arXiv subjects

Wencong Xiao

Publications and source records attributed to Wencong Xiao.

17 recordsLinked to original sources

arXiv2026

TensorHub: Scalable and Elastic Weight Transfer for LLM RL Training

Modern LLM reinforcement learning (RL) workloads require a highly efficient weight transfer system to scale training across heterogeneous computational resources. However, existing weight transfer approaches either fail to provide flexibility for dynamically scaling clusters or incur fundamental data movement overhead, resulting in poor performance. We introduce Reference-Oriented Storage (ROS), a new storage abstraction for RL weight transfer that exploits the highly replicated model weights in place. ROS presents the illusion that certain versions of the model weights are stored and can be fetched on demand. Underneath, ROS does not physically store any copies of the weights; instead, it tracks the workers that hold these weights on GPUs for inference. Upon request, ROS directly uses them to serve reads. We build TensorHub, a production-quality system that extends the ROS idea with topology-optimized transfer, strong consistency, and fault tolerance. Evaluation shows that TensorHub fully saturates RDMA bandwidth and adapts to three distinct rollout workloads with minimal engineering effort. Specifically, TensorHub reduces total GPU stall time by up to 6.7x for standalone rollouts, accelerates weight update for elastic rollout by 4.8x, and cuts cross-datacenter rollout stall time by 19x. TensorHub has been deployed in production to support cutting-edge RL training.

cs.DC↗

arXiv2026

BootSeer: Analyzing and Mitigating Initialization Bottlenecks in Large-Scale LLM Training

Large Language Models (LLMs) have become a cornerstone of modern AI, driving breakthroughs in natural language processing and expanding into multimodal jobs involving images, audio, and video. As with most computational software, it is important to distinguish between ordinary runtime performance and startup overhead. Prior research has focused on runtime performance: improving training efficiency and stability. This work focuses instead on the increasingly critical issue of startup overhead in training: the delay before training jobs begin execution. Startup overhead is particularly important in large, industrial-scale LLMs, where failures occur more frequently and multiple teams operate in iterative update-debug cycles. In one of our training clusters, more than 3.5% of GPU time is wasted due to startup overhead alone. In this work, we present the first in-depth characterization of LLM training startup overhead based on real production data. We analyze the components of startup cost, quantify its direct impact, and examine how it scales with job size. These insights motivate the design of Bootseer, a system-level optimization framework that addresses three primary startup bottlenecks: (a) container image loading, (b) runtime dependency installation, and (c) model checkpoint resumption. To mitigate these bottlenecks, Bootseer introduces three techniques: (a) hot block record-and-prefetch, (b) dependency snapshotting, and (c) striped HDFS-FUSE. Bootseer has been deployed in a production environment and evaluated on real LLM training workloads, demonstrating a 50% reduction in startup overhead.

cs.LG↗

arXiv2025

Fast LLM Post-training via Decoupled and Fastest-of-N Speculation

Rollout dominates the training time in large language model (LLM) post-training, where the trained model is used to generate tokens given a batch of prompts. This work, SpecActor, achieves fast rollout with speculative decoding that deploys a fast draft path to accelerate the unparallelizable generation, while the correctness is guaranteed by fast parallel verification of the outputs with the original model. SpecActor addresses two foundational challenges that hinder speculation efficiency: (1) a Decoupled speculation method that overcomes the computation inefficiency issue when executing speculative decoding with relative large per-worker batch size -- a common configuration in training but unfriendly to speculation, and (2) a Fastest-of-N speculation method that selects and combines different draft methods according to the rollout progress to approximate the optimal draft method even when the best one is unknown a priori. Extensive evaluations on production traces show that SpecActor accelerates mean rollout speed by 2.0--2.4x, with up to 2.7x speedup, over common post-training baselines. The results are consistent across both dense and MoE models and across different RL algorithms. Notably, SpecActor is 1.1--2.6x faster compared to vanilla speculative rollout in different traces. The accelerated rollout achieves 1.4--2.3x faster end-to-end training time.

cs.DC↗

arXiv2025

Robust LLM Training Infrastructure at ByteDance

The training scale of large language models (LLMs) has reached tens of thousands of GPUs and is still continuously expanding, enabling faster learning of larger models. Accompanying the expansion of the resource scale is the prevalence of failures (CUDA error, NaN values, job hang, etc.), which poses significant challenges to training stability. Any large-scale LLM training infrastructure should strive for minimal training interruption, efficient fault diagnosis, and effective failure tolerance to enable highly efficient continuous training. This paper presents ByteRobust, a large-scale GPU infrastructure management system tailored for robust and stable training of LLMs. It exploits the uniqueness of LLM training process and gives top priorities to detecting and recovering failures in a routine manner. Leveraging parallelisms and characteristics of LLM training, ByteRobust enables high-capacity fault tolerance, prompt fault demarcation, and localization with an effective data-driven approach, comprehensively ensuring continuous and efficient training of LLM tasks. ByteRobust is deployed on a production GPU platform and achieves 97% ETTR for a three-month training job on 9,600 GPUs.

cs.LG↗

arXiv2025

Efficient Pre-Training of LLMs via Topology-Aware Communication Alignment on More Than 9600 GPUs

The scaling law for large language models (LLMs) depicts that the path towards machine intelligence necessitates training at large scale. Thus, companies continuously build large-scale GPU clusters, and launch training jobs that span over thousands of computing nodes. However, LLM pre-training presents unique challenges due to its complex communication patterns, where GPUs exchange data in sparse yet high-volume bursts within specific groups. Inefficient resource scheduling exacerbates bandwidth contention, leading to suboptimal training performance. This paper presents Arnold, a scheduling system summarizing our experience to effectively align LLM communication patterns with data center topology at scale. An in-depth characteristic study is performed to identify the impact of physical network topology to LLM pre-training jobs. Based on the insights, we develop a scheduling algorithm to effectively align communication patterns with the physical network topology in modern data centers. Through simulation experiments, we show the effectiveness of our algorithm in reducing the maximum spread of communication groups by up to $1.67$x. In production training, our scheduling system improves the end-to-end performance by $10.6\%$ when training with more than $9600$ GPUs, a significant improvement for our training pipeline.

cs.DC↗

arXiv2025

Mycroft: Tracing Dependencies in Collective Communication Towards Reliable LLM Training

Reliability is essential for ensuring efficiency in LLM training. However, many real-world reliability issues remain difficult to resolve, resulting in wasted resources and degraded model performance. Unfortunately, today's collective communication libraries operate as black boxes, hiding critical information needed for effective root cause analysis. We propose Mycroft, a lightweight distributed tracing and root cause analysis system designed to address previously hidden reliability issues in collective communication. Mycroft's key idea is to trace collective communication states and leverage internal control and data dependencies to resolve reliability problems in LLM training. Mycroft has been deployed at ByteDance for over six months to debug collective communication related issues at runtime. It detected anomalies within 15 seconds in 90% of cases and identified the root cause within 20 seconds in 60% of cases. We also conducted extensive fault injection experiments to demonstrate Mycroft's capability and efficiency.

cs.DC↗

arXiv2025

Enhancing Large-Scale AI Training Efficiency: The C4 Solution for Real-Time Anomaly Detection and Communication Optimization

The emergence of Large Language Models (LLMs) has necessitated the adoption of distributed training techniques, involving the deployment of thousands of GPUs to train a single model. Unfortunately, the efficiency of large-scale distributed training systems is often suboptimal due to the increased likelihood of hardware errors in high-end GPU products and the heightened risk of network traffic collisions. Moreover, any local hardware failure can disrupt training tasks, and the inability to swiftly identify faulty components leads to a significant waste of GPU resources. And, prolonged communication due to traffic collisions can substantially increase GPU waiting times. To address these challenges, we propose a communication-driven solution, namely the C4. The key insights of C4 are twofold. First, the load in distributed training exhibits homogeneous characteristics and is divided into iterations through periodic synchronization, therefore hardware anomalies would incur certain syndrome in collective communication. By leveraging this feature, C4 can rapidly identify the faulty components, swiftly isolate the anomaly, and restart the task, thereby avoiding resource wastage caused by delays in anomaly detection. Second, the predictable communication model of collective communication, involving a limited number of long-lived flows, allows C4 to efficiently execute traffic planning, substantially reducing bandwidth competition among these flows. The C4 has been extensively deployed across real-world production systems in a hyperscale cloud provider, yielding a significant improvement in system efficiency, from 30% to 45%. This enhancement is attributed to a 30% reduction in error-induced overhead and a 15% reduction in communication costs.

cs.DC↗

arXiv2025

Efficient Heterogeneous Large Language Model Decoding with Model-Attention Disaggregation

Transformer-based large language models (LLMs) exhibit impressive performance in generative tasks but also introduce significant challenges in real-world serving due to inefficient use of the expensive, computation-optimized accelerators. Although disaggregated serving architectures have been proposed to split different phases of LLM inference, the efficiency of decoding phase is still low. This is caused by the varying resource demands of different operators in the transformer-based LLMs. Specifically, the attention operator is memory-intensive, exhibiting a memory access pattern that clashes with the strengths of modern accelerators, especially for long context requests. To enhance the efficiency of LLM decoding, we introduce model-attention disaggregation. This approach leverages a collection of cheap, memory-optimized devices for the attention operator while still utilizing high-end accelerators for other parts of the model. This heterogeneous setup ensures that each component is tailored to its specific workload, maximizing overall performance and cost efficiency. Our comprehensive analysis and experiments confirm the viability of splitting the attention computation over multiple devices. Also, the communication bandwidth required between heterogeneous devices proves to be manageable with prevalent networking technologies. To further validate our theory, we develop and deploy Lamina, an LLM inference system that incorporates model-attention disaggregation in a distributed heterogeneous cluster. Experimental results indicate that Lamina can provide 16.1 ~ 90.1% higher estimated throughput than existing solutions with similar costs.

cs.LG↗

arXiv2024

Rubick: Exploiting Job Reconfigurability for Deep Learning Cluster Scheduling

The era of large deep learning models has given rise to advanced training strategies such as 3D parallelism and the ZeRO series. These strategies enable various (re-)configurable execution plans for a training job, which exhibit remarkably different requirements of multiple resource types. Existing cluster scheduling systems, however, treat such reconfigurable training jobs as black boxes: they rely on users to choose execution plans statically, and then make resource allocations without awareness of the chosen plans and their resource requirements. This approach results in mismatches between execution plans and resources, making both training performance and cluster utilization far from optimal. We introduce Rubick, a cluster scheduling system for deep learning training that exploits the reconfigurability to improve job performance and cluster efficiency. Rubick incorporates the job execution planning as a new dimension in cluster scheduling, by continuously reconfiguring jobs' execution plans and tuning multi-resource allocations across jobs jointly. Such a co-optimization is navigated by a performance model that understands the diverse resource requirements and performance characteristics of different jobs and execution plans. Rubick exploits such a model to make performance-aware scheduling decisions to maximize cluster throughput while providing performance guarantees to individual jobs. Evaluations on a 64-GPU high-performance training cluster show that Rubick improves average job completion time and makespan by up to 3.2x and 1.4x, respectively, compared against state-of-the-art systems.

cs.DC↗

arXiv2024

Infinite-LLM: Efficient LLM Service for Long Context with DistAttention and Distributed KVCache

Large Language Models (LLMs) demonstrate substantial potential across a diverse array of domains via request serving. However, as trends continue to push for expanding context sizes, the autoregressive nature of LLMs results in highly dynamic behavior of the attention layers, showcasing significant differences in computational characteristics and memory requirements from the non-attention layers. This presents substantial challenges for resource management and performance optimization in service systems. Existing static model parallelism and resource allocation strategies fall short when dealing with this dynamicity. To address the issue, we propose Infinite-LLM, a novel LLM serving system designed to effectively handle dynamic context lengths. Infinite-LLM disaggregates attention layers from an LLM's inference process, facilitating flexible and independent resource scheduling that optimizes computational performance and enhances memory utilization jointly. By leveraging a pooled GPU memory strategy across a cluster, Infinite-LLM not only significantly boosts system throughput but also supports extensive context lengths. Evaluated on a dataset with context lengths ranging from a few to 2000K tokens across a cluster with 32 A100 GPUs, Infinite-LLM demonstrates throughput improvement of 1.35-3.4x compared to state-of-the-art methods, enabling efficient and elastic LLM deployment.

cs.DC↗

arXiv2024

Llumnix: Dynamic Scheduling for Large Language Model Serving

Inference serving for large language models (LLMs) is the key to unleashing their potential in people's daily lives. However, efficient LLM serving remains challenging today because the requests are inherently heterogeneous and unpredictable in terms of resource and latency requirements, as a result of the diverse applications and the dynamic execution nature of LLMs. Existing systems are fundamentally limited in handling these characteristics and cause problems such as severe queuing delays, poor tail latencies, and SLO violations. We introduce Llumnix, an LLM serving system that reacts to such heterogeneous and unpredictable requests by runtime rescheduling across multiple model instances. Similar to context switching across CPU cores in modern operating systems, Llumnix reschedules requests to improve load balancing and isolation, mitigate resource fragmentation, and differentiate request priorities and SLOs. Llumnix implements the rescheduling with an efficient and scalable live migration mechanism for requests and their in-memory states, and exploits it in a dynamic scheduling policy that unifies the multiple rescheduling scenarios elegantly. Our evaluations show that Llumnix improves tail latencies by an order of magnitude, accelerates high-priority requests by up to 1.5x, and delivers up to 36% cost savings while achieving similar tail latencies, compared against state-of-the-art LLM serving systems. Llumnix is publicly available at https://github.com/AlibabaPAI/llumnix.

cs.AR↗

arXiv2023

EasyScale: Accuracy-consistent Elastic Training for Deep Learning

Distributed synchronized GPU training is commonly used for deep learning. The resource constraint of using a fixed number of GPUs makes large-scale training jobs suffer from long queuing time for resource allocation, and lowers the cluster utilization. Adapting to resource elasticity can alleviate this but often introduces inconsistent model accuracy, due to lacking of capability to decouple model training procedure from resource allocation. We propose EasyScale, an elastic training system that achieves consistent model accuracy under resource elasticity for both homogeneous and heterogeneous GPUs. EasyScale preserves the data-parallel training behaviors strictly, traces the consistency-relevant factors carefully, utilizes the deep learning characteristics for EasyScaleThread abstraction and fast context-switching. To utilize heterogeneous cluster, EasyScale dynamically assigns workers based on the intra-/inter-job schedulers, minimizing load imbalance and maximizing aggregated job throughput. Deployed in an online serving cluster, EasyScale powers the training jobs to utilize idle GPUs opportunistically, improving overall cluster utilization by 62.1%.

cs.DC↗

arXiv2023

FusionAD: Multi-modality Fusion for Prediction and Planning Tasks of Autonomous Driving

Building a multi-modality multi-task neural network toward accurate and robust performance is a de-facto standard in perception task of autonomous driving. However, leveraging such data from multiple sensors to jointly optimize the prediction and planning tasks remains largely unexplored. In this paper, we present FusionAD, to the best of our knowledge, the first unified framework that fuse the information from two most critical sensors, camera and LiDAR, goes beyond perception task. Concretely, we first build a transformer based multi-modality fusion network to effectively produce fusion based features. In constrast to camera-based end-to-end method UniAD, we then establish a fusion aided modality-aware prediction and status-aware planning modules, dubbed FMSPnP that take advantages of multi-modality features. We conduct extensive experiments on commonly used benchmark nuScenes dataset, our FusionAD achieves state-of-the-art performance and surpassing baselines on average 15% on perception tasks like detection and tracking, 10% on occupancy prediction accuracy, reducing prediction error from 0.708 to 0.389 in ADE score and reduces the collision rate from 0.31% to only 0.12%.

cs.CV↗

arXiv2023

MIGPerf: A Comprehensive Benchmark for Deep Learning Training and Inference Workloads on Multi-Instance GPUs

New architecture GPUs like A100 are now equipped with multi-instance GPU (MIG) technology, which allows the GPU to be partitioned into multiple small, isolated instances. This technology provides more flexibility for users to support both deep learning training and inference workloads, but efficiently utilizing it can still be challenging. The vision of this paper is to provide a more comprehensive and practical benchmark study for MIG in order to eliminate the need for tedious manual benchmarking and tuning efforts. To achieve this vision, the paper presents MIGPerf, an open-source tool that streamlines the benchmark study for MIG. Using MIGPerf, the authors conduct a series of experiments, including deep learning training and inference characterization on MIG, GPU sharing characterization, and framework compatibility with MIG. The results of these experiments provide new insights and guidance for users to effectively employ MIG, and lay the foundation for further research on the orchestration of hybrid training and inference workloads on MIGs. The code and results are released on https://github.com/MLSysOps/MIGProfiler. This work is still in progress and more results will be published soon.

cs.LG↗

arXiv2022

Whale: Efficient Giant Model Training over Heterogeneous GPUs

The scaling up of deep neural networks has been demonstrated to be effective in improving model quality, but also encompasses several training challenges in terms of training efficiency, programmability, and resource adaptability. We present Whale, a general and efficient distributed training framework for giant models. To support various parallel strategies and their hybrids, Whale generalizes the programming interface by defining two new primitives in the form of model annotations, allowing for incorporating user hints. The Whale runtime utilizes those annotations and performs graph optimizations to transform a local deep learning DAG graph for distributed multi-GPU execution. Whale further introduces a novel hardware-aware parallel strategy, which improves the performance of model training on heterogeneous GPUs in a balanced manner. Deployed in a production cluster with 512 GPUs, Whale successfully trains an industry-scale multimodal model with over ten trillion model parameters, named M6, demonstrating great scalability and efficiency.

cs.DC↗

arXiv2019

Analysis of Large-Scale Multi-Tenant GPU Clusters for DNN Training Workloads

With widespread advances in machine learning, a number of large enterprises are beginning to incorporate machine learning models across a number of products. These models are typically trained on shared, multi-tenant GPU clusters. Similar to existing cluster computing workloads, scheduling frameworks aim to provide features like high efficiency, resource isolation, fair sharing across users, etc. However Deep Neural Network (DNN) based workloads, predominantly trained on GPUs, differ in two significant ways from traditional big data analytics workloads. First, from a cluster utilization perspective, GPUs represent a monolithic resource that cannot be shared at a fine granularity across users. Second, from a workload perspective, deep learning frameworks require gang scheduling reducing the flexibility of scheduling and making the jobs themselves inelastic to failures at runtime. In this paper we present a detailed workload characterization of a two-month long trace from a multi-tenant GPU cluster in a large enterprise. By correlating scheduler logs with logs from individual jobs, we study three distinct issues that affect cluster utilization for DNN training workloads on multi-tenant clusters: (1) the effect of gang scheduling and locality constraints on queuing, (2) the effect of locality on GPU utilization, and (3) failures during training. Based on our experience running a large-scale operation, we provide design guidelines pertaining to next-generation cluster schedulers for DNN training workloads.

cs.DC↗

arXiv2018

Balanced Sparsity for Efficient DNN Inference on GPU

In trained deep neural networks, unstructured pruning can reduce redundant weights to lower storage cost. However, it requires the customization of hardwares to speed up practical inference. Another trend accelerates sparse model inference on general-purpose hardwares by adopting coarse-grained sparsity to prune or regularize consecutive weights for efficient computation. But this method often sacrifices model accuracy. In this paper, we propose a novel fine-grained sparsity approach, balanced sparsity, to achieve high model accuracy with commercial hardwares efficiently. Our approach adapts to high parallelism property of GPU, showing incredible potential for sparsity in the widely deployment of deep learning services. Experiment results show that balanced sparsity achieves up to 3.1x practical speedup for model inference on GPU, while retains the same high model accuracy as fine-grained sparsity.

cs.CV↗