AdaSplash : Adaptive Sparse Flash Attention
Nuno Gonçalves, Marcos Treviso, André F. T. Martins · 2025
Read AdaSplash : Adaptive Sparse Flash Attention by Nuno Gonçalves, Marcos Treviso, André F. T. Martins (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-2502-12082.txt.
Description
Nuno Gonçalves, Marcos Treviso, and André Martins speed sparse gaze in tongue nets with a fit flash build, in a 2025 paper. Soft gaze costs dear on long texts, which caps net use there. Sparse gaze bends by data, but past builds waste its gaps and gain no speed.
A hybrid root hunt cuts the sparse shift solves by a vast fold. Hand built chip cores then skip the null gaze lots in runs. Tests span class, fetch, and tongue tale tasks on known nets. Run time and chip store both drop hard past past sparse builds. Speed nears the best dense builds while task marks hold.
Catalog record
| CREATOR | Nuno Gonçalves, Marcos Treviso, André F. T. Martins |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-12082