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

1-bit gradient compression and local steps are two representative techniquesthat enable drastic communication reduction in distributed SGD. Their benefits,however, remain an open question on Adam-based large model pre-training (e.g.BERT and GPT). In this paper, we demonstrate the non-linearity in Adam causesslow convergence even when 1-bit compression or local steps are individuallyapplied. To alleviate this limitation, we propose 0/1 Adam that linearizes eachAdam step via approximating its optimizer states using their stale estimatesand linear correlation. 0/1 Adam performs an Adam-like step to preserve theadaptivity, while its linearity allows utilizing 1-bit compression and localsteps simultaneously for wall-clock time speed up. We provide convergenceguarantee for 0/1 Adam on smooth non-convex objectives. On various large-scalebenchmarks such as BERT-Base, BERT-Large, GPT-2 pre-training and ImageNet, wedemonstrate on up to 128 GPUs that 0/1 Adam is able to reduce up to 87% of datavolume, 54% of communication rounds, and achieve up to 2$\times$ highertraining throughput and end-to-end training time reduction compared to thestate-of-the-art baseline 1-bit Adam; while enjoying the same statisticalconvergence speed and end task model accuracy on GLUE dataset and ImageNetvalidation set.

Maximizing Communication Efficiency for Large-scale Training via 0/1 Adam