Efficient ML & Distributed GPU Orchestration
We publish practical research on workload optimization and orchestration across heterogeneous GPUs. Find papers, reproducible benchmarks, grants, and media coverage
PUBLICATION
Peer-reviewed papers and preprints on efficient training, model offloading, inference latency, and GPU schedking.
Highly-efficient billion-scale AI models training and inference using affordable GPUs
ZeRO-Offload and Sentinel for transformers
USENIX ATC’21HPCA’21
DyNN-Offload for Mixture-of-Experts (MoE)
HPCA’24
TECO-Offload on disaggregated memory
SC’24
Billion-scale graph neural network
ASPLOS’23
AI training based on parallelism management
Runtime Concurrency Control and Operation Scheduling
IPDPS’21
Tree structure-aware high performance inference engine
EuroSys’21
AI training using novel hardware
Energy-efficient training on GPU-FPGA accelerators
ICS’21
Processing-in-memory for energy-efficient DNN
MICRO’18
AWARD
Decentralized AI Computing Operating System for Accessible and Cost-Effective AI
NSF-Symbol
MEDIA REPORTS
Microsoft And The University Of California, Merced Introduces ZeRO-Offload, A Novel Heterogeneous DeepLearning...
What’s New in HPC Research: Galaxies, Fugaku, Electron Microscopes & More
Fit More and Train Faster With ZeRO via DeepSpeed and FairScale
Microsoft Releases AI Training Library ZeRO-3 Offload