SiDA: Sparsity-Inspired Data-Aware Serving for Efficient and Scalable Large Mixture-of-Experts Models.
Zhixu DuShiyu LiYuhao WuXiangyu JiangJingwei SunQilin ZhengYongkai WuAng LiHai (Helen) LiYiran ChenPublished in: CoRR (2023)
Keyphrases
- data sets
- experimental data
- statistical methods
- data collection
- prior knowledge
- training data
- raw data
- statistical analysis
- data analysis
- data sources
- probability distribution
- high quality
- accurate models
- historical data
- databases
- domain experts
- original data
- database
- image data
- data points
- data structure
- data quality
- limited memory
- em algorithm
- neural network
- spatial data
- synthetic data
- parameter estimation
- computer systems
- input data
- xml documents
- bayesian networks