English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Deep learning recommendation models (DLRM) rely on large embedding tables tomanage categorical sparse features. Expanding such embedding tables cansignificantly enhance model performance, but at the cost of increasedGPU/CPU/memory usage. Meanwhile, tech companies have built extensivecloud-based services to accelerate training DLRM models at scale. In thispaper, we conduct a deep investigation of the DLRM training platforms atAntGroup and reveal two critical challenges: low resource utilization due tosuboptimal configurations by users and the tendency to encounter abnormalitiesdue to an unstable cloud environment. To overcome them, we introduceDLRover-RM, an elastic training framework for DLRMs designed to increaseresource utilization and handle the instability of a cloud environment.DLRover-RM develops a resource-performance model by considering the uniquecharacteristics of DLRMs and a three-stage heuristic strategy to automaticallyallocate and dynamically adjust resources for DLRM training jobs for higherresource utilization. Further, DLRover-RM develops multiple mechanisms toensure efficient and reliable execution of DLRM training jobs. Our extensiveevaluation shows that DLRover-RM reduces job completion times by 31%, increasesthe job completion rate by 6%, enhances CPU usage by 15%, and improves memoryutilization by 20%, compared to state-of-the-art resource schedulingframeworks. DLRover-RM has been widely deployed at AntGroup and processesthousands of DLRM training jobs on a daily basis. DLRover-RM is open-sourcedand has been adopted by 10+ companies.

DLRover-RM: Resource Optimization for Deep Recommendation Models Training in the Cloud