SiDA: Sparsity-Inspired Data-Aware Serving for Efficient and Scalable Large Mixture-of-Experts Models.
Zhixu DuShiyu LiYuhao WuXiangyu JiangJingwei SunQilin ZhengYongkai WuAng LiHai LiYiran ChenPublished in: MLSys (2024)
Keyphrases
- experimental data
- data sets
- data analysis
- synthetic data
- raw data
- data processing
- data structure
- data points
- knowledge discovery
- image data
- accurate models
- historical data
- original data
- statistical methods
- statistical models
- domain experts
- data collection
- data sources
- domain knowledge
- prior knowledge
- lightweight
- mixture model
- statistical analysis
- input data
- human experts
- data quality
- latent variables
- learning models
- discrete data
- database