MLSys 2025 Papers

Skip to yearly menu bar Skip to main content

Browse

mini compact topic detail

by

Context Parallelism for Scalable Million-Token Inference

ReaL: Efficient RLHF Training of Large Language Models with Parameter Reallocation

SwiftVI: Time-Efficient Planning and Learning with MDPs

Venn: Resource Management Across Federated Learning Jobs

VoLUT: Efficient Volumetric streaming enhanced by LUT-based super-resolution

Training Ultra Long Context Language Model with Fully Pipelined Distributed Transformer

Rethinking Key-Value Cache Compression Techniques for Large Language Model Serving

TileLink: Generating Efficient Compute-Communication Overlapping Kernels using Tile-Centric Primitives

APOLLO: SGD-like Memory, AdamW-level Performance

Youmu: Efficient Columnar Data Pipeline for LLM Training

FLStore: Efficient Federated Learning Storage for non-training workloads

COMET: Fine-grained Computation-communication Overlapping for Mixture-of-Experts

On Distributed Larger-Than-Memory Subset Selection With Pairwise Submodular Functions

ThunderServe: High-performance and Cost-efficient LLM Serving in Cloud Environments

Efficient LLM Inference using Dynamic Input Pruning and Cache-Aware Masking

Uncertainty and Interference-aware Runtime Prediction for Edge Computing with Conformal Matrix Completion

FastTree: Optimizing Attention Kernel and Runtime for Tree-Structured LLM Inference

LServe: Efficient Long-sequence LLM Serving with Unified Sparse Attention

AIOpsLab: A Holistic Framework to Evaluate AI Agents for Enabling Autonomous Clouds

Know Where You’re Uncertain When Planning with Multimodal Foundation Models: A Formal Framework

Graph Learning at Scale: Characterizing and Optimizing Pre-Propagation GNNs

MiLo: Efficient Quantized MoE Inference with Mixture of Low-Rank Compensators

AdaParse: An Adaptive Parallel PDF Parsing and Resource Scaling Engine

DiffServe: Efficiently Serving Text-to-Image Diffusion Models with Query-Aware Model Scaling

Seesaw: High-throughput LLM Inference via Model Re-sharding

ScaleFusion: Scalable Inference of Spatial-Temporal Diffusion Transformers for High-Resolution Long Video Generation

Supply-Chain Attacks in Machine Learning Frameworks

Enabling Unstructured Sparse Acceleration on Structured Sparse Accelerators

A Practical Cross-Layer Approach for ML-Driven Storage Placement in Warehouse-Scale Computers

FedProphet: Memory-Efficient Federated Adversarial Training via Robust and Consistent Cascade Learning

HyC-LoRA: Memory Efficient LoRA Fine-tuning with \textbf{Hy}brid Activation \textbf{C}ompression

FlexInfer: Flexible LLM Inference with CPU Computations

Efficient On-Device Machine Learning with a Biologically-Plausible Forward-Only Algorithm

SystemX: Federated LLM Pre-Training

Lightweight Software Kernels and Hardware Extensions for Efficient Sparse Deep Neural Networks on Microcontrollers

Self-Data Distillation for Recovering Quality in Pruned Large Language Models

Radius: Range-based Gradient Sparsity for Large Foundation Model Pre-training

PipeFill: Using GPUs During Bubbles in Pipeline-parallel LLM Training

SOLA: Optimizing SLO Attainment for Large Language Model Serving with State-Aware Scheduling

LAVA: Lifetime-Aware VM Allocation with Learned Distributions and Adaptation to Mispredictions

Morphling: Exploiting Job Reconfigurability for Deep Learning Cluster Scheduling

MEADOW: Memory-efficient Dataflow and Data Packing for Low Power Edge LLMs

Scaling Deep Learning Training with MPMD Pipeline Parallelism

Marconi: Prefix Caching for the Era of Hybrid LLMs

Balancing Pipeline Parallelism with Vocabulary Parallelism

SparseTransX: Efficient Training of Translation-Based Knowledge Graph Embeddings Using Sparse Matrix Operations

MAS-ATTENTION: MEMORY-AWARE STREAM PROCESSING FOR ATTENTION ACCELERATION ON RESOURCE-CONSTRAINED EDGE DEVICES

XGrammar: Flexible and Efficient Structured Generation Engine for Large Language Models

SPA: SCALING GRAPH NEURAL NETWORK TRAINING ON LARGE GRAPHS VIA PROBABILISTIC SPLITTING

FlexAttention: A Programming Model for Generating Fused Attention Variants.

Lumos: Efficient Performance Modeling and Estimation for Large-scale LLM Training

NEO: Saving GPU Memory Crisis with CPU Offloading for Online LLM Inference

Optimizing LLM Queries in Relational Data Analytics Workloads

AI Metropolis: Scaling Large Language Model-based Multi-Agent Simulation with Out-of-order Execution

LeanAttention: Hardware-Aware Scalable Attention Mechanism for the Decode-Phase of Transformers

TurboAttention: Efficient attention approximation for high throughputs llm

ProtoRAIL: A Risk-cognizant Imitation Agent for Adaptive vCPU Oversubscription In the Cloud

SampleAttention: Near-Lossless Acceleration of Long Context LLM Inference with Adaptive Structured Sparse Attention

QServe:W4A8KV4 Quantization and System Co-design for Efficient LLM Serving

FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving

The Hidden Bloat in Machine Learning Systems