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

Auto-regressive large language models such as GPT-3 require enormouscomputational resources to use. Traditionally, structured pruning methods areemployed to reduce resource usage. However, their application to and efficacyfor generative language models is heavily under-explored. In this paper weconduct an comprehensive evaluation of common structured pruning methods,including magnitude, random, and movement pruning on the feed-forward layers inGPT-type models. Unexpectedly, random pruning results in performance that iscomparable to the best established methods, across multiple natural languagegeneration tasks. To understand these results, we provide a framework formeasuring neuron-level redundancy of models pruned by different methods, anddiscover that established structured pruning methods do not take into accountthe distinctiveness of neurons, leaving behind excess redundancies. In view ofthis, we introduce Globally Unique Movement (GUM) to improve the uniqueness ofneurons in pruned models. We then discuss the effects of our techniques ondifferent redundancy metrics to explain the improved performance.

What Matters In The Structured Pruning of Generative Language Models?