Symbiosis: Multi-Adapter Inference and Fine-Tuning Conference: ; ;

Saransh Gupta Umesh Deshpande Travis Janssen Swaminathan Sundararaman IBM Research, USA · 2025

Read Symbiosis: Multi-Adapter Inference and Fine-Tuning Conference: ; ; by Saransh Gupta Umesh Deshpande Travis Janssen Swaminathan Sundararaman IBM Research, USA (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-03220.txt.

Description

Saransh Gupta, Umesh Deshpande, Travis Janssen, and Swaminathan Sundararaman build a system for serving many model adapters at once in a 2025 study. Adapters are small add-ons that specialize a large model for one task, and popular models have spawned thousands of them. Existing tools waste resources: fine-tuning each adapter demands its own copy of the base model, and inference stacks handle multiple adapters poorly.

Symbiosis lets several adapters share one base model for both training and serving. Sharing cuts memory and cost wherever fleets of specialists run, from enterprise copilots to per-user personalization.

Catalog record

CREATORSaransh Gupta Umesh Deshpande Travis Janssen Swaminathan Sundararaman IBM Research, USA
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE60 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-03220

Browse the full library · About the Tower of Babel