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

Leveraging recent advancements in large language models, modern neural codecompletion models have demonstrated the capability to generate highly accuratecode suggestions. However, their massive size poses challenges in terms ofcomputational costs and environmental impact, hindering their widespreadadoption in practical scenarios. Dynamic inference emerges as a promisingsolution, as it allocates minimal computation during inference whilemaintaining the model's performance. In this research, we explore dynamicinference within the context of code completion. Initially, we conducted anempirical investigation on GPT-2, focusing on the inference capabilities ofintermediate layers for code completion. We found that 54.4% of tokens can beaccurately generated using just the first layer, signifying significantcomputational savings potential. Moreover, despite using all layers, the modelstill fails to predict 14.5% of tokens correctly, and the subsequentcompletions continued from them are rarely considered helpful, with only a 4.2%Acceptance Rate. These findings motivate our exploration of dynamic inferencein code completion and inspire us to enhance it with a decision-makingmechanism that stops the generation of incorrect code. We thus propose a noveldynamic inference method specifically tailored for code completion models. Thismethod aims not only to produce correct predictions with largely reducedcomputation but also to prevent incorrect predictions proactively. Ourextensive evaluation shows that it can averagely skip 1.7 layers out of 16layers in the models, leading to an 11.2% speedup with only a marginal 1.1%reduction in ROUGE-L.

When Neural Code Completion Models Size up the Situation: Attaining Cheaper and Faster Completion through Dynamic Model Inference