arXiv subjects

Yinghao Yu

Publications and source records attributed to Yinghao Yu.

At least 19 recordsLinked to original sources

arXiv2026

From LLM Inference to Agentic Workloads: Characterization and Implications for Serving Systems

Agentic applications are shifting AI serving from isolated model inference to long-running workloads in which LLMs coordinate tools, environments, and persistent state. However, the system behavior of these workloads---where latency, cost, and bottlenecks arise---remains poorly characterized, leaving serving systems to rely on assumptions built for conventional inference. We present AgentSysBench, a benchmark suite and measurement toolkit with ten representative agentic applications and unified systems-level instrumentation. Across controlled deployments and production traces, we identify six properties that distinguish agentic workloads from conventional LLM serving: (1) execution is heavyweight and stateful, with non-LLM components dominating latency in 5 of 10 applications and sandbox working-set memory peaking at 28 GB per session; (2) applications compose components with heterogeneous resource affinity---GPU-bound inference, memory-bound retrieval, CPU-bound sandboxes---whose task latencies diverge by up to 32x; (3) bottlenecks shift across requests, models, and deployments; (4) production sessions hold state idle for minutes to hours between active steps; (5) a control-plane tax---auxiliary LLM calls and context overhead from tool schemas and observations---crowds out productive compute and context; and (6) production traces from three applications reveal heavy cross-request redundancy in search queries and web fetches, exposing a large caching opportunity. Four design explorations demonstrate that these findings are actionable: task-aware serving reduces latency by 29--40%, communication-aware placement by up to 4.5x, state offloading reduces memory usage by 4.6x, and tool-result caching removes 35.2% of redundant search calls and saves 19.3% of aggregate search latency.

cs.OS↗

arXiv2026

Bifrost: Empowering Pretrained Language Model with Fallibility Representation for Log-Based Fault Diagnosis

Log-based fault diagnosis is crucial for runtime debugging and maintenance. Existing fault diagnosis methods use language models pre-trained on natural language (PLMs) for log representation. However, system faults are reflected in the multi-level structure of system logs. PLMs pre-trained on natural language struggle to comprehensively capture multi-level fault information, failing to meet the requirements of fault diagnosis. We refer to this information as fallibility representations. To address this problem, we propose a novel log representation learning method, Bifrost. It draws inspiration from the log analysis experience of Site Reliability Engineers and meticulously designs strategies based on self-supervised contrastive learning to learn the fallibility representations of logs. Across three public systems and one industrial ML-as-a-Service system, the log representations produced by Bifrost outperform existing PLMs by average margins of 9.83% in F1 for anomaly detection, 18.28% in HR@k for root cause localization, and 20.88% in Macro-F1 for fault identification.

cs.SE↗

arXiv2026

Are LLM-Generated GPU Kernels Production-Ready? A Trace-Driven Benchmark and Optimization Agent

Existing GPU kernel generation benchmarks draw problems from synthetic or curated sources that diverge from deployed workloads. We present Atrex-Bench, a benchmark whose 30 operators and 440 shapes are sampled directly from full-cluster production inference traces of compute-limited, memory-rich GPUs. Each problem carries an importance weight derived from its share of observed GPU time, weighted by application card-hours and computed separately for the serving phases in which it runs, together with a per-problem roofline ceiling, so the aggregate score emphasizes the kernels that consume the most serving time. Evaluating six frontier coding agents on Atrex-Bench shows that even the best vanilla model reaches only ${\sim}10\%$ of the hardware roofline on production operators; and correctness alone overstates capability, since much of the apparent pass rate comes from PyTorch fallbacks rather than kernels the model wrote. To close this gap, we co-release Atrex-Kernel-Agent (AKA), a profile-driven kernel-optimization agent that combines iterative measure-revise search, optimization dropout for escaping stalled search contexts, and a layered GPU-optimization knowledge base (298 reference-kernel files and 244 optimization-knowledge documents, plus external upstream reference projects for API/ISA lookup). In a controlled case study, the agent converts zero-FlyDSL fallbacks into real kernels that match or exceed hand-tuned production baselines.

cs.AI↗

arXiv2026

WorldOlympiad: Can Your World Model Survive a Triathlon?

We introduce WorldOlympiad, a benchmark for diagnosing video-based world models across physical faithfulness, geometric consistency, and interaction fidelity. While existing benchmarks often focus on visual quality, semantic alignment, or short-term temporal coherence, they provide limited insight into whether generated videos obey physical rules, preserve coherent 3D structure, and sustain controllable interactions over long horizons. To address this gap, WorldOlympiad decomposes world-model evaluation into three complementary dimensions. The physical track uses object segmentation and MLLM-as-judge to assess whether generated videos follow interpretable rules in mechanics, thermal phenomena, and material properties. The geometry track reconstructs generated videos with Gaussian splatting and evaluates structural consistency, cross-view coherence, and camera-trajectory alignment. The interaction track assesses whether generated rollouts follow complex action prompts and maintain smooth, coherent transitions across consecutive video chunks. WorldOlympiad further covers three major downstream scenarios, including gaming, robotics, and general real-world videos, capturing diverse challenges from interactive control and embodied manipulation to open-domain motion and camera dynamics. Together, these tracks and scenarios form a scalable and interpretable evaluation suite that exposes failure modes beyond generic video quality. Experiments on state-of-the-art models reveal substantial gaps in physical reasoning, 3D consistency, and long-horizon interaction, underscoring the need for more structured evaluation protocols for generative world models.

cs.CV↗

arXiv2026

RTP-LLM: High-Performance Alibaba LLM Inference Engine

Large Language Models (LLMs) have revolutionized AI applications, but deploying them at scale presents significant challenges. We present RTP-LLM, a high-performance inference engine for industrial-scale LLM deployment, successfully deployed across Alibaba Group serving over 100 million users. RTP-LLM addresses fundamental bottlenecks through integrated design. It optimizes model loading via file-order-driven I/O and parallel I/O-communication overlapping. The Prefill-Decode Disaggregation architecture decouples compute-intensive prefill from memory-bound decode phases, combined with hierarchical multi-tiered KV cache management enabling efficient cache reuse. In addition, RTP-LLM incorporates modular speculative decoding supporting multiple algorithms, adaptive KV cache quantization, and decoupled multimodal processing, with support for multi-level parallelism. Comprehensive evaluations across diverse model architectures (8B-235B parameters) have been conducted, where both controlled benchmarks and real production workloads are used. The results demonstrate RTP-LLM's superior performance against vLLM and SGLang: 4.7x-6.3x model loading speedup, 35-37% TTFT P95 latency reduction with 215% cache reuse improvement in production traffic scheduling, 1.12x-2.48x and 1.86x-2.52x throughput improvements in speculative decoding and multimodal inference, respectively, and 35-40% batch latency reduction with 1.9x-3.0x TTFT improvement in quantized inference. RTP-LLM's production-proven architecture and open-source availability make it a comprehensive solution for industrial LLM deployment.

cs.OS↗

arXiv2026

LegoDiffusion: Micro-Serving Text-to-Image Diffusion Workflows

Text-to-image generation executes a diffusion workflow comprising multiple models centered on a base diffusion model. Existing serving systems treat each workflow as an opaque monolith, provisioning, placing, and scaling all constituent models together, which obscures internal dataflow, prevents model sharing, and enforces coarse-grained resource management. In this paper, we make a case for micro-serving diffusion workflows with LegoDiffusion, a system that decomposes a workflow into loosely coupled model-execution nodes that can be independently managed and scheduled. By explicitly managing individual model inference, LegoDiffusion unlocks cluster-scale optimizations, including per-model scaling, model sharing, and adaptive model parallelism. Collectively, LegoDiffusion outperforms existing diffusion workflow serving systems, sustaining up to 3x higher request rates and tolerating up to 8x higher burst traffic.

cs.DC↗

arXiv2026

Dissecting Outlier Dynamics in LLM NVFP4 Pretraining

Training large language models using 4-bit arithmetic enhances throughput and memory efficiency. Yet, the limited dynamic range of FP4 increases sensitivity to outliers. While NVFP4 mitigates quantization error via hierarchical microscaling, a persistent loss gap remains compared to BF16. This study conducts a longitudinal analysis of outlier dynamics across architecture during NVFP4 pretraining, focusing on where they localize, why they occur, and how they evolve temporally. We find that, compared with Softmax Attention (SA), Linear Attention (LA) reduces per-tensor heavy tails but still exhibits persistent block-level spikes under block quantization. Our analysis attributes outliers to specific architectural components: Softmax in SA, gating in LA, and SwiGLU in FFN, with "post-QK" operations exhibiting higher sensitivity to quantization. Notably, outliers evolve from transient spikes early in training to a small set of persistent hot channels (i.e., channels with persistently large magnitudes) in later stages. Based on these findings, we introduce Hot-Channel Patch (HCP), an online compensation mechanism that identifies hot channels and reinjects residuals using hardware-efficient kernels. We then develop CHON, an NVFP4 training recipe integrating HCP with post-QK operation protection. On GLA-1.3B model trained for 60B tokens, CHON reduces the loss gap to BF16 from 0.94% to 0.58% while maintaining downstream accuracy.

cs.LG↗

arXiv2025

Diving into 3D Parallelism with Heterogeneous Spot Instance GPUs: Design and Implications

The rapid growth of large language models (LLMs) and the continuous release of new GPU products have significantly increased the demand for distributed training across heterogeneous GPU environments. In this paper, we present a comprehensive analysis of the challenges involved in implementing 3D parallelism in such environments, addressing critical issues such as the need for symmetric tensor parallelism, efficient gradient synchronization in asymmetric pipeline parallelism, and the trade-offs between memory utilization and computational efficiency. Building upon these insights, we introduce AutoHet, a novel system that automatically identifies the optimal parallelism plan for distributed training on heterogeneous GPUs. AutoHet supports asymmetric 3D parallelism structures and facilitates fine-grained workload distribution. We propose a theoretical model that frames the device grouping and load balancing as an optimization problem to minimize per-iteration training time, thus effectively balancing computing power and memory usage across GPUs with diverse capabilities. To enable elastic training upon spot instance preemption, AutoHet presents an efficient recovery strategy that prioritizes to retrieve training states from local nodes, and only downloads the missing checkpoints from the cloud storage. Our extensive evaluation, conducted on three large-scale models and utilizing combinations of three different GPU types, demonstrates that AutoHet outperforms existing DNN training systems, achieving up to a 1.79$\times$ speedup in training throughput compared with Megatron-LM and Whale, and a 4.38$\times$ speedup of recovery speed compared to a spot instance baseline.

cs.DC↗

arXiv2025

RollMux: Phase-Level Multiplexing for Disaggregated RL Post-Training

Rollout-training disaggregation is emerging as the standard architecture for Reinforcement Learning (RL) post-training, where memory-bound rollout and compute-bound training are physically disaggregated onto purpose-built clusters to maximize hardware efficiency. However, the strict synchronization required by on-policy algorithms introduces severe dependency bubbles, forcing one cluster to idle while the dependent phase is running on the other. We present RollMux, a cluster scheduling framework that reclaims these bubbles through cross-cluster orchestration. RollMux is built on the insight that the structural idleness of one job can be effectively utilized by the active phase of another. To realize this, we introduce the co-execution group abstraction, which partitions the cluster into isolated locality domains. This abstraction enables a two-tier scheduling architecture: an inter-group scheduler that optimizes job placement using conservative stochastic planning, and an intra-group scheduler that orchestrates a provably optimal round-robin schedule. The group abstraction also imposes a residency constraint, ensuring that massive model states remain cached in host memory to enable "warm-star" context switching. We evaluate RollMux on a production-scale testbed with 328 H20 and 328 H800 GPUs. RollMux improves cost efficiency by 1.84x over standard disaggregation and 1.38x over state-of-the-art co-located baselines, all while achieving 100% SLO attainment.

cs.DC↗

arXiv2025

Inferix: A Block-Diffusion based Next-Generation Inference Engine for World Simulation

World models serve as core simulators for fields such as agentic AI, embodied AI, and gaming, capable of generating long, physically realistic, and interactive high-quality videos. Moreover, scaling these models could unlock emergent capabilities in visual perception, understanding, and reasoning, paving the way for a new paradigm that moves beyond current LLM-centric vision foundation models. A key breakthrough empowering them is the semi-autoregressive (block-diffusion) decoding paradigm, which merges the strengths of diffusion and autoregressive methods by generating video tokens in block-applying diffusion within each block while conditioning on previous ones, resulting in more coherent and stable video sequences. Crucially, it overcomes limitations of standard video diffusion by reintroducing LLM-style KV Cache management, enabling efficient, variable-length, and high-quality generation. Therefore, Inferix is specifically designed as a next-generation inference engine to enable immersive world synthesis through optimized semi-autoregressive decoding processes. This dedicated focus on world simulation distinctly sets it apart from systems engineered for high-concurrency scenarios (like vLLM or SGLang) and from classic video diffusion models (such as xDiTs). Inferix further enhances its offering with interactive video streaming and profiling, enabling real-time interaction and realistic simulation to accurately model world dynamics. Additionally, it supports efficient benchmarking through seamless integration of LV-Bench, a new fine-grained evaluation benchmark tailored for minute-long video generation scenarios. We hope the community will work together to advance Inferix and foster world model exploration.

cs.CV↗

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

InstGenIE: Generative Image Editing Made Efficient with Mask-aware Caching and Scheduling

Generative image editing using diffusion models has become a prevalent application in today's AI cloud services. In production environments, image editing typically involves a mask that specifies the regions of an image template to be edited. The use of masks provides direct control over the editing process and introduces sparsity in the model inference. In this paper, we present InstGenIE, a system that efficiently serves image editing requests. The key insight behind InstGenIE is that image editing only modifies the masked regions of image templates while preserving the original content in the unmasked areas. Driven by this insight, InstGenIE judiciously skips redundant computations associated with the unmasked areas by reusing cached intermediate activations from previous inferences. To mitigate the high cache loading overhead, InstGenIE employs a bubble-free pipeline scheme that overlaps computation with cache loading. Additionally, to reduce queuing latency in online serving while improving the GPU utilization, InstGenIE proposes a novel continuous batching strategy for diffusion model serving, allowing newly arrived requests to join the running batch in just one step of denoising computation, without waiting for the entire batch to complete. As heterogeneous masks induce imbalanced loads, InstGenIE also develops a load balancing strategy that takes into account the loads of both computation and cache loading. Collectively, InstGenIE outperforms state-of-the-art diffusion serving systems for image editing, achieving up to 3x higher throughput and reducing average request latency by up to 14.7x while ensuring image quality.

cs.DC↗

arXiv2025

Adaptra: Straggler-Resilient Hybrid-Parallel Training with Pipeline Adaptation

Training large Deep Neural Network (DNN) models at scale often encounters straggler issues, mostly in communications due to network congestion, RNIC/switch defects, or topological asymmetry. Under advanced pipeline parallelism, even minor communication delays can induce significant training slowdowns. This occurs because (1) slow communication disrupts the pipeline schedule, creating cascading "bubbles" in a domino effect, and (2) current GPU kernel scheduling is susceptible to head-of-line blocking, where slow communication blocks subsequent computations, further adding to these bubbles. To address these challenges, we present ADAPTRA, a straggler-resilient training system with two key optimizations. First, it optimally adapts the pipeline schedule in the presence of stragglers to absorb communication delays without inducing cascading bubbles, using a simple yet effective algorithm guided by an analytical model. Second, upon detecting slow communication, ADAPTRA offloads communication operations from GPU to host memory and utilizes CPU-side RDMA for data transfer. This eliminates head-of-line blocking as subsequent computation kernels can be scheduled immediately on GPUs. Together, these optimizations effectively reduce pipeline stalls in the presence of communication stragglers, improving the training iteration time by 1.2-3.5x in our experiments under various settings.

cs.DC↗

arXiv2024

FALCON: Pinpointing and Mitigating Stragglers for Large-Scale Hybrid-Parallel Training

Fail-slows, or stragglers, are common but largely unheeded problems in large-scale hybrid-parallel training that spans thousands of GPU servers and runs for weeks to months. Yet, these problems are not well studied, nor can they be quickly detected and effectively mitigated. In this paper, we first present a characterization study on a shared production cluster with over 10,000 GPUs1. We find that fail-slows are caused by various CPU/GPU computation and cross-node networking issues, lasting from tens of seconds to nearly ten hours, and collectively delaying the average job completion time by 1.34%. The current practice is to manually detect these fail-slows and simply treat them as fail-stops using a checkpoint-and-restart failover approach, which are labor-intensive and time-consuming. In this paper, we propose FALCON, a framework that rapidly identifies fail-slowed GPUs and/or communication links, and effectively tackles them with a novel multi-level mitigation mechanism, all without human intervention. We have applied FALCON to detect human-labeled fail-slows in a production cluster with over 99% accuracy. Cluster deployment further demonstrates that FALCON effectively handles manually injected fail-slows, mitigating the training slowdown by 60.1%.

cs.DC↗

arXiv2024

SwiftDiffusion: Efficient Diffusion Model Serving with Add-on Modules

Text-to-image (T2I) generation using diffusion models has become a blockbuster service in today's AI cloud. A production T2I service typically involves a serving workflow where a base diffusion model is augmented with various "add-on" modules, notably ControlNet and LoRA, to enhance image generation control. Compared to serving the base model alone, these add-on modules introduce significant loading and computational overhead, resulting in increased latency. In this paper, we present SwiftDiffusion, a system that efficiently serves a T2I workflow through a holistic approach. SwiftDiffusion decouples ControNet from the base model and deploys it as a separate, independently scaled service on dedicated GPUs, enabling ControlNet caching, parallelization, and sharing. To mitigate the high loading overhead of LoRA serving, SwiftDiffusion employs a bounded asynchronous LoRA loading (BAL) technique, allowing LoRA loading to overlap with the initial base model execution by up to k steps without compromising image quality. Furthermore, SwiftDiffusion optimizes base model execution with a novel latent parallelism technique. Collectively, these designs enable SwiftDiffusion to outperform the state-of-the-art T2I serving systems, achieving up to 7.8x latency reduction and 1.6x throughput improvement in serving SDXL models on H800 GPUs, without sacrificing image quality.

cs.DC↗

arXiv2017

LERC: Coordinated Cache Management for Data-Parallel Systems

Memory caches are being aggressively used in today's data-parallel frameworks such as Spark, Tez and Storm. By caching input and intermediate data in memory, compute tasks can witness speedup by orders of magnitude. To maximize the chance of in-memory data access, existing cache algorithms, be it recency- or frequency-based, settle on cache hit ratio as the optimization objective. However, unlike the conventional belief, we show in this paper that simply pursuing a higher cache hit ratio of individual data blocks does not necessarily translate into faster task completion in data-parallel environments. A data-parallel task typically depends on multiple input data blocks. Unless all of these blocks are cached in memory, no speedup will result. To capture this all-or-nothing property, we propose a more relevant metric, called effective cache hit ratio. Specifically, a cache hit of a data block is said to be effective if it can speed up a compute task. In order to optimize the effective cache hit ratio, we propose the Least Effective Reference Count (LERC) policy that persists the dependent blocks of a compute task as a whole in memory. We have implemented the LERC policy as a memory manager in Spark and evaluated its performance through Amazon EC2 deployment. Evaluation results demonstrate that LERC helps speed up data-parallel jobs by up to 37% compared with the widely employed least-recently-used (LRU) policy.

cs.DC↗

arXiv2017

LRC: Dependency-Aware Cache Management for Data Analytics Clusters

Memory caches are being aggressively used in today's data-parallel systems such as Spark, Tez, and Piccolo. However, prevalent systems employ rather simple cache management policies--notably the Least Recently Used (LRU) policy--that are oblivious to the application semantics of data dependency, expressed as a directed acyclic graph (DAG). Without this knowledge, memory caching can at best be performed by "guessing" the future data access patterns based on historical information (e.g., the access recency and/or frequency), which frequently results in inefficient, erroneous caching with low hit ratio and a long response time. In this paper, we propose a novel cache replacement policy, Least Reference Count (LRC), which exploits the application-specific DAG information to optimize the cache management. LRC evicts the cached data blocks whose reference count is the smallest. The reference count is defined, for each data block, as the number of dependent child blocks that have not been computed yet. We demonstrate the efficacy of LRC through both empirical analysis and cluster deployments against popular benchmarking workloads. Our Spark implementation shows that, compared with LRU, LRC speeds up typical applications by 60%.

cs.DC↗