DaiFu : In-Situ Crash Recovery for Deep Learning Systems DOI: XXXXXXX.XXXXXXX CCS: Software and its engineering Error handling and recovery

Zilong He, Pengfei Chen, Hongyu Zhang, Xiaoyun Li, Guangba Yu, Hongyang Chen, Zibin Zheng · 2025

Read DaiFu : In-Situ Crash Recovery for Deep Learning Systems DOI: XXXXXXX.XXXXXXX CCS: Software and its engineering Error handling and recovery by Zilong He, Pengfei Chen, Hongyu Zhang, Xiaoyun Li, Guangba Yu, Hongyang Chen, Zibin Zheng (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-2507-01628.txt.

Description

Zilong He and co-authors present DaiFu, a framework that lets deep learning systems recover from crashes in place, in a 2025 paper. Crashes are common in these systems because their software stacks are complex, and each crash wastes computing time. The standard fix, restarting from a checkpoint, is heavy and slow for small coding slips or brief glitches.

DaiFu adds a light code change that lets a running program catch its own crash. It can then patch code, settings, or data on the fly and resume at once. In tests it cut recovery time by a factor of more than a thousand against top rivals, with overhead under half a percent. A new benchmark covers seven crash types.

Catalog record

CREATORZilong He, Pengfei Chen, Hongyu Zhang, Xiaoyun Li, Guangba Yu, Hongyang Chen, Zibin Zheng
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-01628

Browse the full library · About the Tower of Babel