computation to hide the immense cost of data movement. We present MoE-SpeQ, a new inference system built on a novel co-design of speculative execution and expert offloading. MoE-SpeQ employs a small, on-device draft model to predict the sequence of required experts for future tokens. This foresight enables a runtime orchestrator to prefetch these experts from host memory, effectively overlapping the expensive I/O with useful computation and hiding the latency from the critical path. To maximize performance, an adaptive governor, guided by an Amortization Roofline Model, dynamically tunes the speculation strategy to the underlying hardware. Our evaluation on memory-constrained devices shows that for the Phi-MoE model, MoE-SpeQ achieves at most 2.34x speedup over the state-of-the-art offloading framework. Our work establishes a new, principled approach for managing data-dependent memory access in resource-limited environments, making MoE inference more accessible on commodity hardware.

cs.LG↗

arXiv2025

A Survey on Inference Optimization Techniques for Mixture of Experts Models

The emergence of large-scale Mixture of Experts (MoE) models represents a significant advancement in artificial intelligence, offering enhanced model capacity and computational efficiency through conditional computation. However, deploying and running inference on these models presents significant challenges in computational resources, latency, and energy efficiency. This comprehensive survey analyzes optimization techniques for MoE models across the entire system stack. We first establish a taxonomical framework that categorizes optimization approaches into model-level, system-level, and hardware-level optimizations. At the model level, we examine architectural innovations including efficient expert design, attention mechanisms, various compression techniques such as pruning, quantization, and knowledge distillation, as well as algorithm improvement including dynamic routing strategies and expert merging methods. At the system level, we investigate distributed computing approaches, load balancing mechanisms, and efficient scheduling algorithms that enable scalable deployment. Furthermore, we delve into hardware-specific optimizations and co-design strategies that maximize throughput and energy efficiency. This survey provides both a structured overview of existing solutions and identifies key challenges and promising research directions in MoE inference optimization. To facilitate ongoing updates and the sharing of cutting-edge advances in MoE inference optimization research, we have established a repository accessible at https://github.com/MoE-Inf/awesome-moe-inference/.

cs.LG↗

arXiv2024

Membership Information Leakage in Federated Contrastive Learning

Federated Contrastive Learning (FCL) represents a burgeoning approach for learning from decentralized unlabeled data while upholding data privacy. In FCL, participant clients collaborate in learning a global encoder using unlabeled data, which can serve as a versatile feature extractor for diverse downstream tasks. Nonetheless, FCL is susceptible to privacy risks, such as membership information leakage, stemming from its distributed nature, an aspect often overlooked in current solutions. This study delves into the feasibility of executing a membership inference attack on FCL and proposes a robust attack methodology. The attacker's objective is to determine if the data signifies training member data by accessing the model's inference output. Specifically, we concentrate on attackers situated within a client framework, lacking the capability to manipulate server-side aggregation methods or discern the training status of other clients. We introduce two membership inference attacks tailored for FCL: the \textit{passive membership inference attack} and the \textit{active membership inference attack}, contingent on the attacker's involvement in local model training. Experimental findings across diverse datasets validate the effectiveness of our attacks and underscore the inherent privacy risks associated with the federated contrastive learning paradigm.

cs.CR↗

arXiv2023

Electrical contact properties between Yb and few-layer WS$_2$

Charge injection mechanism from contact electrodes into two-dimensional (2D) dichalcogenides is an essential topic for exploiting electronics based on 2D channels, but remains not well understood. Here, low-work-function metal ytterbium (Yb) was employed as contacts for tungsten disulfide (WS$_2$) to understand the realistic injection mechanism. The contact properties in WS$_2$ with variable temperature (T) and channel thickness (tch) were synergetically characterized. It is found that the Yb/WS$_2$ interfaces exhibit a strong pinning effect between energy levels and a low contact resistance ($R_\rm{C}$) value down to $5\,kΩ\cdotμ$m. Cryogenic electrical measurements reveal that $R_\rm{C}$ exhibits weakly positive dependence on T till 77 K, as well as a weakly negative correlation with tch. In contrast to the non-negligible $R_\rm{C}$ values extracted, an unexpectedly low effective thermal injection barrier of 36 meV is estimated, indicating the presence of significant tunneling injection in subthreshold regime and the inapplicability of the pure thermionic emission model to estimate the height of injection barrier.

cond-mat.mtrl-sci↗

arXiv2018

Anomalous increasing of the intensity of field dependence optical mode ferromagnetic resonance in the exchange coupled bilayer system

Acoustic and optical ferromagnetic resonance (FMR) in the interlayer exchange coupled Fe$_{20}$Ni$_{80}$/Co bilayer have been investigated. In the optical mode, unexpected increasing tendencies of peak value at the resonance frequency has been observed under an increasing magnetic field. We presented analytical calculations with which the exchange coupling between Co and Fe$_{20}$Ni$_{80}$ layers, the magnetization and the in-plane uniaxial anisotropy are taken into account, to interpret the increasing of the maximum values of the optical permeability. Both experimental measurements and theoretical calculation show that such tendencies are dependent on the layer thickness t, and that there is a critical field above which the optical peak value begins to decrease. These results might help us to understand the mechanism of interlayer exchange coupling induced optical FMR and might enlighten us to find new possibility of high frequency applications of magnetic materials.

cond-mat.mes-hall↗

arXiv2011

b-> s gamma and b-> d gamma (B factories)

The photon spectrum in B->X_{s,d} γdecay, where X_s(d) is any strange (non-strange) hadronic state, is studied using data samples of $e^+e^- \rightarrow Υ(4S) \to B B-bar decays collected by the Babar and Belle experiments. Here I present the latest measurements of the branching fraction and spectral mo ments from B -> X_s gamma$ decays by Belle and the direct CP asymmetry of B->X_{s+d} gamma measured at Babar. The determination of |V_{td}/V_{ts}|^2$ i also presented.

hep-ex↗