arXiv subjects

Fahao Chen

Publications and source records attributed to Fahao Chen.

14 recordsLinked to original sources

arXiv2026

On the Fundamental Limits of Prefill-Based Decoding Length Prediction

Predicting how many tokens a large language model (LLM) will decode, using only prefill-phase hidden states, is a key primitive for inference scheduling. We show that current benchmarks conflate predictor quality with task hardness: when the target model, dataset, or sampling configuration changes, reported accuracy collapses even though the predictor is unchanged. We trace this instability to an overlooked error source---the inherent stochasticity of autoregressive decoding---and derive bounds that every prefill-side method must obey. Our analysis treats decoding as a stopped stochastic process. The per-step stopping probability of a response is not constant but ramps up over hundreds of steps, so the decode length inherits a central limit theorem from the many local decisions that determine where a response ends; this yields per-prompt normality with an explicit error term, together with an affine variance law $σ=αμ+b$ whose slope is the coefficient of variation of the number of response segments. Empirically, the normal law reproduces the per-prompt MAE and the central mass of the decode-length distribution to within a few percent, and the residual departures from normality (mild right skew, slight excess kurtosis) are the ones predicted by the sampling-noise correction. Under fixed model, data, and sampling conditions, a broad design space of predictors (MLP, EGTP, FiLM, soft-label, and Mixture-of-Experts variants) converges to similar performance close to the bound, confirming that \emph{system error dominates model error}. We argue for a benchmarking protocol that fixes the model, data, and sampling parameters and that reports relative-error metrics alongside absolute ones. Code and analysis tools are available at https://github.com/ShwStone/DecodingPredict.

cs.AI↗

arXiv2026

Request-Adaptive Execution for Efficient Hybrid LLM Inference

Hybrid large language models (LLMs) have emerged as a promising way to combine the modeling capability of the Attention mechanism with the efficiency of state space models (SSMs). However, most existing hybrid LLMs still inherit a static design philosophy: once the Attention-to-SSM layout is chosen, every request must follow the same hybrid execution. This one-layout-fits-all design is inefficient for real-world serving workloads, where different requests rely on contextual information in different ways. This paper presents Morph, a serving system that co-designs the model architecture and runtime for request-adaptive hybrid LLM inference. Its model architecture pairs Attention and SSM layers within each block and uses prefill-oriented gating to select a request-specific layout that remains fixed throughout decoding. Its serving runtime preserves batching efficiency under divergent layer choices through selective batching and enables correct prefix reuse through hybrid-aware context caching. Experiments on six datasets show that Morph achieves a better quality-efficiency trade-off than existing static hybrid baselines. System-level evaluation further shows that Morph improves the serving throughput by up to 3.87x over baselines while substantially reducing inference latency.

cs.DC↗

arXiv2026

When Messages Go Wrong: Understanding Communication Faults in LLM-Based Multi-Agent Systems

LLM-based multi-agent systems (LLM-MAS) use specialized agents that exchange messages to accomplish complex tasks. Their reliance on networked protocols, remote tools, and model endpoints exposes communication to message loss, delay, and context expiration. Because agents often exchange open-ended natural language, anomalies can alter task-relevant information while leaving messages processable, allowing semantic errors to propagate through subsequent interactions. Systematic studies of faults in inter-agent communication remain limited. This paper presents a large-scale empirical study of how communication faults manifest, propagate, and affect task outcomes in LLM-MAS. We construct a cross-layer taxonomy linking lower-stack disturbances to application-visible faults and develop a fault-injection and tracing pipeline. Our evaluation covers more than 29,000 executions across 100 tasks, five domains, three workflow topologies, and three model configurations. We find that 20.7% of failures with complete detection records occur without an immediate failure signal, reaching 76.2% for partial-content faults. Flat maintains 53.0% task success under communication faults, compared with 3.3% and 3.2% for Sequential and Hierarchical, respectively. This advantage primarily arises when an alternative information path preserves the evidence required by downstream agents. We also evaluate execution-side and evidence protection mechanisms and find that they recover different failure cases. Evidence recovery depends on whether trustworthy information remains accessible through an unaffected path. These observations provide empirical insights into communication-fault propagation, workflow resilience, and protection design for LLM-MAS.

cs.SE↗

arXiv2026

Protecting Shared Agent Workflows from Invalidation Attacks with Signed Trajectories and Runtime Verification

Large language model (LLM)-based agents increasingly rely on shared workflow repositories to transfer experience across tasks and agents. By abstracting past executions into reusable workflow graphs, such systems significantly improve planning efficiency and generalization. However, this emerging paradigm introduces a new and underexplored security risk: if the shared repository is compromised, tampered workflows can systematically mislead downstream agents, inducing persistent execution failures or unsafe behaviors. In this work, we present a principled protection framework for shared agent workflows based on signed trajectories and runtime verification. We require all cloud-collected trajectories used for workflow generation to be cryptographically signed, with each trajectory representing a concrete execution instance. Prior to repository inclusion, trajectories undergo a unified screening process to ensure correctness and validity. When an agent retrieves a workflow from the shared repository, it also obtains a small set of signed supporting trajectories and verifies both their cryptographic integrity and their semantic consistency with the workflow. To make this practical for branching workflows, we further propose a runtime, path-aware retrieval strategy that dynamically fetches only the signed trajectories relevant to the execution path actually taken by the agent, substantially reducing communication and verification overhead. Extensive experiments across multiple reasoning benchmarks show that workflow invalidation attacks achieve up to 90.12% success rates, degrading agent inference performance by up to 73.16 percentage points in accuracy without effective workflow protection. In addition, the proposed protection achieves an average F1 score of 85.26% and a detection rate of 79.97%, outperforming existing baselines by about 2.02x.

cs.CR↗

arXiv2026

SparLeak: Privacy Leakage from Sparse Attention in LLM Inference on Shared GPUs

Sparse attention is widely used to accelerate long-context inference in modern large language models (LLMs), but its input-dependent execution behavior introduces previously unexplored privacy risks. We identify a new GPU micro-architectural side channel, termed Sparsity-Induced Memory Access (SIMA), which arises from secret-dependent key-value cache access patterns induced by sparse attention. Based on this observation, we present SparLeak, a phase-aware side-channel attack that extracts SIMA traces during LLM inference and enables two practical privacy extractions: query attribute inference from prefill-phase traces and autoregressive response reconstruction from decoding-phase traces. By reconstructing approximate token-level sparsity profiles from page-level observations and applying profiling-based learning, SparLeak accurately recovers sensitive information, including user-query attributes and private LLM response content. Extensive evaluation across three LLM architectures, three sparse attention mechanisms, and three privacy-sensitive datasets shows that SparLeak achieves average attack success rates of 90.9% for attribute inference and 87.3% for response reconstruction under real-world LLM serving settings, highlighting the significance to account for SIMA leakage when deploying sparse-attention-based LLM systems. We provide anonymized SIMA traces, trained attack models, evaluation scripts, and documentation as artifacts at https://anonymous.4open.science/r/Janus_artifacts/.

cs.LG↗

arXiv2026

Director: Accelerating Distributed MoE Serving via Online Proactive Expert Placement

Expert parallelism has become the prevailing paradigm to serve Mixture-of-Experts (MoE) models. Its efficiency depends on the communication and computation latencies of the GPUs, which are linked to the placement of experts in the GPUs. Existing works for optimizing expert placement focus on leveraging past requests' expert activation patterns. However, they demonstrate deficiencies facing diverse and rapidly changing request patterns, calling for an online, proactive approach. Implementing such an approach requires addressing several challenges: the uncertainty associated with incoming requests' expert activation, the cost of expert migration, and the NP-hard complexity in optimization. Therefore, we present Director, a new distributed MoE serving system that minimizes end-to-end latency via prediction-driven, online expert placement. Director uses either a lightweight cascaded predictor or a low-bit quantized replica for expert activation patterns of incoming requests. An online migration module then enacts the changes with near-zero downtime by executing migrations in compute-bound phases, keeping disruption bounded. At its core, a relaxation-based expert placement optimizer operates under capacity constraints, runs in polynomial time, and achieves a $(1+ε)$ approximation ratio. Finally, we implement a prototype and demonstrate, through extensive experiments, a reduction in end-to-end latency of $11\sim55\%$ for popular MoE models (e.g., Mistral, DeepSeek and Qwen) compared to existing work.

cs.LG↗

arXiv2026

GraphFlow: A Graph-Based Workflow Management for Efficient LLM-Agent Serving

Large Language Model (LLM)-based agents demonstrate strong reasoning and execution capabilities on complex tasks when guided by structured instructions, commonly referred to as workflows. However, existing workflow-assisted agent serving systems typically rely on predefined templates and shallow matching mechanisms, which limit their ability to capture deep semantic relationships and generalize to previously unseen tasks. To address these limitations, we propose a new workflow management paradigm that represents workflows using a unified graph, termed wGraph, where each node corresponds to an atomic operation. wGraph serves as a shared substrate from which task-specific workflows are dynamically instantiated. Building on wGraph primitives, we introduce GraphFlow, a system that efficiently integrates workflows into agent serving through two key designs. First, adaptive workflow generation dynamically constructs workflows from wGraph based on task semantics and constraint requirements. Second, workflow state management exploits wGraph structure to efficiently manage Key-Value (KV) caches, reducing redundant computation during agent serving. Extensive experiments across five benchmark datasets show that GraphFlow consistently outperforms state-of-the-art methods, yielding an average performance improvement of approximately 4.95 percentage points, while achieving an approximately 4$\times$ reduction in memory footprint.

cs.LG↗

arXiv2025

Internet of Agents: Fundamentals, Applications, and Challenges

With the rapid proliferation of large language models and vision-language models, AI agents have evolved from isolated, task-specific systems into autonomous, interactive entities capable of perceiving, reasoning, and acting without human intervention. As these agents proliferate across virtual and physical environments, from virtual assistants to embodied robots, the need for a unified, agent-centric infrastructure becomes paramount. In this survey, we introduce the Internet of Agents (IoA) as a foundational framework that enables seamless interconnection, dynamic discovery, and collaborative orchestration among heterogeneous agents at scale. We begin by presenting a general IoA architecture, highlighting its hierarchical organization, distinguishing features relative to the traditional Internet, and emerging applications. Next, we analyze the key operational enablers of IoA, including capability notification and discovery, adaptive communication protocols, dynamic task matching, consensus and conflict-resolution mechanisms, and incentive models. Finally, we identify open research directions toward building resilient and trustworthy IoA ecosystems.

cs.MA↗

arXiv2025

Federated Fine-Tuning of Sparsely-Activated Large Language Models on Resource-Constrained Devices

Federated fine-tuning of Mixture-of-Experts (MoE)-based large language models (LLMs) is challenging due to their massive computational requirements and the resource constraints of participants. Existing working attempts to fill this gap through model quantization, computation offloading, or expert pruning. However, they cannot achieve desired performance due to impractical system assumptions and a lack of consideration for MoE-specific characteristics. In this paper, we propose FLUX, a system designed to enable federated fine-tuning of MoE-based LLMs across participants with constrained computing resources (e.g., consumer-grade GPUs), aiming to minimize time-to-accuracy. FLUX introduces three key innovations: (1) quantization-based local profiling to estimate expert activation with minimal overhead, (2) adaptive layer-aware expert merging to reduce resource consumption while preserving accuracy, and (3) dynamic expert role assignment using an exploration-exploitation strategy to balance tuning and non-tuning experts. Extensive experiments on LLaMA-MoE and DeepSeek-MoE with multiple benchmark datasets demonstrate that FLUX significantly outperforms existing methods, achieving up to 4.75X speedup in time-to-accuracy.

cs.DC↗

arXiv2025

Semi-Clairvoyant Scheduling of Speculative Decoding Requests to Minimize LLM Inference Latency

Speculative decoding accelerates Large Language Model (LLM) inference by employing a small speculative model (SSM) to generate multiple candidate tokens and verify them using the LLM in parallel. This technique has been widely integrated into LLM inference serving systems. However, inference requests typically exhibit uncertain execution time, which poses a significant challenge of efficiently scheduling requests in these systems. Existing work estimates execution time based solely on predicted output length, which could be inaccurate because execution time depends on both output length and token acceptance rate of verification by the LLM. In this paper, we propose a semi-clairvoyant request scheduling algorithm called Least-Attained/Perceived-Service for Speculative Decoding (LAPS-SD). Given a number of inference requests, LAPS-SD can effectively minimize average inference latency by adaptively scheduling requests according to their features during decoding. When the token acceptance rate is dynamic and execution time is difficult to estimate, LAPS-SD maintains multiple priority queues and allows request execution preemption across different queues. Once the token acceptance rate becomes stable, LAPS-SD can accurately estimate the execution time and schedule requests accordingly. Extensive experiments show that LAPS-SD reduces inference latency by approximately 39\% compared to state-of-the-art scheduling methods.

cs.CL↗

arXiv2025

SPIN: Accelerating Large Language Model Inference with Heterogeneous Speculative Models

Speculative decoding has been shown as an effective way to accelerate Large Language Model (LLM) inference by using a Small Speculative Model (SSM) to generate candidate tokens in a so-called speculation phase, which are subsequently verified by the LLM in a verification phase. However, current state-of-the-art speculative decoding approaches have three key limitations: handling requests with varying difficulty using homogeneous SSMs, lack of robust support for batch processing, and insufficient holistic optimization for both speculation and verification phases. In this paper, we introduce SPIN, an efficient LLM inference serving system based on speculative decoding, designed to address these challenges through three main innovations. First, SPIN improves token speculation by using multiple heterogeneous SSMs, with a learning-based algorithm for SSM selection that operates without prior knowledge of request difficulty. Second, SPIN employs a request decomposition method to minimize batching overhead during LLM verification. Finally, SPIN orchestrates speculation and verification phases by pipelining their executions on GPUs to achieve further acceleration. Experimental results demonstrate that SPIN significantly outperforms state-of-the-art methods, achieving a performance increase of approximately 2.28X.

cs.DC↗

arXiv2024

Communication-Efficient Sparsely-Activated Model Training via Sequence Migration and Token Condensation

Mixture-of-Experts (MoE) is an emerging technique for scaling large models with sparse activation. MoE models are typically trained in a distributed manner with an expert parallelism scheme, where experts in each MoE layer are distributed across multiple GPUs. However, the default expert parallelism suffers from the heavy network burden due to the all-to-all intermediate data exchange among GPUs before and after the expert run. Some existing works have proposed to reduce intermediate data exchanges by transferring experts to reduce the network loads, however, which would decrease parallelism level of expert execution and make computation inefficient. The weaknesses of existing works motivate us to explore whether it is possible to reduce inter-GPU traffic while maintaining a high degree of expert parallelism. This paper gives a positive response by presenting Luffy, a communication-efficient distributed MoE training system with two new techniques. First, Luffy migrates sequences among GPUs to hide heavy token pulling paths within GPUs and avoid copying experts over GPUs. Second, we propose token condensation that identifies similar tokens and then eliminates redundant transmissions. We implement Luffy based on PyTorch and evaluate its performance on a testbed of 16 V100 GPUs. Luffy system can achieve a speedup of up to 2.73x compared to state-of-the-art MoE training systems.

cs.DC↗

arXiv2023

DGC: Training Dynamic Graphs with Spatio-Temporal Non-Uniformity using Graph Partitioning by Chunks

Dynamic Graph Neural Network (DGNN) has shown a strong capability of learning dynamic graphs by exploiting both spatial and temporal features. Although DGNN has recently received considerable attention by AI community and various DGNN models have been proposed, building a distributed system for efficient DGNN training is still challenging. It has been well recognized that how to partition the dynamic graph and assign workloads to multiple GPUs plays a critical role in training acceleration. Existing works partition a dynamic graph into snapshots or temporal sequences, which only work well when the graph has uniform spatio-temporal structures. However, dynamic graphs in practice are not uniformly structured, with some snapshots being very dense while others are sparse. To address this issue, we propose DGC, a distributed DGNN training system that achieves a 1.25x - 7.52x speedup over the state-of-the-art in our testbed. DGC's success stems from a new graph partitioning method that partitions dynamic graphs into chunks, which are essentially subgraphs with modest training workloads and few inter connections. This partitioning algorithm is based on graph coarsening, which can run very fast on large graphs. In addition, DGC has a highly efficient run-time, powered by the proposed chunk fusion and adaptive stale aggregation techniques. Extensive experimental results on 3 typical DGNN models and 4 popular dynamic graph datasets are presented to show the effectiveness of DGC.

cs.DC↗

arXiv2021

FedGraph: Federated Graph Learning with Intelligent Sampling

Federated learning has attracted much research attention due to its privacy protection in distributed machine learning. However, existing work of federated learning mainly focuses on Convolutional Neural Network (CNN), which cannot efficiently handle graph data that are popular in many applications. Graph Convolutional Network (GCN) has been proposed as one of the most promising techniques for graph learning, but its federated setting has been seldom explored. In this paper, we propose FedGraph for federated graph learning among multiple computing clients, each of which holds a subgraph. FedGraph provides strong graph learning capability across clients by addressing two unique challenges. First, traditional GCN training needs feature data sharing among clients, leading to risk of privacy leakage. FedGraph solves this issue using a novel cross-client convolution operation. The second challenge is high GCN training overhead incurred by large graph size. We propose an intelligent graph sampling algorithm based on deep reinforcement learning, which can automatically converge to the optimal sampling policies that balance training speed and accuracy. We implement FedGraph based on PyTorch and deploy it on a testbed for performance evaluation. The experimental results of four popular datasets demonstrate that FedGraph significantly outperforms existing work by enabling faster convergence to higher accuracy.

cs.LG↗