LIBRA: Enabling Workload-aware Multi-dimensional Network Topology Optimization for Distributed Training of Large AI Models
This work addresses a critical performance issue for researchers and engineers scaling up AI model training, though it appears incremental as it builds on existing multi-dimensional network concepts.
The paper tackles the communication bottleneck in distributed training of large AI models by proposing LIBRA, a framework for optimizing multi-dimensional network topologies, which enhances bandwidth allocation and resource utilization.
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.