Official PDF Translation•Engineering Information Technology & Electronic Engineering

FastCheck: fast checkpointing and recovery for DNN training via parallel transmission and compression

Authors: Yun TENG; Dawei SUN; Shipeng HU; Zhiyue LI; Guangyan ZHANG; Haidong TIAN; Rui CHANG

DOI: 10.1631/ENG_ITEE_2025_0034Status: Verified Translated Edition

Sponsored AdvertisementAd Placement Area

reCAPTCHA Bot Shield Active

Preparing Secure Academic Download

Verifying human reader & generating high-resolution document...

Verifying Document Integrity15s remaining

← Back to Article

Protected by Google reCAPTCHA v3.•Privacy•Terms

Sponsored ContentAdSense In-Feed Ad Slot

Key Findings in This Report

• FastCheck reduces checkpointing time by up to 78.42% and recovery time by up to 77.41% compared to baselines. • It partitions checkpoints into shards and leverages multiple nodes for parallel transmission and recovery. • Delta compression for weights and index compression for momentum reduce checkpoint size and overhead. • Lightweight health status maintenance prevents checkpoint transmission to failed nodes, improving reliability.