arXiv subjects

Xuanzhe Liu

Publications and source records attributed to Xuanzhe Liu.

At least 37 records · Page 2Linked to original sources

arXiv2026

A First Look at Bugs in LLM Inference Engines

Large language model-specific inference engines (in short as \emph{LLM inference engines}) have become a fundamental component of modern AI infrastructure, enabling the deployment of LLM-powered applications (LLM apps) across cloud and local devices. Despite their critical role, LLM inference engines are prone to bugs due to the immense resource demands of LLMs and the complexities of cross-platform compatibility. However, a systematic understanding of these bugs remains lacking. To bridge this gap, we present the first empirical study on bugs in LLM inference engines. We mine official repositories of 5 widely adopted LLM inference engines, constructing a comprehensive dataset of 929 real-world bugs. Through a rigorous open coding process, we analyze these bugs to uncover their symptoms, root causes, commonality, fix effort, fix strategies, and temporal evolution. Our findings reveal six bug symptom types and a taxonomy of 28 root causes, shedding light on the key challenges in bug detection and location within LLM inference engines. Based on these insights, we propose a series of actionable implications for researchers, inference engine vendors, and LLM app developers, along with general guidelines for developing LLM inference engines.

cs.SE↗

arXiv2025

BAMAS: Structuring Budget-Aware Multi-Agent Systems

Large language model (LLM)-based multi-agent systems have emerged as a powerful paradigm for enabling autonomous agents to solve complex tasks. As these systems scale in complexity, cost becomes an important consideration for practical deployment. However, existing work rarely addresses how to structure multi-agent systems under explicit budget constraints. In this paper, we propose BAMAS, a novel approach for building multi-agent systems with budget awareness. BAMAS first selects an optimal set of LLMs by formulating and solving an Integer Linear Programming problem that balances performance and cost. It then determines how these LLMs should collaborate by leveraging a reinforcement learning-based method to select the interaction topology. Finally, the system is instantiated and executed based on the selected agents and their collaboration topology. We evaluate BAMAS on three representative tasks and compare it with state-of-the-art agent construction methods. Results show that BAMAS achieves comparable performance while reducing cost by up to 86%.

cs.MA↗

arXiv2025

MegaScale-MoE: Large-Scale Communication-Efficient Training of Mixture-of-Experts Models in Production

We present MegaScale-MoE, a production system tailored for the efficient training of large-scale mixture-of-experts (MoE) models. MoE emerges as a promising architecture to scale large language models (LLMs) to unprecedented sizes, thereby enhancing model performance. However, existing MoE training systems experience a degradation in training efficiency, exacerbated by the escalating scale of MoE models and the continuous evolution of hardware. Recognizing the pivotal role of efficient communication in enhancing MoE training, MegaScale-MoE customizes communication-efficient parallelism strategies for attention and FFNs in each MoE layer and adopts a holistic approach to overlap communication with computation at both inter- and intra-operator levels. Additionally, MegaScale-MoE applies communication compression with adjusted communication patterns to lower precision, further improving training efficiency. When training a 352B MoE model on 1,440 NVIDIA Hopper GPUs, MegaScale-MoE achieves a training throughput of 1.41M tokens/s, improving the efficiency by 1.88$\times$ compared to Megatron-LM. We share our operational experience in accelerating MoE training and hope that by offering our insights in system design, this work will motivate future research in MoE systems.

cs.LG↗

arXiv2025

Elastic On-Device LLM Service

On-device Large Language Models (LLMs) are transforming mobile AI, catalyzing applications like UI automation without privacy concerns. Nowadays the common practice is to deploy a single yet powerful LLM as a general task solver for multiple requests. We identify a key system challenge in this paradigm: current LLMs lack the elasticity to serve requests that have diversified Service-Level Objectives (SLOs) on inference latency. To tackle this, we present \sys, an on-device LLM service that elasticizes both the model and the prompt dimension of a full LLM. It incorporates (1) a one-shot neuron-reordering method, which leverages the intrinsic permutation consistency in transformer models to generate high-quality elasticized sub-models with minimal runtime switching overhead; (2) a dual-head tiny language model, which efficiently and effectively refines the prompt and orchestrates the elastification between model and prompt. We implement such an elastic on-device LLM service on multiple COTS smartphones, and evaluate \sys on both standalone NLP/mobile-agent datasets and end-to-end synthesized traces. On diverse SLOs, \sys outperforms 7 strong baselines in (absolute) accuracy by up to 14.83\% and 10.45\% on average, with <1\% TTFT switching overhead, on-par memory consumption and <100 offline GPU hours.

cs.DC↗

arXiv2025

HydraServe: Minimizing Cold Start Latency for Serverless LLM Serving in Public Clouds

With the proliferation of large language model (LLM) variants, developers are turning to serverless computing for cost-efficient LLM deployment. However, public cloud providers often struggle to provide performance guarantees for serverless LLM serving due to significant cold start latency caused by substantial model sizes and complex runtime dependencies. To address this problem, we present HydraServe, a serverless LLM serving system designed to minimize cold start latency in public clouds. HydraServe proactively distributes models across servers to quickly fetch them, and overlaps cold-start stages within workers to reduce startup latency. Additionally, HydraServe strategically places workers across GPUs to avoid network contention among cold-start instances. To minimize resource consumption during cold starts, HydraServe further introduces pipeline consolidation that can merge groups of workers into individual serving endpoints. Our comprehensive evaluations under diverse settings demonstrate that HydraServe reduces the cold start latency by 1.7$\times$-- 4.7$\times$ and improves service level objective attainment by 1.43$\times$--1.74$\times$ compared to baselines.

cs.DC↗

arXiv2025

TokenLake: A Unified Segment-level Prefix Cache Pool for Fine-grained Elastic Long-Context LLM Serving

Prefix caching is crucial to accelerate multi-turn interactions and requests with shared prefixes. At the cluster level, existing prefix caching systems are tightly coupled with request scheduling to optimize cache efficiency and computation performance together, leading to load imbalance, data redundancy, and memory fragmentation of caching systems across instances. To address these issues, memory pooling is promising to shield the scheduler from the underlying cache management so that it can focus on the computation optimization. However, because existing prefix caching systems only transfer increasingly longer prefix caches between instances, they cannot achieve low-latency memory pooling. To address these problems, we propose a unified segment-level prefix cache pool, TokenLake. It uses a declarative cache interface to expose requests' query tensors, prefix caches, and cache-aware operations to TokenLake for efficient pooling. Powered by this abstraction, TokenLake can manage prefix cache at the segment level with a heavy-hitter-aware load balancing algorithm to achieve better cache load balance, deduplication, and defragmentation. TokenLake also transparently minimizes the communication volume of query tensors and new caches. Based on TokenLake, the scheduler can schedule requests elastically by using existing techniques without considering prefix cache management. Evaluations on real-world workloads show that TokenLake can improve throughput by up to 2.6$\times$ and 2.0$\times$ and boost hit rate by 2.0$\times$ and 2.1$\times$, compared to state-of-the-art cache-aware routing and cache-centric PD-disaggregation solutions, respectively.

cs.DC↗

arXiv2025

MegaScale-Infer: Serving Mixture-of-Experts at Scale with Disaggregated Expert Parallelism

Mixture-of-Experts (MoE) showcases tremendous potential to scale large language models (LLMs) with enhanced performance and reduced computational complexity. However, its sparsely activated architecture shifts feed-forward networks (FFNs) from being compute-intensive to memory-intensive during inference, leading to substantially lower GPU utilization and increased operational costs. We present MegaScale-Infer, an efficient and cost-effective system for serving large-scale MoE models. MegaScale-Infer disaggregates attention and FFN modules within each model layer, enabling independent scaling, tailored parallelism strategies, and heterogeneous deployment for both modules. To fully exploit disaggregation in the presence of MoE's sparsity, MegaScale-Infer introduces ping-pong pipeline parallelism, which partitions a request batch into micro-batches and shuttles them between attention and FFNs for inference. Combined with distinct model parallelism for each module, MegaScale-Infer effectively hides communication overhead and maximizes GPU utilization. To adapt to disaggregated attention and FFN modules and minimize data transmission overhead (e.g., token dispatch), MegaScale-Infer provides a high-performance M2N communication library that eliminates unnecessary GPU-to-CPU data copies, group initialization overhead, and GPU synchronization. Experimental results indicate that MegaScale-Infer achieves up to 1.90x higher per-GPU throughput than state-of-the-art solutions.

cs.DC↗

arXiv2025

WebANNS: Fast and Efficient Approximate Nearest Neighbor Search in Web Browsers

Approximate nearest neighbor search (ANNS) has become vital to modern AI infrastructure, particularly in retrieval-augmented generation (RAG) applications. Numerous in-browser ANNS engines have emerged to seamlessly integrate with popular LLM-based web applications, while addressing privacy protection and challenges of heterogeneous device deployments. However, web browsers present unique challenges for ANNS, including computational limitations, external storage access issues, and memory utilization constraints, which state-of-the-art (SOTA) solutions fail to address comprehensively. We propose WebANNS, a novel ANNS engine specifically designed for web browsers. WebANNS leverages WebAssembly to overcome computational bottlenecks, designs a lazy loading strategy to optimize data retrieval from external storage, and applies a heuristic approach to reduce memory usage. Experiments show that WebANNS is fast and memory efficient, achieving up to $743.8\times$ improvement in 99th percentile query latency over the SOTA engine, while reducing memory usage by up to 39\%. Note that WebANNS decreases query time from 10 seconds to the 10-millisecond range in browsers, making in-browser ANNS practical with user-acceptable latency.

cs.IR↗

arXiv2025

FinMME: Benchmark Dataset for Financial Multi-Modal Reasoning Evaluation

Multimodal Large Language Models (MLLMs) have experienced rapid development in recent years. However, in the financial domain, there is a notable lack of effective and specialized multimodal evaluation datasets. To advance the development of MLLMs in the finance domain, we introduce FinMME, encompassing more than 11,000 high-quality financial research samples across 18 financial domains and 6 asset classes, featuring 10 major chart types and 21 subtypes. We ensure data quality through 20 annotators and carefully designed validation mechanisms. Additionally, we develop FinScore, an evaluation system incorporating hallucination penalties and multi-dimensional capability assessment to provide an unbiased evaluation. Extensive experimental results demonstrate that even state-of-the-art models like GPT-4o exhibit unsatisfactory performance on FinMME, highlighting its challenging nature. The benchmark exhibits high robustness with prediction variations under different prompts remaining below 1%, demonstrating superior reliability compared to existing datasets. Our dataset and evaluation protocol are available at https://huggingface.co/datasets/luojunyu/FinMME and https://github.com/luo-junyu/FinMME.

cs.CL↗

arXiv2025

SeaLLM: Service-Aware and Latency-Optimized Resource Sharing for Large Language Model Inference

Large language models (LLMs) with different architectures and sizes have been developed. Serving each LLM with dedicated GPUs leads to resource waste and service inefficiency due to the varying demand of LLM requests. A common practice is to share multiple LLMs. However, existing sharing systems either do not consider the autoregressive pattern of LLM services, or only focus on improving the throughput, which impairs the sharing performance, especially the serving latency. We present SeaLLM, which enables service-aware and latency-optimized LLM sharing. SeaLLM improves the overall sharing performance by (1) a latency-optimized scheduling algorithm utilizing the characteristics of LLM services, (2) a placement algorithm to determine the placement plan and an adaptive replacement algorithm to decide the replacement interval, and (3) a unified key-value cache to share GPU memory among LLM services efficiently. Our evaluation under real-world traces and LLM services demonstrates that SeaLLM improves the normalized latency by up to $13.60\times$, the tail latency by up to $18.69\times$, and the SLO attainment by up to $3.64\times$ compared to existing solutions.

cs.DC↗

arXiv2025

SCOPE: Performance Testing for Serverless Computing

Serverless computing is a popular cloud computing paradigm that has found widespread adoption across various online workloads. It allows software engineers to develop cloud applications as a set of functions (called serverless functions). However, accurately measuring the performance (i.e., end-to-end response latency) of serverless functions is challenging due to the highly dynamic nature of the environment in which they run. To tackle this problem, a potential solution is to apply checks of performance testing techniques to determine how many repetitions of a given serverless function across a range of inputs are needed to cater to the performance fluctuation. However, the available literature lacks performance testing approaches designed explicitly for serverless computing. In this paper, we propose SCOPE, the first serverless computing-oriented performance testing approach. SCOPE takes into account the unique performance characteristics of serverless functions, such as their short execution durations and on-demand triggering. As such, SCOPE is designed as a fine-grained analysis approach. SCOPE incorporates the accuracy check and the consistency check to obtain the accurate and reliable performance of serverless functions. The evaluation shows that SCOPE provides testing results with 97.25% accuracy, 33.83 percentage points higher than the best currently available technique. Moreover, the superiority of SCOPE over the state-of-the-art holds on all functions that we study.

cs.SE↗

arXiv2024

Fast On-device LLM Inference with NPUs

On-device inference for Large Language Models (LLMs), driven by increasing privacy concerns and advancements of mobile-sized models, has gained significant interest. However, even mobile-sized LLMs (e.g., Gemma-2B) encounter unacceptably high inference latency, often bottlenecked by the prefill stage in tasks like screen UI understanding. We present llm.npu, the first LLM inference system utilizing on-device Neural Processing Unit (NPU) offloading to reduce prefill latency. llm.npu enhances NPU offloading efficiency by re-constructing the prompt and model in three levels: (1) At prompt level, it divides variable-length prompts into multiple fixed-sized chunks while maintaining data dependencies; (2) At tensor level, it identifies and extracts significant outliers to run on the CPU/GPU in parallel with minimal overhead; (3) At block level, it schedules Transformer blocks in an out-of-order manner to the CPU/GPU and NPU based on their hardware affinity and sensitivity to accuracy. Compared to competitive baselines, llm.npu achieves 22.4x faster prefill speed and 30.7$\times$ energy savings on average, and up to 32.8x speedup in an end-to-end real-world application. For the first time, llm.npu achieves more than 1,000 tokens/sec prefilling for a billion-sized model.

cs.AI↗

arXiv2024

LoongServe: Efficiently Serving Long-Context Large Language Models with Elastic Sequence Parallelism

The context window of large language models (LLMs) is rapidly increasing, leading to a huge variance in resource usage between different requests as well as between different phases of the same request. Restricted by static parallelism strategies, existing LLM serving systems cannot efficiently utilize the underlying resources to serve variable-length requests in different phases. To address this problem, we propose a new parallelism paradigm, elastic sequence parallelism (ESP), to elastically adapt to the variance between different requests and phases. Based on ESP, we design and build LoongServe, an LLM serving system that (1) improves computation efficiency by elastically adjusting the degree of parallelism in real-time, (2) improves communication efficiency by reducing key-value cache migration overhead and overlapping partial decoding communication with computation, and (3) improves GPU memory efficiency by reducing key-value cache fragmentation across instances. Our evaluation under diverse real-world datasets shows that LoongServe improves the maximum throughput by up to 3.85$\times$ compared to the chunked prefill and 5.81$\times$ compared to the prefill-decoding disaggregation.

cs.DC↗

arXiv2024

Research on WebAssembly Runtimes: A Survey

WebAssembly (abbreviated as Wasm) was initially introduced for the Web but quickly extended its reach into various domains beyond the Web. To create Wasm applications, developers can compile high-level programming languages into Wasm binaries or manually convert equivalent textual formats into Wasm binaries. Regardless of whether it is utilized within or outside the Web, the execution of Wasm binaries is supported by the Wasm runtime. Such a runtime provides a secure, memory-efficient, and sandboxed execution environment designed explicitly for Wasm applications. This paper provides a comprehensive survey of research on WebAssembly runtimes. It covers 98 articles on WebAssembly runtimes and characterizes existing studies from two different angles, including the "internal" research of Wasm runtimes(Wasm runtime design, testing, and analysis) and the "external" research(applying Wasm runtimes to various domains). This paper also proposes future research directions about WebAssembly runtimes.

cs.SE↗

arXiv2024

Bias Behind the Wheel: Fairness Testing of Autonomous Driving Systems

This paper conducts fairness testing of automated pedestrian detection, a crucial but under-explored issue in autonomous driving systems. We evaluate eight state-of-the-art deep learning-based pedestrian detectors across demographic groups on large-scale real-world datasets. To enable thorough fairness testing, we provide extensive annotations for the datasets, resulting in 8,311 images with 16,070 gender labels, 20,115 age labels, and 3,513 skin tone labels. Our findings reveal significant fairness issues, particularly related to age. The proportion of undetected children is 20.14% higher compared to adults. Furthermore, we explore how various driving scenarios affect the fairness of pedestrian detectors. We find that pedestrian detectors demonstrate significant gender biases during night time, potentially exacerbating the prevalent societal issue of female safety concerns during nighttime out. Moreover, we observe that pedestrian detectors can demonstrate both enhanced fairness and superior performance under specific driving conditions, which challenges the fairness-performance trade-off theory widely acknowledged in the fairness literature. We publicly release the code, data, and results to support future research on fairness in autonomous driving.

cs.CY↗

arXiv2024

Fast Distributed Inference Serving for Large Language Models

Large language models (LLMs) power a new generation of interactive AI applications exemplified by ChatGPT. The interactive nature of these applications demands low latency for LLM inference. Existing LLM serving systems use run-to-completion processing for inference jobs, which suffers from head-of-line blocking and long latency. We present FastServe, a distributed inference serving system for LLMs. FastServe exploits the autoregressive pattern of LLM inference to enable preemption at the granularity of each output token. FastServe uses preemptive scheduling to minimize latency with a novel skip-join Multi-Level Feedback Queue scheduler. Based on the new semi-information-agnostic setting of LLM inference, the scheduler leverages the input length information to assign an appropriate initial queue for each arrival job to join. The higher priority queues than the joined queue are skipped to reduce demotions. We design an efficient GPU memory management mechanism that proactively offloads and uploads intermediate state between GPU memory and host memory for LLM inference. We build a system prototype of FastServe and experimental results show that compared to the state-of-the-art solution vLLM, FastServe improves the throughput by up to 31.4x and 17.9x under the same average and tail latency requirements, respectively.

cs.LG↗

arXiv2024

A Survey of Resource-efficient LLM and Multimodal Foundation Models

Large foundation models, including large language models (LLMs), vision transformers (ViTs), diffusion, and LLM-based multimodal models, are revolutionizing the entire machine learning lifecycle, from training to deployment. However, the substantial advancements in versatility and performance these models offer come at a significant cost in terms of hardware resources. To support the growth of these large models in a scalable and environmentally sustainable way, there has been a considerable focus on developing resource-efficient strategies. This survey delves into the critical importance of such research, examining both algorithmic and systemic aspects. It offers a comprehensive analysis and valuable insights gleaned from existing literature, encompassing a broad array of topics from cutting-edge model architectures and training/serving algorithms to practical system designs and implementations. The goal of this survey is to provide an overarching understanding of how current approaches are tackling the resource challenges posed by large foundation models and to potentially inspire future breakthroughs in this field.

cs.LG↗

arXiv2024

Anatomizing Deep Learning Inference in Web Browsers

Web applications have increasingly adopted Deep Learning (DL) through in-browser inference, wherein DL inference performs directly within Web browsers. The actual performance of in-browser inference and its impacts on the quality of experience (QoE) remain unexplored, and urgently require new QoE measurements beyond traditional ones, e.g., mainly focusing on page load time. To bridge this gap, we make the first comprehensive performance measurement of in-browser inference to date. Our approach proposes new metrics to measure in-browser inference: responsiveness, smoothness, and inference accuracy. Our extensive analysis involves 9 representative DL models across Web browsers of 50 popular PC devices and 20 mobile devices. The results reveal that in-browser inference exhibits a substantial latency gap, averaging 16.9 times slower on CPU and 4.9 times slower on GPU compared to native inference on PC devices. The gap on mobile CPU and mobile GPU is 15.8 times and 7.8 times, respectively. Furthermore, we identify contributing factors to such latency gap, including underutilized hardware instruction sets, inherent overhead in the runtime environment, resource contention within the browser, and inefficiencies in software libraries and GPU abstractions. Additionally, in-browser inference imposes significant memory demands, at times exceeding 334.6 times the size of the DL models themselves, partly attributable to suboptimal memory management. We also observe that in-browser inference leads to a significant 67.2% increase in the time it takes for GUI components to render within Web browsers, significantly affecting the overall user QoE of Web applications reliant on this technology

cs.LG↗