arXiv subjects

Xueze Kang

Publications and source records attributed to Xueze Kang.

8 recordsLinked to original sources

arXiv2026

Zellige: Moldable Sequence Placement for Mixed Image-Video DiT Training

High-quality video generation requires training Diffusion Transformers (DiTs) jointly on image and video data, posing a mixed-length sequence training problem across GPUs. Existing systems rely on data parallelism (DP), context parallelism (CP), or their combination; we model these designs as disjoint-group placement and prove that they face a fundamental tradeoff between inter-group load imbalance and intra-group communication redundancy. We present Zellige, a moldable sequence placement system that jointly selects each sequence's parallelism configuration and participating ranks. Zellige consists of three components: a hardware profiler that estimates the execution time and memory consumption of candidate placements, a two-stage planner that balances compute-heavy anchor sequences and packs lighter filler sequences into the remaining capacity, and a coalesced attention engine that efficiently executes whole sequences alongside distributed-attention shards. Across 21 plans, the hardware profile predicts step makespan and peak allocated memory with mean absolute percentage errors of $3.4%$ and $1.5%$, respectively. The two-stage planner solves each batch in 33--119 ms, significantly faster than a joint-placement reference that optimizes all sequences together, while their modeled makespans differ by at most $0.32%$. In end-to-end evaluations, Zellige outperforms KnapFormer by $1.12$--$1.48\times$ on 16 A800 GPUs and $1.27$--$1.54\times$ on 32 A6000 GPUs.

cs.DC↗

arXiv2026

Xema: Efficient Diffusion Serving through Fine-Grained Memory Management and Auto-Configuration

Diffusion models are increasingly deployed as production visual-generation services, where serving high-resolution image and long video generation is often limited by GPU memory. Popular memory-saving techniques such as weight offloading, sharding, and VAE slicing are often not practical because they tend to introduce significant performance overhead. In this paper, we present Xema, a memory-efficient diffusion serving system that exploits predictable tensor lifetimes for trace-guided memory optimization. For each request template, Xema derives an offline memory trace to identify short memory-pressure intervals and applies memory mitigation only within these intervals and only by the amount needed to fit the target GPU budget. Xema further constructs a static memory layout for tensors with predictable lifetimes, reducing fragmentation-induced reserved memory and making offline memory reasoning reliable at runtime. Built on this memory optimization layer, Xema introduces an offline planner that jointly selects parallelism, concurrency, and memory control under GPU memory and SLO constraints. The selected plan is stored in a plan table and directly used by the online serving runtime. We implement Xema on production diffusion pipelines and evaluate it with Flux.2, CogVideoX-5B, and LTX-2. Compared with existing serving configurations, Xema improves SLO attainment by up to 3.7x and reduces planning cost from 6.3 hours to 197 seconds compared with grid search.

cs.DC↗

arXiv2026

KernelFlume: Elastic Core-Attention Scaling for Agentic Long-Context Decoding

LLM serving is increasingly dominated by long and dynamic decode workloads from agents, reasoning models, and extended conversations. When bursty long-context demand exceeds deployed capacity, existing serving systems typically scale out by launching additional serving instances with model replicas. This instance-level elasticity increases KV capacity only by provisioning another full copy of the model, inheriting startup latency, memory overhead, and batch fragmentation. We present KernelFlume, a decode-centric architecture that disaggregates the stable projection/FFN path from core-attention computation: weight nodes execute dense projection/FFN kernels, while weightless attention nodes store token-range KV partitions and scale with request-state demand. To make this separation elastic, KernelFlume maintains a routing table that maps token ranges to attention-node endpoints. It updates routes at token boundaries and uses host-visible graph signals to drive pre-registered UCX endpoint communication outside the captured CUDA Graph. To preserve low per-token latency after disaggregation, KernelFlume combines query-first core-attention dispatch with inter-layer kernel pipelining, overlapping remote attention and communication with local projection/FFN work. On real GPU testbeds (intra-node A6000 and cross-node H100), under a dynamic long-context agentic workload serving Llama-3.1-8B, KernelFlume sustains flat p99 TPOTs of ~74 ms on A6000 and ~34 ms on H100, while lowering cost per million output tokens by up to 32% and 61%, respectively, relative to full-instance elastic scaling with ServerlessLLM, a state-of-the-art instance-startup method. Replaying the same trace at larger model scale in simulation projects a 56--66% cost reduction over ServerlessLLM, widening to 80--85% with cheaper heterogeneous attention-node hardware and persisting into the million-token context range.

cs.DC↗

arXiv2026

On the Spectral Flattening of Quantized Embeddings

Training Large Language Models (LLMs) at ultra-low precision is critically impeded by instability rooted in the conflict between discrete quantization constraints and the intrinsic heavy-tailed spectral nature of linguistic data. By formalizing the connection between Zipfian statistics and random matrix theory, we prove that the power-law decay in the singular value spectra of embeddings is a fundamental requisite for semantic encoding. We derive theoretical bounds showing that uniform quantization introduces a noise floor that disproportionately truncates this spectral tail, which induces spectral flattening and a strictly provable increase in the stable rank of representations. Empirical validation across diverse architectures including GPT-2 and TinyLlama corroborates that this geometric degradation precipitates representational collapse. This work not only quantifies the spectral sensitivity of LLMs but also establishes spectral fidelity as a necessary condition for stable low-bit optimization.

cs.LG↗

arXiv2025

ElasWave: An Elastic-Native System for Scalable Hybrid-Parallel Training

Large-scale LLM pretraining now runs across $10^5$--$10^6$ accelerators, making failures routine and elasticity mandatory. We posit that an elastic-native training system must jointly deliver (i) parameter consistency, (ii) low mean time to recovery (MTTR), (iii) high post-change throughput, and (iv) computation consistency. No prior system achieves all four simultaneously. To achieve these goals, we present ElasWave, which delivers per-step fault tolerance via multi-dimensional scheduling across graph, dataflow, DVFS, and RNG. ElasWave reshapes and reshards micro-batches while preserving the global batch size and gradient scale. It performs online pipeline resharding with asynchronous parameter migration and interleaves ZeRO partitions, reducing parameter recovery processes to disjoint rank-to-rank transfers. It further leverages DVFS to absorb pipeline bubbles and reshards RNG to keep computation consistency. Together, a dynamic communicator enables in-place communication group edits, while per-step in-memory snapshots support online verification and redistribution. We evaluate ElasWave on 96 NPUs and benchmark it against state-of-the-art baselines: throughput improves by $1.35\times$ over ReCycle and $1.60\times$ over TorchFT; communicator recovery completes within one second (up to $82\times/3.6\times$ faster than full/partial rebuilds); migration MTTR drops by as much as $51\%$; and convergence deviation is reduced by approximately $78\%$.

cs.DC↗

arXiv2025

BurstGPT: A Real-world Workload Dataset to Optimize LLM Serving Systems

Serving systems for Large Language Models (LLMs) are often optimized to improve quality of service (QoS) and throughput. However, due to the lack of open-source LLM serving workloads, these systems are frequently evaluated under unrealistic workload assumptions. Consequently, performance may degrade when systems are deployed in real-world scenarios. This work presents BurstGPT, an LLM serving workload with 10.31 million traces from regional Azure OpenAI GPT services over 213 days. BurstGPT captures LLM serving characteristics from user, model and system perspectives: (1) User request concurrency: burstiness variations of requests in Azure OpenAI GPT services, revealing diversified concurrency patterns in different services and model types. (2) User conversation patterns: counts and intervals within conversations for service optimizations. (3) Model response lengths: auto-regressive serving processes of GPT models, showing statistical relations between requests and their responses. (4) System response failures: failures of conversation and API services, showing intensive resource needs and limited availability of LLM services in Azure. The details of the characteristics can serve multiple purposes in LLM serving optimizations, such as system evaluation and trace provisioning. In our demo evaluation with BurstGPT, frequent variations in BurstGPT reveal declines in efficiency, stability, or reliability in realistic LLM serving. We identify that the generalization of KV cache management, scheduling and disaggregation optimizations can be improved under realistic workload evaluations. BurstGPT is publicly available now at https://github.com/HPMLL/BurstGPT and is widely used to develop prototypes of LLM serving frameworks in the industry.

cs.DC↗

arXiv2024

FusionLLM: A Decentralized LLM Training System on Geo-distributed GPUs with Adaptive Compression

To alleviate hardware scarcity in training large deep neural networks (DNNs), particularly large language models (LLMs), we present FusionLLM, a decentralized training system designed and implemented for training DNNs using geo-distributed GPUs across different computing clusters or individual devices. Decentralized training faces significant challenges regarding system design and efficiency, including: 1) the need for remote automatic differentiation (RAD), 2) support for flexible model definitions and heterogeneous software, 3) heterogeneous hardware leading to low resource utilization or the straggler problem, and 4) slow network communication. To address these challenges, in the system design, we represent the model as a directed acyclic graph of operators (OP-DAG). Each node in the DAG represents the operator in the DNNs, while the edge represents the data dependency between operators. Based on this design, 1) users are allowed to customize any DNN without caring low-level operator implementation; 2) we enable the task scheduling with the more fine-grained sub-tasks, offering more optimization space; 3) a DAG runtime executor can implement RAD withour requiring the consistent low-level ML framework versions. To enhance system efficiency, we implement a workload estimator and design an OP-Fence scheduler to cluster devices with similar bandwidths together and partition the DAG to increase throughput. Additionally, we propose an AdaTopK compressor to adaptively compress intermediate activations and gradients at the slowest communication links. To evaluate the convergence and efficiency of our system and algorithms, we train ResNet-101 and GPT-2 on three real-world testbeds using 48 GPUs connected with 8 Mbps~10 Gbps networks. Experimental results demonstrate that our system and method can achieve 1.45 - 9.39x speedup compared to baseline methods while ensuring convergence.

cs.DC↗

arXiv2024

Fault-Tolerant Hybrid-Parallel Training at Scale with Reliable and Efficient In-memory Checkpointing

To efficiently scale large model (LM) training, researchers transition from data parallelism (DP) to hybrid parallelism (HP) on GPU clusters, which frequently experience hardware and software failures. Existing works introduce in-memory checkpointing optimizations that snapshot parameters to device memory for rapid failure recovery. However, these methods introduce severe resource competition between checkpointing and training, which can work under DP but can hardly scale under resource-intensive HP. To ensure low checkpointing overhead for hybrid-parallel training, this paper introduces a distributed in-memory checkpointing system with near-zero in-memory saving overhead. It strives from two aspects to mitigate the on-host resource competition caused by in-memory checkpointing: (1) It introduces Hierarchical Asynchronous Snapshotting Coordination in the checkpoint saving stage. This approach uses three-level asynchronous on-device scheduling to enhance parallelism between snapshotting and training, thereby minimizing snapshotting overhead. (2) It proposes Hybrid In-memory Checkpoint Protection to enhance checkpoint completeness during hardware failures. Unlike methods that require inter-node communications, which may block training under HP, it creates intra-node redundancy with efficient resource utilization, protecting training against hardware failures with minimal overhead. With these methods, this work enables fast restart for failed HP training with Distributed In-memory Checkpoint Loading, bypassing inefficiencies in NFS reads. In our evaluation, we achieve zero in-memory checkpoint saving overhead on Frontier while training Llama-2-34B on 256 MI250X devices (512 GPUs).

cs.DC↗