Skywork-MoE: A Deep Dive into Training Techniques for Mixture-of-Experts Language Models

TL;DR

This paper presents Skywork-MoE, a large language model with 146 billion parameters, focusing on innovative training techniques like gating logit normalization and adaptive auxiliary loss, achieving strong benchmark performance.

Contribution

The paper introduces new training methodologies for MoE models, including gating logit normalization and adaptive auxiliary loss, validated through extensive experiments.

Findings

Gating logit normalization enhances expert diversification.

Adaptive auxiliary loss coefficients improve training stability.

Upcycling from dense checkpoints can be effective depending on resources.

Abstract

In this technical report, we introduce the training methodologies implemented in the development of Skywork-MoE, a high-performance mixture-of-experts (MoE) large language model (LLM) with 146 billion parameters and 16 experts. It is initialized from the pre-existing dense checkpoints of our Skywork-13B model. We explore the comparative effectiveness of upcycling versus training from scratch initializations. Our findings suggest that the choice between these two approaches should consider both the performance of the existing dense checkpoints and the MoE training budget. We highlight two innovative techniques: gating logit normalization, which improves expert diversification, and adaptive auxiliary loss coefficients, allowing for layer-specific adjustment of auxiliary loss coefficients. Our experimental results validate the effectiveness of these methods. Leveraging these techniques and insights, we trained our upcycled Skywork-MoE on a condensed subset of our SkyPile corpus. The evaluation results demonstrate that our model delivers strong performance across a wide range of benchmarks.