Matches in SemOpenAlex for { <https://semopenalex.org/work/W3022719076> ?p ?o ?g. }
- W3022719076 abstract "Large transformer-based language models have been shown to be very effective in many classification tasks. However, their computational complexity prevents their use in applications requiring the classification of a large set of candidates. While previous works have investigated approaches to reduce model size, relatively little attention has been paid to techniques to improve batch throughput during inference. In this paper, we introduce the Cascade Transformer, a simple yet effective technique to adapt transformer-based models into a cascade of rankers. Each ranker is used to prune a subset of candidates in a batch, thus dramatically increasing throughput at inference time. Partial encodings from the transformer model are shared among rerankers, providing further speed-up. When compared to a state-of-the-art transformer model, our approach reduces computation by 37% with almost no impact on accuracy, as measured on two English Question Answering datasets." @default.
- W3022719076 created "2020-05-13" @default.
- W3022719076 creator A5056376686 @default.
- W3022719076 creator A5060844217 @default.
- W3022719076 date "2020-05-05" @default.
- W3022719076 modified "2023-09-27" @default.
- W3022719076 title "The Cascade Transformer: an Application for Efficient Answer Sentence Selection" @default.
- W3022719076 cites W1522301498 @default.
- W3022719076 cites W1821462560 @default.
- W3022719076 cites W1966443646 @default.
- W3022719076 cites W2000431947 @default.
- W3022719076 cites W2091364465 @default.
- W3022719076 cites W2091379987 @default.
- W3022719076 cites W2101210369 @default.
- W3022719076 cites W2120735855 @default.
- W3022719076 cites W2130618701 @default.
- W3022719076 cites W2132324454 @default.
- W3022719076 cites W2133556223 @default.
- W3022719076 cites W2138382875 @default.
- W3022719076 cites W2186615578 @default.
- W3022719076 cites W2251818205 @default.
- W3022719076 cites W2338364780 @default.
- W3022719076 cites W2760753016 @default.
- W3022719076 cites W2767857566 @default.
- W3022719076 cites W2803820154 @default.
- W3022719076 cites W2806019191 @default.
- W3022719076 cites W2865914962 @default.
- W3022719076 cites W2891602716 @default.
- W3022719076 cites W2908332126 @default.
- W3022719076 cites W2912924812 @default.
- W3022719076 cites W2921848006 @default.
- W3022719076 cites W2923890923 @default.
- W3022719076 cites W2940744433 @default.
- W3022719076 cites W2948947170 @default.
- W3022719076 cites W2951528484 @default.
- W3022719076 cites W2960289873 @default.
- W3022719076 cites W2962776659 @default.
- W3022719076 cites W2962777840 @default.
- W3022719076 cites W2963326042 @default.
- W3022719076 cites W2963341956 @default.
- W3022719076 cites W2963403868 @default.
- W3022719076 cites W2963430224 @default.
- W3022719076 cites W2963854351 @default.
- W3022719076 cites W2964054038 @default.
- W3022719076 cites W2964110616 @default.
- W3022719076 cites W2964213727 @default.
- W3022719076 cites W2964303116 @default.
- W3022719076 cites W2965046076 @default.
- W3022719076 cites W2965373594 @default.
- W3022719076 cites W2970120757 @default.
- W3022719076 cites W2970597249 @default.
- W3022719076 cites W2972324944 @default.
- W3022719076 cites W2973727699 @default.
- W3022719076 cites W2975059944 @default.
- W3022719076 cites W2978017171 @default.
- W3022719076 cites W2997090102 @default.
- W3022719076 cites W3008618223 @default.
- W3022719076 cites W3102286003 @default.
- W3022719076 hasPublicationYear "2020" @default.
- W3022719076 type Work @default.
- W3022719076 sameAs 3022719076 @default.
- W3022719076 citedByCount "1" @default.
- W3022719076 countsByYear W30227190762022 @default.
- W3022719076 crossrefType "posted-content" @default.
- W3022719076 hasAuthorship W3022719076A5056376686 @default.
- W3022719076 hasAuthorship W3022719076A5060844217 @default.
- W3022719076 hasConcept C11413529 @default.
- W3022719076 hasConcept C119599485 @default.
- W3022719076 hasConcept C119857082 @default.
- W3022719076 hasConcept C127413603 @default.
- W3022719076 hasConcept C137293760 @default.
- W3022719076 hasConcept C154945302 @default.
- W3022719076 hasConcept C165801399 @default.
- W3022719076 hasConcept C2776214188 @default.
- W3022719076 hasConcept C2777530160 @default.
- W3022719076 hasConcept C34146451 @default.
- W3022719076 hasConcept C41008148 @default.
- W3022719076 hasConcept C42360764 @default.
- W3022719076 hasConcept C44291984 @default.
- W3022719076 hasConcept C45374587 @default.
- W3022719076 hasConcept C66322947 @default.
- W3022719076 hasConceptScore W3022719076C11413529 @default.
- W3022719076 hasConceptScore W3022719076C119599485 @default.
- W3022719076 hasConceptScore W3022719076C119857082 @default.
- W3022719076 hasConceptScore W3022719076C127413603 @default.
- W3022719076 hasConceptScore W3022719076C137293760 @default.
- W3022719076 hasConceptScore W3022719076C154945302 @default.
- W3022719076 hasConceptScore W3022719076C165801399 @default.
- W3022719076 hasConceptScore W3022719076C2776214188 @default.
- W3022719076 hasConceptScore W3022719076C2777530160 @default.
- W3022719076 hasConceptScore W3022719076C34146451 @default.
- W3022719076 hasConceptScore W3022719076C41008148 @default.
- W3022719076 hasConceptScore W3022719076C42360764 @default.
- W3022719076 hasConceptScore W3022719076C44291984 @default.
- W3022719076 hasConceptScore W3022719076C45374587 @default.
- W3022719076 hasConceptScore W3022719076C66322947 @default.
- W3022719076 hasLocation W30227190761 @default.
- W3022719076 hasOpenAccess W3022719076 @default.
- W3022719076 hasPrimaryLocation W30227190761 @default.
- W3022719076 hasRelatedWork W1544902885 @default.