AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Out of the Memory Barrier: A Highly Memory-Efficient Training System for LLMs with Million-Token Contexts

Wenhao Li, Daohai Yu, Gen Luo, Yuxin Zhang, Yifan Wu, Jiaxin Liu, Ziyang Gong, Zimu Liao, Fei Chao, Rongrong Ji

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use OOMB dense attention instead of FlashAttention baselineQwen2.5-7B training · 64K contextper-iteration latency: 76.1 -> 68.9Improved
add CPU offload (ablation)OOMB dense attention · 256K contextpeak memory (MB): 76650 -> 37129Improved
add CPU offload (ablation)OOMB dense attention · 256K contextper-iteration latency: 1015.9 -> 985.8No clear effect
use page-level sparse attention (4K+512) instead of dense (ablation)OOMB · 256K contextper-iteration latency: 1015.9 -> 75.7Improved
add CPU offload to sparse attention (ablation)OOMB sparse 4K+512 · 256K contextper-iteration latency: 75.7 -> 96.9Hurt
use OOMB sparse attention instead of Ring Flash Attentionlong-context training · 256K contextper-device throughput: 218.41 -> 1301.60Improved
use sparse attention with larger retrieval budget (ablation)OOMB · 16K-256K contexttraining loss: 1.50000 -> 1.32031Improved
use sparse attention instead of dense (ablation)OOMB · 1M contexttraining loss: 4.31250 -> 4.78125Hurt
chunk size above 4096 (ablation)OOMB · chunk size ablationthroughputNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy