Metis: Fast Automatic Distributed Training on Heterogeneous GPUs.
Taegeon UmByungsoo OhMinyoung KangWoo-Yeon LeeGoeun KimDongseob KimYoungtaek KimMohd MuzzammilMyeongjae JeonPublished in: USENIX ATC (2024)
Keyphrases
- heterogeneous environments
- distributed systems
- loosely coupled
- heterogeneous data
- distributed search
- distributed data
- distributed information systems
- training process
- distributed environment
- transparent access
- heterogeneous databases
- training phase
- training algorithm
- cooperative
- general purpose
- supervised learning
- distributed data sources
- training samples
- computing environments
- lightweight
- online learning
- neural network
- training examples
- multi agent
- training set
- parallel programming
- pairwise
- hidden markov models
- peer to peer
- parallel processing
- fully automatic
- semi automatic