Linear-MoE: Linear Sequence Modeling Meets Mixture-of-Experts
This work addresses the problem of training and deploying large-scale AI models more efficiently for researchers and practitioners, though it appears incremental as it combines existing LSM and MoE techniques.
The paper tackles the challenge of scaling large-scale models efficiently by integrating Linear Sequence Modeling (LSM) with Mixture-of-Experts (MoE) into a system called Linear-MoE, achieving efficiency gains while maintaining competitive performance on benchmarks for models ranging from 0.3B to 7B parameters.
Linear Sequence Modeling (LSM) like linear attention, state space models and linear RNNs, and Mixture-of-Experts (MoE) have recently emerged as significant architectural improvements. In this paper, we introduce Linear-MoE, a production-level system for modeling and training large-scale models that integrate LSM with MoE. Linear-MoE leverages the advantages of both LSM modules for linear-complexity sequence modeling and MoE layers for sparsely activation, aiming to offer high performance with efficient training. The Linear-MoE system comprises: 1) Modeling subsystem, which provides a unified framework supporting all instances of LSM. and 2) Training subsystem, which facilitates efficient training by incorporating various advanced parallelism technologies, particularly Sequence Parallelism designed for Linear-MoE models. Additionally, we explore hybrid models that combine Linear-MoE layers with standard Transformer-MoE layers with its Sequence Parallelism to further enhance model flexibility and performance. Evaluations on two model series, A0.3B-2B and A1B-7B, demonstrate Linear-MoE achieves efficiency gains while maintaining competitive performance on various benchmarks, showcasing its potential as a next-generation foundational model architecture. Code: https://github.com/OpenSparseLLMs/Linear-MoE.