Saved in:
Bibliographic Details
Main Authors: Li, Wenhao, Yu, Daohai, Luo, Gen, Zhang, Yuxin, Chao, Fei, Ji, Rongrong, Wu, Yifan, Liu, Jiaxin, Gong, Ziyang, Liao, Zimu
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.02108
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917301777858560
author Li, Wenhao
Yu, Daohai
Luo, Gen
Zhang, Yuxin
Chao, Fei
Ji, Rongrong
Wu, Yifan
Liu, Jiaxin
Gong, Ziyang
Liao, Zimu
author_facet Li, Wenhao
Yu, Daohai
Luo, Gen
Zhang, Yuxin
Chao, Fei
Ji, Rongrong
Wu, Yifan
Liu, Jiaxin
Gong, Ziyang
Liao, Zimu
contents Training Large Language Models (LLMs) on long contexts is severely constrained by prohibitive GPU memory overhead, not training time. The primary culprits are the activations, whose memory footprints scale linearly with sequence length. We introduce OOMB, a highly memory-efficient training system that directly confronts this barrier. Our approach employs a chunk-recurrent training framework with on-the-fly activation recomputation, which maintains a constant activation memory footprint (O(1)) and shifts the primary bottleneck to the growing KV cache. To manage the KV cache, OOMB integrates a suite of synergistic optimizations: a paged memory manager for both the KV cache and its gradients to eliminate fragmentation, asynchronous CPU offloading to hide data transfer latency, and page-level sparse attention to reduce both computational complexity and communication overhead. The synergy of these techniques yields exceptional efficiency. Our empirical results show that for every additional 10K tokens of context, the end-to-end training memory overhead increases by a mere 10MB for Qwen2.5-7B. This allows training Qwen2.5-7B with a 4M-token context on a single H200 GPU, a feat that would otherwise require a large cluster using context parallelism. This work represents a substantial advance in resource efficiency for long-context LLM training. The source code is available at https://github.com/wenhaoli-xmu/OOMB.
format Preprint
id arxiv_https___arxiv_org_abs_2602_02108
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Out of the Memory Barrier: A Highly Memory Efficient Training System for LLMs with Million-Token Contexts
Li, Wenhao
Yu, Daohai
Luo, Gen
Zhang, Yuxin
Chao, Fei
Ji, Rongrong
Wu, Yifan
Liu, Jiaxin
Gong, Ziyang
Liao, Zimu
Computation and Language
Training Large Language Models (LLMs) on long contexts is severely constrained by prohibitive GPU memory overhead, not training time. The primary culprits are the activations, whose memory footprints scale linearly with sequence length. We introduce OOMB, a highly memory-efficient training system that directly confronts this barrier. Our approach employs a chunk-recurrent training framework with on-the-fly activation recomputation, which maintains a constant activation memory footprint (O(1)) and shifts the primary bottleneck to the growing KV cache. To manage the KV cache, OOMB integrates a suite of synergistic optimizations: a paged memory manager for both the KV cache and its gradients to eliminate fragmentation, asynchronous CPU offloading to hide data transfer latency, and page-level sparse attention to reduce both computational complexity and communication overhead. The synergy of these techniques yields exceptional efficiency. Our empirical results show that for every additional 10K tokens of context, the end-to-end training memory overhead increases by a mere 10MB for Qwen2.5-7B. This allows training Qwen2.5-7B with a 4M-token context on a single H200 GPU, a feat that would otherwise require a large cluster using context parallelism. This work represents a substantial advance in resource efficiency for long-context LLM training. The source code is available at https://github.com/wenhaoli-xmu/OOMB.
title Out of the Memory Barrier: A Highly Memory Efficient Training System for LLMs with Million-Token Contexts
topic Computation and Language
url https://arxiv.org/abs/2602.02108