AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

TawPipe: Topology-Aware Weight Pipeline Parallelism for Accelerating Long-Context Large Models Training

Houming Wu, Ling Chen

AAAI 2026 · Architectures and efficiency › RNNs and normalization · approach: Distributed LLM inference systems · 0 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
add group-based weight pipeline scheduler (GWPS) (ablation)TawPipe · C4, 24 GPUs, H=1024, S=16384throughput (kilo tokens/s): 8.59 (-3.6%) -> 8.91Improved
add communication-computation overlap (CCO) (ablation)TawPipe · C4, 24 GPUs, H=1024, S=16384throughput (kilo tokens/s): 8.22 (-7.7%) -> 8.91Improved
add group-based weight pipeline scheduler (GWPS) (ablation)TawPipe · C4, 24 GPUs, H=2048, S=16384throughput (kilo tokens/s): 3.91 (-6.5%) -> 4.18Improved
add communication-computation overlap (CCO) (ablation)TawPipe · C4, 24 GPUs, H=2048, S=16384throughput (kilo tokens/s): 3.47 (-17.0%) -> 4.18Improved
add group-based weight pipeline scheduler (GWPS) (ablation)TawPipe · C4, 24 GPUs, H=4096, S=16384throughput (kilo tokens/s): 1.26 (-8.7%) -> 1.38Improved
add communication-computation overlap (CCO) (ablation)TawPipe · C4, 24 GPUs, H=4096, S=16384throughput (kilo tokens/s): 1.14 (-17.4%) -> 1.38Improved
use device-bound storage instead of ring-based weight buffersTawPipe · weight buffer memoryweight buffer memory: 2M_W -> M_WImproved
use TawPipe instead of WeiPipe as hierarchy-aware weight passingTawPipe vs WeiPipe · C4, 24 GPUs, S=16384throughput: 6.0% to 11.8% as H scales from 1024 to 4096Improved
use TawPipe instead of WeiPipe for communicationTawPipe vs WeiPipe · C4, 24 GPUs, H=1024, S=16384NCCL execution time: up to 82.1%Improved
use TawPipe instead of FSDP for communicationTawPipe vs FSDP · C4, 24 GPUs, H=1024, S=16384NCCL execution time: 16.8%Improved
use TawPipe instead of 1F1B or FSDP at the most demanding configurationTawPipe vs 1F1B/FSDP · C4, 24 GPUs, H=4096, S=16384throughput: 23.6%, 44.1%Improved

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