A Comprehensive Analysis of Adapter Efficiency.
- Nandini Mundra,
- Sumanth Doddapaneni,
- Raj Dabre,
- Anoop Kunchukuttan,
- ,
- Mitesh M. Khapra
- Indian Institute of Technology Madras,
- National Institute Of Information And Communications Technology, Japan,
- Microsoft India,
- Agency for Science, Technology and Research (A*Star)
Publikation:
Konference artikel i Proceeding eller bog/rapport kapitel
Konferencebidrag i proceedings
Forskning
Peer-reviewOpen Access
Resume
Adapters have been positioned as a parameter-efficient fine-tuning (PEFT) approach, whereby a minimal number of parameters are added to the model and fine-tuned. However, adapters have not been sufficiently analyzed to understand if PEFT translates to benefits in training/deployment efficiency and maintainability/extensibility. Through extensive experiments on many adapters, tasks, and languages in supervised and cross-lingual zero-shot settings, we clearly show that for Natural Language Understanding (NLU) tasks, the parameter efficiency in adapters does not translate to efficiency gains compared to full fine-tuning of models. More precisely, adapters are relatively expensive to train and have slightly higher deployment latency. Furthermore, the maintainability/extensibility benefits of adapters can be achieved with simpler approaches like multi-task training via full fine-tuning, which also provide relatively faster training times. We, therefore, recommend that for moderately sized models for NLU tasks, practitioners should rely on full fine-tuning or multi-task training rather than using adapters. Our code is available at https://github.com/AI4Bharat/adapter-efficiency.
Publikation information
Produktionstype
Publikation:
Konference artikel i Proceeding eller bog/rapport kapitel
Konferencebidrag i proceedings
Forskning
Peer-reviewOriginalsprog
EngelskSider fra-til (Antal sider)
Sider 136-154 (19 sider)Publikationsmilepæle
- Udgivet - 2024
Publikationsstatus
Udgivet - 2024
Forlag
Association for Computing Machinery, USAPublikations IDer
- Scopus: 85183581297
Titel på værtspublikation
CODS-COMAD '24: Proceedings of the 7th Joint International Conference on Data Science & Management of Data (11th ACM IKDD CODS and 29th COMAD)Metrikker
PlumX, åbner i en ny fane
Hentninger
17
Antal citationer
13
Relateret event
Titel
International Conference on Data Science & Management of Data
Begivenhedstype
KonferenceGrad af anerkendelse
International begivenhedDato
04/01/2024 - 07/01/2024Lokation
BangaloreIndien
