zFLoRA: Zero-Latency Fused Low-Rank Adapters
Dhananjaya Gowda, Seoha SongHarshith GokaJunhyun LeeSamsung Research {d.gowda, seoha.song, h9399.goka, junhyun8.lee}@samsung.com · 2025
Read zFLoRA: Zero-Latency Fused Low-Rank Adapters by Dhananjaya Gowda, Seoha SongHarshith GokaJunhyun LeeSamsung Research {d.gowda, seoha.song, h9399.goka, junhyun8.lee}@samsung.com (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-2510-25784.txt.
Description
Dhananjaya Gowda, Seoha Song, Harshith Goka, and Junhyun Lee propose zFLoRA. Their 2025 paper fuses low-rank adapters with zero latency. Large language models deploy with task-certain adapters widely. Adapters serve multiple downstream uses simultaneously. Adapter settings are under one percent of base model. Yet their inference compute is disproportionately great. Overhead can reach two and a half times base.
Their zero-latency fused low-rank adapter removes that overhead. Zero or negligible latency is added over base model. Tests used three model sizes and standard rivals. Eighteen tasks across three categories were tested. Commonsense thinking, math thinking, and summary-dialogue categories appeared. Latency readings ran on phone and GPU platforms. A Samsung Galaxy device and NVIDIA H100 were used. Proposed adapters present zero to negligible latency overhead. Favorable quality comparisons against rivals were found. Quick multi-adapter deployment becomes more useful on devices.
Catalog record
| CREATOR | Dhananjaya Gowda, Seoha SongHarshith GokaJunhyun LeeSamsung Research {d.gowda, seoha.song, h9399.goka, junhyun8.lee}@samsung.com |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-25784