AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

SAPipe: Staleness-Aware Pipeline for Data Parallel DNN Training

Yangrui Chen, Cong Xie, Meng Ma, Juncheng Gu, Yanghua Peng, Haibin Lin, Chuan Wu, Yibo Zhu

NeurIPS 2022 · 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 staleness compensation (SAPipe) over 1-step stale PipeSGDPipeSGD · Transformer Multi30Kaccuracy: 0.531 -> 0.668Improved
add staleness compensation over PipeSGDPipeSGD · ResNet50 CIFAR-10accuracy: 0.893 -> 0.932Improved
add staleness compensation over PipeSGDPipeSGD · VGG16 CIFAR-10accuracy: 0.906 -> 0.926Improved
add WP-OPT3 over PipeSGDPipeSGD · GPT-2 WikiText-2perplexity: 22.36 -> 20.23Improved
use delay compensation only (SAPipe-DC) instead of PipeSGDPipeSGD · Transformer Multi30Kaccuracy: 0.531 -> 0.526No clear effect
WP-OPT2 instead of PipeSGDPipeSGD · GPT-2 WikiText-2perplexity: 22.36 -> 22.60Hurt
WP-OPT1 vs OPT3 option choiceSAPipe · ImageNet VGG16accuracy: 0.718 -> 0.735Mixed
SAPipe-DC versus non-stale BytePSSAPipe-DC · ResNet50 CIFAR-10accuracy: 0.932 -> 0.894Hurt
SAPipe-WP-OPT3 versus BytePSSAPipe-WP-OPT3 · VGG16 ImageNetaccuracy: 0.731 -> 0.735Improved
SAPipe throughput versus BytePSBytePS · throughput, 8-64 GPUsspeedup: up to 57%Improved
SAPipe throughput versus HorovodHorovod · throughput, 8-64 GPUsspeedup: up to 203%Improved
remove runtime optimization (kernel fusion, buffers) (ablation)SAPipe · throughput, 32 GPUstraining speed: decreases the training speed by 7% to 24%Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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