Making Asynchronous Stochastic Gradient Descent Work for Transformers
Asynchronous stochastic gradient descent (SGD) is attractive from a speed perspective because workers do not wait for synchronization. However, the Transformer model converges poorly with asynchronous SGD, resulting in substantially lower quality compared to synchronous SGD. To investigate why this is the case, we isolate differences between asynchronous and synchronous methods to investigate batch size and staleness effects. We find that summing several asynchronous updates, rather than applying them immediately, restores convergence behavior. With this hybrid method, Transformer training for neural machine translation task reaches a near-convergence level 1.36x faster in single-node multi-GPU training with no impact on model quality.
Code (0)
등록된 구현이 없습니다.
Tasks
GPUMachine TranslationTranslationMethods 이 논문이 사용한 방법론
Similar Papers 제목 키워드 기반
Accelerating Asynchronous Stochastic Gradient Descent for Neural Machine Translation
In order to extract the best possible performance from asynchronous stochastic gradient descent one must increase the mini-batch size and scale the learning rate accordingly. In order to achieve further speedup we introd…
Machine TranslationTranslationDecoupled Asynchronous Proximal Stochastic Gradient Descent with Variance Reduction
In the era of big data, optimizing large scale machine learning problems becomes a challenging task and draws significant attention. Asynchronous optimization algorithms come out as a promising solution. Recently, decoup…
Asynchronous Stochastic Gradient Descent with Variance Reduction for Non-Convex Optimization
We provide the first theoretical analysis on the convergence rate of the asynchronous stochastic variance reduced gradient (SVRG) descent algorithm on non-convex optimization. Recent studies have shown that the asynchron…
Stochastic modified equations for the asynchronous stochastic gradient descent
We propose a stochastic modified equations (SME) for modeling the asynchronous stochastic gradient descent (ASGD) algorithms. The resulting SME of Langevin type extracts more information about the ASGD dynamics and eluci…
Asynchronous decentralized accelerated stochastic gradient descent
In this work, we introduce an asynchronous decentralized accelerated stochastic gradient descent type of method for decentralized stochastic optimization, considering communication and synchronization are the major bottl…
Stochastic Optimization