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

Accelerating deep model training and inference is crucial in practice.Existing deep learning frameworks usually concentrate on optimizing trainingspeed and pay fewer attentions to inference-specific optimizations. Actually,model inference differs from training in terms of computation, e.g. parametersare refreshed each gradient update step during training, but kept invariantduring inference. These special characteristics of model inference open newopportunities for its optimization. In this paper, we propose a hardware-awareoptimization framework, namely Woodpecker-DL (WPK), to accelerate inference bytaking advantage of multiple joint optimizations from the perspectives of graphoptimization, automated searches, domain-specific language (DSL) compilertechniques and system-level exploration. In WPK, we investigated two newautomated search approaches based on genetic algorithm and reinforcementlearning, respectively, to hunt the best operator code configurations targetingspecific hardware. A customized DSL compiler is further attached to thesesearch algorithms to generate efficient codes. To create an optimized inferenceplan, WPK systematically explores high-speed operator implementations fromthird-party libraries besides our automatically generated codes and singles outthe best implementation per operator for use. Extensive experimentsdemonstrated that on a Tesla P100 GPU, we can achieve the maximum speedup of5.40 over cuDNN and 1.63 over TVM on individual convolution operators, and runup to 1.18 times faster than TensorRT for end-to-end model inference.

Woodpecker-DL: Accelerating Deep Neural Networks via Hardware-Aware Multifaceted Optimizations