Cephalo: Harnessing Heterogeneous GPU Clusters for Training Transformer Models

TL;DR

Cephalo is a system designed to efficiently train transformer models on heterogeneous GPU clusters by optimizing resource utilization, leading to higher throughput and support for larger models compared to existing methods.

Contribution

Cephalo introduces a novel approach that decouples compute distribution from training state assignment to better utilize diverse GPU resources.

Findings

Cephalo achieves higher training throughput than state-of-the-art methods.

Supports larger models and batch sizes effectively.

Outperforms existing strategies in heterogeneous GPU environments.

Abstract

Training transformer models requires substantial GPU compute and memory resources. In homogeneous clusters, distributed strategies allocate resources evenly, but this approach is inefficient for heterogeneous clusters, where GPUs differ in power and memory. As high-end GPUs are costly and limited in availability, heterogeneous clusters with diverse GPU types are becoming more common. Existing methods attempt to balance compute across GPUs based on capacity but often underutilize compute due to memory constraints. We present Cephalo, a system that optimizes compute and memory usage by decoupling compute distribution from training state assignment. Cephalo outperforms state-of-the-art methods by achieving significantly higher training throughput while supporting larger models and batch sizes.