LocMoE: A Low-Overhead MoE for Large Language Model Training
This work addresses efficiency issues in distributed training of large language models, representing an incremental improvement over existing MoE methods.
The paper tackles performance limitations in Mixtures-of-Experts (MoE) models for large language models, such as load imbalance and high communication latency, by proposing a novel routing strategy that combines load balance and locality, resulting in a reduction of training time per epoch by 12.68% to 22.24% without affecting model accuracy.
The Mixtures-of-Experts (MoE) model is a widespread distributed and integrated learning method for large language models (LLM), which is favored due to its ability to sparsify and expand models efficiently. However, the performance of MoE is limited by load imbalance and high latency of All-to-All communication, along with relatively redundant computation owing to large expert capacity. Load imbalance may result from existing routing policies that consistently tend to select certain experts. The frequent inter-node communication in the All-to-All procedure also significantly prolongs the training time. To alleviate the above performance problems, we propose a novel routing strategy that combines load balance and locality by converting partial inter-node communication to that of intra-node. Notably, we elucidate that there is a minimum threshold for expert capacity, calculated through the maximal angular deviation between the gating weights of the experts and the assigned tokens. We port these modifications on the PanGu-Sigma model based on the MindSpore framework with multi-level routing and conduct experiments on Ascend clusters. The experiment results demonstrate that the proposed LocMoE reduces training time per epoch by 12.68% to 22.24% compared to classical routers, such as hash router and switch router, without impacting the model accuracy.