# EuroSys 2026 ## Meta Info Homepage: Paper list: ### Acceptance Rate * Spring: 19.6% (= 79 / 404) ## Papers ### Large Language Models (LLMs) * LLM Training * MoE Training * MegaScale-MoE: Large-Scale Communication-Efficient Training of Mixture-of-Experts Models in Production \[[arXiv](https://arxiv.org/abs/2505.11432)] * PKU & ByteDance * Present **MegaScale-MoE**, a production system for efficient large-scale MoE training. * Co-design communication-efficient parallelism, inter- and intra-operator communication-computation overlap, and lower-precision communication compression for MoE layers. * LoRA Fine-Tuning * LoRAFusion: Efficient LoRA Fine-Tuning for LLMs \[[Paper](https://doi.org/10.1145/3767295.3769331)] \[[arXiv](https://arxiv.org/abs/2510.00206)] * UofT & Vector Institute & NVIDIA * Present **LoRAFusion**, a system that improves LoRA fine-tuning by optimizing both fused kernels and multi-job training schedules. * Combine graph-splitting-based kernel fusion with multi-job adaptive batching to reduce memory traffic, improve communication overlap, and mitigate pipeline bubbles. * Federated Fine-Tuning of Sparsely-Activated Large Language Models on Resource-Constrained Devices * Shandong University & XJTU * Data Pipeline * MegaScale-Data: Scaling DataLoader for Multi-Source Large Foundation Model Training \[[arXiv](https://arxiv.org/abs/2504.09844)] * HKU & ByteDance * Present **MegaScale-Data**, an industrial-grade distributed data loading architecture for multi-source large foundation model training. * Disaggregate preprocessing with role-specific actors and use a centralized declarative data plane to support scalable multi-source orchestration under heterogeneous preprocessing costs. * Scheduling and Parallelism * STAlloc: Enhancing Memory Efficiency in Large-Scale Model Training with Spatio-Temporal Planning * THU & Infinigence-AI & SJTU * Zeppelin: Balancing Variable-length Workloads in Data Parallel Large Model Training * PKU & ETH & CUHK & Shanghai AI Lab & MIT * Arena: Efficiently Training Large Models via Dynamic Scheduling and Adaptive Parallelism Co-Design * SJTU & Lenovo Research & Microsoft & Guizhou University & NUS * HARP: Orchestrating Automated Parallel Training on Heterogeneous GPU Clusters * Fudan & Shandong Computer Science Center * HetAuto: Cross-Cluster Auto-Parallelism for Heterogeneous Distributed Training * HKU & Meituan * Efficient and Adaptable Overlapping for Computation and Communication via Signaling and Reordering * THU & PKU & Infinigence-AI & SJTU * Crimson: Collaborative Parameter Updates for Efficient Pipeline Training of Large Language Models * SYSU & HKUST & Pengcheng Laboratory * Suika: Efficient and High-quality Re-scheduling of 3D-parallelized LLM Training Jobs in Shared Clusters * SJTU & TeleAI & Huawei * Runtime Modeling * Maya: Optimizing Deep Learning Training Workloads using GPU Runtime Emulation \[[arXiv](https://arxiv.org/abs/2503.20191)] * Georgia Tech & NVIDIA * Present **Maya**, a performance modeling system for deep learning training based on transparent GPU device emulation. * Intercept device API calls from unmodified training code to observe low-level operations without workload translation or code modification. * Multimodal Training * MegaScale-Omni: A Hyper-Scale, Workload-Resilient System for MultiModal LLM Training in Production * SJTU & ByteDance * Fault Tolerance * Handling Network Faults in Distributed AI Training: Failover is Now an Option * NUS & ByteDance * LLM Inference * Speculative Decoding * AdaServe: Accelerating Multi-SLO LLM Serving with SLO-Customized Speculative Decoding * CMU & Princeton & EPFL & AWS & Purdue * Request Scheduling * FlexPipe: Adapting Dynamic LLM Serving Through Inflight Pipeline Refactoring in Fragmented Serverless Clusters \[[Paper](https://doi.org/10.1145/3767295.3769316)] \[[arXiv](https://arxiv.org/abs/2510.11938)] * SIAT, CAS & UCAS & UCSD & University of Macau * TokenFlow: Responsive LLM Text Streaming Serving under Request Burst via Preemptive Scheduling * SJTU & GMU & China Telecom Shanghai * AdaGen: Workload-Adaptive Cluster Scheduler for Latency-Optimal LLM Inference Serving * UVA & HPE Labs & UC Riverside * SkyWalker: A Locality-Aware Cross-Region Load Balancer for LLM Inference * UC Berkeley & RUC & Rice * PiLLM: Resource-Efficient LLM Inference Using Workload Prediction * ShanghaiTech & SenseTime & Beihang * KV Cache and Memory Management * Taming Latency-Memory Trade-Off in MoE-Based LLM Serving via Fine-Grained Expert Offloading \[[Paper](https://doi.org/10.1145/3767295.3769319)] \[[arXiv](https://arxiv.org/abs/2502.05370)] * Stevens Institute of Technology & Waterloo & Rutgers * KUNSERVE: Parameter-centric Memory Management for Efficient Memory Overloading Handling in LLM Serving * SJTU * High Throughput and Low Latency LLM Serving via Adaptive KV Caching * University of Macau & SIAT, CAS & NTU * Multiplexing * MFS: An Efficient Model Family Serving System for LLMs * HKUST & USTC & Inspur * Efficient Multimodal Serving via Module Multiplexing * HKUST & SYSU & XJTU & MetaX * Sparsity * SAS: Sparse Attention Synthesizer for Efficient Language Model Inference * Amazon * Heterogeneous Environment * Scaling LLM Test-Time Compute with Mobile NPU on Smartphones * THU & USTC & MSR & AIR, THU * TailorLLM: Collaborative End-Cloud Inference of Large and Small Language Models Based on Low-Rank Adaptation * BUPT * Trusted Execution * TZ-LLM: Protecting On-Device Large Language Models with Arm TrustZone * SJTU * LLM-based Applications * AIMS: A Cost-Efficient Framework for LLM-based Agent Deployment in Cloud-Edge Hybrid Environments * UVA & Microsoft * From Imperative to Declarative: Towards LLM-friendly OS Interfaces for Boosted Computer-Use Agents * IS, CAS & UCAS & SJTU ### Diffusion Models * Image Editing * FlashPS: Efficient Generative Image Editing with Mask-aware Caching and Scheduling \[[arXiv](https://arxiv.org/abs/2505.20600)] \[[Code](https://github.com/Sylvia-16/FlashPS)] * HKUST & Alibaba * **Our work!** ### Model Serving * Automated End-to-End Model Serving with Cooperative Compilation and Scheduling * NJU & Hunan University ### Resource Management * Serverless Computing * Efficient Data Passing for Serverless Inference Workflows: A GPU-Centric Approach * HUST & CUHK-Shenzhen & TeleAI & HKUST * iRoute: Local Routing Table-based Workflow Management in Serverless Computing * TJU & THU & IEIT Systems & Inspur * DROPS: Managing Serverless Resource Pools in Microsoft Azure Functions * Waterloo & MSR & Microsoft * Squeezy: Rapid VM Memory Reclamation for Serverless Functions * NTUA & UIUC * Demystifying Serverless Costs on Public Platforms: Bridging Billing, Architecture, and OS Scheduling * UBC & Johns Hopkins * Fix: externalizing network I/O in serverless computing * Stanford * GPU Cluster Management * Bridging the GPU Utilization Gap: Predictive Multi-Dimensional Resource Scheduling for AI Workloads * THU & Alibaba & SJTU * Untangling GPU Power Consumption: Job-Level Inference in Cloud Shared Settings \[[Paper](https://hal.science/hal-05291033v1/file/GPU_power_Eurosys.pdf)] * ÉTS & Inria & OVHcloud & CNRS * Present practical job-level power estimation methods for GPUs under temporal sharing, spatial sharing, and passthrough deployment modes in cloud environments. * Show that GPU sharing can improve energy efficiency for small AI workloads, and identify substantial GPU underutilization in an IaaS GPU cluster. ## Acronyms * LLM: Large Language Model * MoE: Mixture-of-Experts * LoRA: Low-Rank Adaptation