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/

Global: Zendy Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Modern machine learning workloads use large models, with complex structures,that are very expensive to execute. The devices that execute complex models arebecoming increasingly heterogeneous as we see a flourishing of domain-specificaccelerators being offered as hardware accelerators in addition to CPUs. Thesetrends necessitate distributing the workload across multiple devices. Recentwork has shown that significant gains can be obtained with model parallelism,i.e, partitioning a neural network's computational graph onto multiple devices.In particular, this form of parallelism assumes a pipeline of devices, which isfed a stream of samples and yields high throughput for training and inferenceof DNNs. However, for such settings (large models and multiple heterogeneousdevices), we require automated algorithms and toolchains that can partition theML workload across devices. In this paper, we identify and isolate thestructured optimization problem at the core of device placement of DNNoperators, for both inference and training, especially in modern pipelinedsettings. We then provide algorithms that solve this problem to optimality. Wedemonstrate the applicability and efficiency of our approaches using severalcontemporary DNN computation graphs.

Efficient Algorithms for Device Placement of DNN Graph Operators