LIBRA: Enabling Workload-aware Multi-dimensional Network Topology Optimization for Distributed Training of Large AI Models

TL;DR

LIBRA is a framework that optimizes multi-dimensional network topologies to reduce communication bottlenecks in distributed training of large AI models, improving resource utilization and training efficiency.

Contribution

The paper introduces LIBRA, a novel framework for optimizing multi-dimensional network fabrics tailored for distributed AI training, addressing bandwidth allocation and architecture design.

Findings

LIBRA effectively enhances network bandwidth utilization.

Optimized fabrics reduce training communication overhead.

Framework enables co-optimization of network architecture and training performance.

Abstract

As model sizes in machine learning continue to scale, distributed training is necessary to accommodate model weights within each device and to reduce training time. However, this comes with the expense of increased communication overhead due to the exchange of gradients and activations, which become the critical bottleneck of the end-to-end training process. In this work, we motivate the design of multi-dimensional networks within machine learning systems as a cost-efficient mechanism to enhance overall network bandwidth. We also identify that optimal bandwidth allocation is pivotal for multi-dimensional networks to ensure efficient resource utilization. We introduce LIBRA, a framework specifically focused on optimizing multi-dimensional fabric architectures. Through case studies, we demonstrate the value of LIBRA, both in architecting optimized fabrics under diverse constraints and in enabling co-optimization opportunities.