arXiv subjects

Shiyou Qian

Publications and source records attributed to Shiyou Qian.

18 recordsLinked to original sources

arXiv2026

Self-Correcting Long-Horizon Search Agents via Tree-Structured Memory

Large language model (LLM)-based search agents answer questions through multi-step interactions with external environments. However, providing complete execution trajectories to the LLM causes unbounded context growth and introduces noise. Existing compression methods reduce context at the cost of important details and often replace erroneous facts without repairing downstream reasoning derived from them. To address this problem, we propose ReTree, a self-correcting tree-structured memory mechanism for search agents. ReTree constructs a bounded per-step reasoning context while preserving source-linked evidence. It models search as an evidence tree whose nodes store bounded summaries, evidence, and revision histories. When newly retrieved evidence contradicts an earlier claim, ReTree traces back to the node where the claim was introduced, replaces outdated evidence, regenerates summaries, prunes affected branches, and resumes search. Source-grounded evidence provenance supports reliable conflict localization and keeps final claims traceable to retrieved passages. Experiments on four public question-answering and search benchmarks show that ReTree consistently outperforms Full-Trajectory ReAct, improving answer accuracy by up to 25.6 percentage points (pp); the average maximum per-step reasoning context of Full-Trajectory ReAct is $1.27$--$1.51\times$ that of ReTree. These results establish ReTree as an effective self-correcting memory abstraction for long-horizon search.

cs.AI↗

arXiv2026

PipeWeave: Synergizing Analytical and Learning Models for Unified GPU Performance Prediction

The rapid expansion of Transformer-based large language models has dramatically increased the need for high-performance GPUs. As a result, there is growing demand for fast, accurate, and widely generalizable GPU performance models to support next-generation hardware selection and system-level exploration. However, current data-driven methods are limited, exhibiting poor generalization across hardware and inadequate modeling of complex production-level kernels common in modern inference stacks. To address these issues, we present PipeWeave, a unified GPU modeling framework. This approach first employs an analytical model to quantify a given kernel's demands on the GPU's heterogeneous instruction pipelines. These analytical features are then fed into a machine learning (ML) model to capture complex cross-pipeline interactions and resource dependencies, enabling high-fidelity performance prediction. Our evaluation across 11 GPU types from four generations of major architectures on two widely-used serving systems demonstrates that PipeWeave delivers high fidelity and strong generalizability. It achieves accurate predictions, with only 6.1% average error at the kernel level and 8.5% for end-to-end inference -- reducing the error of state-of-the-art methods by 6.7x and 4.4x, respectively. We also demonstrate PipeWeave's value "beyond simulation" by utilizing its performance ceiling to diagnose implementation shortcomings and guide the optimization of a production fused MoE Triton kernel, achieving up to 1.7x speedup. Code is available https://github.com/zksainx/pipeweave.

cs.PF↗

arXiv2026

WaveTune: Wave-aware Bilinear Modeling for Efficient GPU Kernel Auto-tuning

The rapid adoption of Large Language Models (LLMs) has made GPU inference efficiency an increasingly critical system concern. The runtime of LLM workloads is largely dominated by tile-based kernels, particularly General Matrix Multiplications (GEMMs). Although these kernels are highly optimized, their performance remains sensitive to a large space of runtime parameters, such as tile sizes and pipeline stages. The interaction between these parameters and hardware resources leads to a non-convex optimization landscape. Existing approaches to parameter configuration -- including search-based auto-tuning, heuristic rules, and learned cost models -- face a fundamental trade-off between performance optimality and runtime efficiency. In this paper, we present WaveTune, a wave-aware framework for runtime kernel auto-tuning. First, we introduce a unified mapping method to handle input diversity and decompose the configuration space to manage high dimensionality. Second, we develop an analytical wave-aware bilinear model that accurately predicts kernel latency. Third, we design a sparse sampling scheme based on wave structures and a lightweight dual-table retrieval mechanism to minimize runtime overhead. As a result, WaveTune enables precise and efficient runtime configuration for GPU kernels. Across three representative kernels and five GPU architectures, WaveTune consistently achieves near-optimal kernel performance, delivering up to 1.83x kernel-level speedup and up to 1.33x end-to-end TTFT reduction, while reducing runtime decision overhead by five orders of magnitude compared to exhaustive search. These results demonstrate that WaveTune effectively eliminates the traditional trade-off between configuration latency and execution optimality, providing a practical and robust solution for high-performance LLM inference.

cs.PF↗

arXiv2026

Hestia: Hyperthread-Level Scheduling for Cloud Microservices with Interference-Aware Attention

Modern cloud servers routinely co-locate multiple latency-sensitive microservice instances to improve resource efficiency. However, the diversity of microservice behaviors, coupled with mutual performance interference under simultaneous multithreading (SMT), makes large-scale placement increasingly complex. Existing interference aware schedulers and isolation techniques rely on coarse core-level profiling or static resource partitioning, leaving asymmetric hyperthread-level heterogeneity and SMT contention dynamics largely unmodeled. We present Hestia, a hyperthread-level, interference-aware scheduling framework powered by self-attention. Through an extensive analysis of production traces encompassing 32,408 instances across 3,132 servers, we identify two dominant contention patterns -- sharing-core (SC) and sharing-socket (SS) -- and reveal strong asymmetry in their impact. Guided by these insights, Hestia incorporates (1) a self-attention-based CPU usage predictor that models SC/SS contention and hardware heterogeneity, and (2) an interference scoring model that estimates pairwise contention risks to guide scheduling decisions. We evaluate Hestia through large-scale simulation and a real production deployment. Hestia reduces the 95th-percentile service latency by up to 80\%, lowers overall CPU consumption by 2.3\% under the same workload, and surpasses five state-of-the-art schedulers by up to 30.65\% across diverse contention scenarios.

cs.DC↗

arXiv2026

TaNG: Modeling Packet Classification with TSS-assisted Neural Networks on GPUs

Packet classification is a core function in software-defined networks, and learning-based methods have recently shown significant throughput gains on large-scale rulesets. However, existing learning-based approaches struggle with overlapping rules, leading to incomplete model coverage or excessive rule replication. Their limited GPU integration also hampers performance with large-scale rulesets. To address these issues, we propose TaNG, which utilizes a single neural network trained on multi-dimensional features to ensure complete coverage without duplicating rules. TaNG employs a semi-structured design that combines the neural network model with a tuple space, reducing model complexity. Furthermore, we develop a mechanism based on the semi-structure for rule updates. Finally, we implement a CPU-GPU hybrid streaming framework tailored for learning-based methods, further enhancing throughput. On our GPU-based classification framework with 512k rulesets, TaNG achieves 12.19x and 9.37x higher throughput and 98.84x and 156.98x higher performance stability compared to two state-of-the-art learning methods NuevoMatch and NeuTree, respectively.

cs.NI↗

arXiv2025

DABL: Detecting Semantic Anomalies in Business Processes Using Large Language Models

Detecting anomalies in business processes is crucial for ensuring operational success. While many existing methods rely on statistical frequency to detect anomalies, it's important to note that infrequent behavior doesn't necessarily imply undesirability. To address this challenge, detecting anomalies from a semantic viewpoint proves to be a more effective approach. However, current semantic anomaly detection methods treat a trace (i.e., process instance) as multiple event pairs, disrupting long-distance dependencies. In this paper, we introduce DABL, a novel approach for detecting semantic anomalies in business processes using large language models (LLMs). We collect 143,137 real-world process models from various domains. By generating normal traces through the playout of these process models and simulating both ordering and exclusion anomalies, we fine-tune Llama 2 using the resulting log. Through extensive experiments, we demonstrate that DABL surpasses existing state-of-the-art semantic anomaly detection methods in terms of both generalization ability and learning of given processes. Users can directly apply DABL to detect semantic anomalies in their own datasets without the need for additional training. Furthermore, DABL offers the capability to interpret the causes of anomalies in natural language, providing valuable insights into the detected anomalies.

cs.CL↗

arXiv2025

LLM-SrcLog: Towards Proactive and Unified Log Template Extraction via Large Language Models

Log parsing transforms raw logs into structured templates containing constants and variables. It underpins anomaly detection, failure diagnosis, and other AIOps tasks. Current parsers are mostly reactive and log-centric. They only infer templates from logs, mostly overlooking the source code. This restricts their capacity to grasp dynamic log structures or adjust to evolving systems. Moreover, per-log LLM inference is too costly for practical deployment. In this paper, we propose LLM-SrcLog, a proactive and unified framework for log template parsing. It extracts templates directly from source code prior to deployment and supplements them with data-driven parsing for logs without available code. LLM-SrcLog integrates a cross-function static code analyzer to reconstruct meaningful logging contexts, an LLM-based white-box template extractor with post-processing to distinguish constants from variables, and a black-box template extractor that incorporates data-driven clustering for remaining unmatched logs. Experiments on two public benchmarks (Hadoop and Zookeeper) and a large-scale industrial system (Sunfire-Compute) show that, compared to two LLM-based baselines, LLM-SrcLog improves average F1-score by 2-17% and 8-35%. Meanwhile, its online parsing latency is comparable to data-driven methods and about 1,000 times faster than per-log LLM parsing. LLM-SrcLog achieves a near-ideal balance between speed and accuracy. Finally, we further validate the effectiveness of LLM-SrcLog through practical case studies in a real-world production environment.

cs.SE↗

arXiv2025

PAST: A Primary-Auxiliary Spatio-Temporal Network for Traffic Time Series Imputation

Traffic time series imputation is crucial for the safety and reliability of intelligent transportation systems, while diverse types of missing data, including random, fiber, and block missing make the imputation task challenging. Existing models often focus on disentangling and separately modeling spatial and temporal patterns based on relationships between data points. However, these approaches struggle to adapt to the random missing positions, and fail to learn long-term and large-scale dependencies, which are essential in extensive missing conditions. In this paper, patterns are categorized into two types to handle various missing data conditions: primary patterns, which originate from internal relationships between data points, and auxiliary patterns, influenced by external factors like timestamps and node attributes. Accordingly, we propose the Primary-Auxiliary Spatio-Temporal network (PAST). It comprises a graph-integrated module (GIM) and a cross-gated module (CGM). GIM captures primary patterns via dynamic graphs with interval-aware dropout and multi-order convolutions, and CGM extracts auxiliary patterns through bidirectional gating on embedded external features. The two modules interact via shared hidden vectors and are trained under an ensemble self-supervised framework. Experiments on three datasets under 27 missing data conditions demonstrate that the imputation accuracy of PAST outperforms seven state-of-the-art baselines by up to 26.2% in RMSE and 31.6% in MAE.

cs.LG↗

arXiv2025

EDGC: Entropy-driven Dynamic Gradient Compression for Efficient LLM Training

Training large language models (LLMs) poses significant challenges regarding computational resources and memory capacity. Although distributed training techniques help mitigate these issues, they still suffer from considerable communication overhead. Existing approaches primarily rely on static gradient compression to enhance communication efficiency; however, these methods neglect the dynamic nature of evolving gradients during training, leading to performance degradation. Accelerating LLM training via compression without sacrificing performance remains a challenge. In this paper, we propose an entropy-driven dynamic gradient compression framework called EDGC. The core concept is to adjust the compression rate during LLM training based on the evolving trends of gradient entropy, taking into account both compression efficiency and error. EDGC consists of three key components.First, it employs a down-sampling method to efficiently estimate gradient entropy, reducing computation overhead. Second, it establishes a theoretical model linking compression rate with gradient entropy, enabling more informed compression decisions. Lastly, a window-based adjustment mechanism dynamically adapts the compression rate across pipeline stages, improving communication efficiency and maintaining model performance. We implemented EDGC on a 32-NVIDIA-V100 cluster and a 64-NVIDIA-H100 cluster to train GPT2-2.5B and GPT2-12.1B, respectively. The results show that EDGC significantly reduces communication latency and training time by up to 46.45% and 16.13% while preserving LLM accuracy.

cs.LG↗

arXiv2025

GFS: A Preemption-aware Scheduling Framework for GPU Clusters with Predictive Spot Instance Management

The surge in large language models (LLMs) has fundamentally reshaped the landscape of GPU usage patterns, creating an urgent need for more efficient management strategies. While cloud providers employ spot instances to reduce costs for low-priority (LP) tasks, existing schedulers still grapple with high eviction rates and lengthy queuing times. To address these limitations, we present GFS, a novel preemptive scheduling framework that enhances service-level objective (SLO) compliance for high-priority (HP) tasks while minimizing preemptions to LP tasks. Firstly, GFS utilizes a lightweight forecasting model that predicts GPU demand among different tenants, enabling proactive resource management. Secondly, GFS employs a dynamic allocation mechanism to adjust the spot quota for LP tasks with guaranteed durations. Lastly, GFS incorporates a preemptive scheduling policy that prioritizes HP tasks while minimizing the impact on LP tasks. We demonstrate the effectiveness of GFS through both real-world implementation and simulations. The results show that GFS reduces eviction rates by 33.0\%, and cuts queuing delays by 44.1\% for LP tasks. Furthermore, GFS enhances the GPU allocation rate by up to 22.8\% in real production clusters. In a production cluster of more than 10,000 GPUs, GFS yields roughly \$459,715 in monthly benefits.

cs.DC↗

arXiv2025

Atys: An Efficient Profiling Framework for Identifying Hotspot Functions in Large-scale Cloud Microservices

To handle the high volume of requests, large-scale services are comprised of thousands of instances deployed in clouds. These services utilize diverse programming languages and are distributed across various nodes as encapsulated containers. Given their vast scale, even minor performance enhancements can lead to significant cost reductions. In this paper, we introduce Atys1, an efficient profiling framework specifically designed to identify hotspot functions within large-scale distributed services. Atys presents four key features. First, it implements a language-agnostic adaptation mechanism for multilingual microservices. Second, a two-level aggregation method is introduced to provide a comprehensive overview of flamegraphs. Third, we propose a function selective pruning (FSP) strategy to enhance the efficiency of aggregating profiling results. Finally, we develop a frequency dynamic adjustment (FDA) scheme that dynamically modifies sampling frequency based on service status, effectively minimizing profiling cost while maintaining accuracy. Cluster-scale experiments on two benchmarks show that the FSP strategy achieves a 6.8% reduction in time with a mere 0.58% mean average percentage error (MAPE) in stack traces aggregation. Additionally, the FDA scheme ensures that the mean squared error (MSE) remains on par with that at high sampling rates, while achieving an 87.6% reduction in cost.

cs.PF↗

arXiv2025

LKD-KGC: Domain-Specific KG Construction via LLM-driven Knowledge Dependency Parsing

Knowledge Graphs (KGs) structure real-world entities and their relationships into triples, enhancing machine reasoning for various tasks. While domain-specific KGs offer substantial benefits, their manual construction is often inefficient and requires specialized knowledge. Recent approaches for knowledge graph construction (KGC) based on large language models (LLMs), such as schema-guided KGC and reference knowledge integration, have proven efficient. However, these methods are constrained by their reliance on manually defined schema, single-document processing, and public-domain references, making them less effective for domain-specific corpora that exhibit complex knowledge dependencies and specificity, as well as limited reference knowledge. To address these challenges, we propose LKD-KGC, a novel framework for unsupervised domain-specific KG construction. LKD-KGC autonomously analyzes document repositories to infer knowledge dependencies, determines optimal processing sequences via LLM driven prioritization, and autoregressively generates entity schema by integrating hierarchical inter-document contexts. This schema guides the unsupervised extraction of entities and relationships, eliminating reliance on predefined structures or external knowledge. Extensive experiments show that compared with state-of-the-art baselines, LKD-KGC generally achieves improvements of 10% to 20% in both precision and recall rate, demonstrating its potential in constructing high-quality domain-specific KGs.

cs.CL↗

arXiv2025

LogLLM: Log-based Anomaly Detection Using Large Language Models

Software systems often record important runtime information in logs to help with troubleshooting. Log-based anomaly detection has become a key research area that aims to identify system issues through log data, ultimately enhancing the reliability of software systems. Traditional deep learning methods often struggle to capture the semantic information embedded in log data, which is typically organized in natural language. In this paper, we propose LogLLM, a log-based anomaly detection framework that leverages large language models (LLMs). LogLLM employs BERT for extracting semantic vectors from log messages, while utilizing Llama, a transformer decoder-based model, for classifying log sequences. Additionally, we introduce a projector to align the vector representation spaces of BERT and Llama, ensuring a cohesive understanding of log semantics. Unlike conventional methods that require log parsers to extract templates, LogLLM preprocesses log messages with regular expressions, streamlining the entire process. Our framework is trained through a novel three-stage procedure designed to enhance performance and adaptability. Experimental results across four public datasets demonstrate that LogLLM outperforms state-of-the-art methods. Even when handling unstable logs, it effectively captures the semantic meaning of log messages and detects anomalies accurately.

cs.SE↗

arXiv2024

Large Language Models for Constructing and Optimizing Machine Learning Workflows: A Survey

Building effective machine learning (ML) workflows to address complex tasks is a primary focus of the Automatic ML (AutoML) community and a critical step toward achieving artificial general intelligence (AGI). Recently, the integration of Large Language Models (LLMs) into ML workflows has shown great potential for automating and enhancing various stages of the ML pipeline. This survey provides a comprehensive and up-to-date review of recent advancements in using LLMs to construct and optimize ML workflows, focusing on key components encompassing data and feature engineering, model selection and hyperparameter optimization, and workflow evaluation. We discuss both the advantages and limitations of LLM-driven approaches, emphasizing their capacity to streamline and enhance ML workflow modeling process through language understanding, reasoning, interaction, and generation. Finally, we highlight open challenges and propose future research directions to advance the effective application of LLMs in ML workflows.

cs.LG↗

arXiv2024

Mitigating Interference of Microservices with a Scoring Mechanism in Large-scale Clusters

Co-locating latency-critical services (LCSs) and best-effort jobs (BEJs) constitute the principal approach for enhancing resource utilization in production. Nevertheless, the co-location practice hurts the performance of LCSs due to resource competition, even when employing isolation technology. Through an extensive analysis of voluminous real trace data derived from two production clusters, we observe that BEJs typically exhibit periodic execution patterns and serve as the primary sources of interference to LCSs. Furthermore, despite occupying the same level of resource consumption, the diverse compositions of BEJs can result in varying degrees of interference on LCSs. Subsequently, we propose PISM, a proactive Performance Interference Scoring and Mitigating framework for LCSs through the optimization of BEJ scheduling. Firstly, PISM adopts a data-driven approach to establish a characterization and classification methodology for BEJs. Secondly, PISM models the relationship between the composition of BEJs on servers and the response time (RT) of LCSs. Thirdly, PISM establishes an interference scoring mechanism in terms of RT, which serves as the foundation for BEJ scheduling. We assess the effectiveness of PISM on a small-scale cluster and through extensive data-driven simulations. The experiment results demonstrate that PISM can reduce cluster interference by up to 41.5%, and improve the throughput of long-tail LCSs by 76.4%.

cs.DC↗

arXiv2024

Humas: A Heterogeneity- and Upgrade-aware Microservice Auto-scaling Framework in Large-scale Data Centers

An effective auto-scaling framework is essential for microservices to ensure performance stability and resource efficiency under dynamic workloads. As revealed by many prior studies, the key to efficient auto-scaling lies in accurately learning performance patterns, i.e., the relationship between performance metrics and workloads in data-driven schemes. However, we notice that there are two significant challenges in characterizing performance patterns for large-scale microservices. Firstly, diverse microservices demonstrate varying sensitivities to heterogeneous machines, causing difficulty in quantifying the performance difference in a fixed manner. Secondly, frequent version upgrades of microservices result in uncertain changes in performance patterns, known as pattern drifts, leading to imprecise resource capacity estimation issues. To address these challenges, we propose Humas, a heterogeneity- and upgrade-aware auto-scaling framework for large-scale microservices. Firstly, Humas quantifies the difference in resource efficiency among heterogeneous machines for various microservices online and normalizes their resources in standard units. Additionally, Humas develops a least squares density-difference (LSDD) based algorithm to identify pattern drifts caused by upgrades. Lastly, Humas generates capacity adjustment plans for microservices based on the latest performance patterns and predicted workloads. The experiment results conducted on 50 real microservices with over 11,000 containers demonstrate that Humas improves resource efficiency and performance stability by approximately 30.4% and 48.0%, respectively, compared to state-of-the-art approaches.

cs.DC↗

arXiv2024

MOPAR: A Model Partitioning Framework for Deep Learning Inference Services on Serverless Platforms

With its elastic power and a pay-as-you-go cost model, the deployment of deep learning inference services (DLISs) on serverless platforms is emerging as a prevalent trend. However, the varying resource requirements of different layers in DL models hinder resource utilization and increase costs, when DLISs are deployed as a single function on serverless platforms. To tackle this problem, we propose a model partitioning framework called MOPAR. This work is based on the two resource usage patterns of DLISs: global differences and local similarity, due to the presence of resource dominant (RD) operators and layer stacking. Considering these patterns, MOPAR adopts a hybrid approach that initially divides the DL model vertically into multiple slices composed of similar layers to improve resource efficiency. Slices containing RD operators are further partitioned into multiple sub-slices, enabling parallel optimization to reduce inference latency. Moreover, MOPAR comprehensively employs data compression and share-memory techniques to offset the additional time introduced by communication between slices. We implement a prototype of MOPAR and evaluate its efficacy using four categories of 12 DL models on OpenFaaS and AWS Lambda. The experiment results show that MOPAR can improve the resource efficiency of DLISs by 27.62\% on average, while reducing latency by about 5.52\%. Furthermore, based on Lambda's pricing, the cost of running DLISs is reduced by about 2.58 $\times$ using MOPAR.

cs.DC↗

arXiv2023

GDOD: Effective Gradient Descent using Orthogonal Decomposition for Multi-Task Learning

Multi-task learning (MTL) aims at solving multiple related tasks simultaneously and has experienced rapid growth in recent years. However, MTL models often suffer from performance degeneration with negative transfer due to learning several tasks simultaneously. Some related work attributed the source of the problem is the conflicting gradients. In this case, it is needed to select useful gradient updates for all tasks carefully. To this end, we propose a novel optimization approach for MTL, named GDOD, which manipulates gradients of each task using an orthogonal basis decomposed from the span of all task gradients. GDOD decomposes gradients into task-shared and task-conflict components explicitly and adopts a general update rule for avoiding interference across all task gradients. This allows guiding the update directions depending on the task-shared components. Moreover, we prove the convergence of GDOD theoretically under both convex and non-convex assumptions. Experiment results on several multi-task datasets not only demonstrate the significant improvement of GDOD performed to existing MTL models but also prove that our algorithm outperforms state-of-the-art optimization methods in terms of AUC and Logloss metrics.

cs.LG↗