FreeScale: Distributed Training for Sequence Recommendation Models with Minimal Scaling Cost
For industrial recommendation systems, FreeScale improves training efficiency by minimizing idle GPU time due to stragglers and communication bottlenecks.
FreeScale addresses computational inefficiencies in distributed training of sequence recommendation models by reducing stragglers, overlapping communication with computation, and avoiding GPU resource contention. It achieves up to 90.3% reduction in computational bubbles on 256 H100 GPUs.
Modern industrial Deep Learning Recommendation Models typically extract user preferences through the analysis of sequential interaction histories, subsequently generating predictions based on these derived interests. The inherent heterogeneity in data characteristics frequently result in substantial under-utilization of computational resources during large-scale training, primarily due to computational bubbles caused by severe stragglers and slow blocking communications. This paper introduces FreeScale, a solution designed to (1) mitigate the straggler problem through meticulously load balanced input samples (2) minimize the blocking communication by overlapping prioritized embedding communications with computations (3) resolve the GPU resource competition during computation and communication overlapping by communicating through SM-Free techniques. Empirical evaluation demonstrates that FreeScale achieves up to 90.3% reduction in computational bubbles when applied to real-world workloads running on 256 H100 GPUs.