AquilaMoE: Efficient Training for MoE Models with Scale-Up and Scale-Out Strategies

TL;DR

AquilaMoE introduces an efficient two-stage training methodology for large-scale MoE language models, leveraging knowledge transfer from smaller models to reduce data needs and improve training efficiency.

Contribution

The paper presents a novel EfficientScale training approach for MoE models, combining scale-up and scale-out strategies to enhance performance and reduce resource requirements.

Findings

Successful training of a 16B model using knowledge transfer

AquilaMoE achieves reduced loss during continuous pretraining

Demonstrated improved training efficiency over traditional methods

Abstract

In recent years, with the rapid application of large language models across various fields, the scale of these models has gradually increased, and the resources required for their pre-training have grown exponentially. Training an LLM from scratch will cost a lot of computation resources while scaling up from a smaller model is a more efficient approach and has thus attracted significant attention. In this paper, we present AquilaMoE, a cutting-edge bilingual 8*16B Mixture of Experts (MoE) language model that has 8 experts with 16 billion parameters each and is developed using an innovative training methodology called EfficientScale. This approach optimizes performance while minimizing data requirements through a two-stage process. The first stage, termed Scale-Up, initializes the larger model with weights from a pre-trained smaller model, enabling substantial knowledge transfer and continuous pretraining with significantly less data. The second stage, Scale-Out, uses a pre-trained dense model to initialize the MoE experts, further enhancing knowledge transfer and performance. Extensive validation experiments on 1.8B and 7B models compared various initialization schemes, achieving models that maintain and reduce loss during continuous pretraining. Utilizing the optimal scheme, we successfully trained a 16B model and subsequently the 8*16B AquilaMoE model, demonstrating significant improvements in performance and training efficiency.