Saved in:
Bibliographic Details
Main Authors: Zhao, Bohan, Wang, Yuanhong, Liu, Chenglin, Pan, Jiagi, Yang, Guang, Liu, Ruitao, Zhang, Tingrui, Luo, Kai, Xu, Wei
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.03644
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911300122050560
author Zhao, Bohan
Wang, Yuanhong
Liu, Chenglin
Pan, Jiagi
Yang, Guang
Liu, Ruitao
Zhang, Tingrui
Luo, Kai
Xu, Wei
author_facet Zhao, Bohan
Wang, Yuanhong
Liu, Chenglin
Pan, Jiagi
Yang, Guang
Liu, Ruitao
Zhang, Tingrui
Luo, Kai
Xu, Wei
contents Recent developments in large language models (LLMs) have introduced new requirements for efficient and robust training. As LLM clusters scale, node failures, lengthy recoveries, and bulky checkpoints erode efficiency. Infrequent asynchronous checkpoints trigger costly rollbacks, yet higher frequencies add prohibitive overhead. To address these challenges, we propose FFTrainer, a system designed for robust LLM training. FFTrainer leverages surplus network capacity to quickly save and load states, thereby preventing rollbacks and accelerating recovery. Compared with prior checkpointing approaches, FFTrainer reduces recovery time by up to 98% and mitigates GPU utilization loss by up to 68% without hindering normal training.
format Preprint
id arxiv_https___arxiv_org_abs_2512_03644
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle FFTrainer: Fast Failover in Large-Language Model Training with Almost-Free State Management
Zhao, Bohan
Wang, Yuanhong
Liu, Chenglin
Pan, Jiagi
Yang, Guang
Liu, Ruitao
Zhang, Tingrui
Luo, Kai
Xu, Wei
Distributed, Parallel, and Cluster Computing
Recent developments in large language models (LLMs) have introduced new requirements for efficient and robust training. As LLM clusters scale, node failures, lengthy recoveries, and bulky checkpoints erode efficiency. Infrequent asynchronous checkpoints trigger costly rollbacks, yet higher frequencies add prohibitive overhead. To address these challenges, we propose FFTrainer, a system designed for robust LLM training. FFTrainer leverages surplus network capacity to quickly save and load states, thereby preventing rollbacks and accelerating recovery. Compared with prior checkpointing approaches, FFTrainer reduces recovery time by up to 98% and mitigates GPU utilization loss by up to 68% without hindering normal training.
title FFTrainer: Fast Failover in Large-Language Model Training with Almost-Free State Management
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2512.03644