AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

DeepNet: Scaling Transformers to 1,000 Layers

Hongyu Wang, Shuming Ma, Li Dong, Shaohan Huang, Dongdong Zhang, Furu Wei

TPAMI 2024 · Architectures and efficiency › Attention and transformers · approach: Transformer block and depth design

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
scale depth from 12 to 48 layers (ablation)baseline multilingual NMT · OPUS-100average BLEU: 24.5 -> 27.7Improved
scale DeepNet depth from 200 to 1000 layers (ablation)DeepNet · OPUS-100average BLEU: 31.1 -> 32.1Improved
use DeepNorm at 1000 layers instead of 48-layer baselinemultilingual NMT · OPUS-100average BLEU: 27.7 -> 32.1Improved
use deep narrow DeepNet 200L 3.2B instead of M2M-100 48L 12Bmultilingual NMT · FloresBLEU: 13.6 -> 18.6Improved
use DeepNet instead of M2M-100multilingual NMT · WMTBLEU: 31.9 -> 33.9Improved
use DeepNet instead of M2M-100multilingual NMT · OPUSBLEU: 18.4 -> 23.0Improved
use DeepNet instead of M2M-100multilingual NMT · TEDBLEU: 18.7 -> 20.1Improved
use DeepNorm instead of Post-LN at 100L-100LTransformer · WMT-17 En-DeBLEU: diverged -> 28.9Improved
use DeepNorm instead of Pre-LN at 50L-50LTransformer · WMT-17 En-DeBLEU: 28.0 -> 29.0Improved
use DeepNorm instead of Post-LN at 6L-6LTransformer · WMT-17 En-DeBLEU: 28.1 -> 27.8No clear effect
increase DeepNet depth from 18L-18L to 50L-50L (ablation)DeepNet · WMT-17 En-DeBLEU: 28.8 -> 29.0Improved
increase R-Fixup depth from 18L-18L to 50L-50L (ablation)R-Fixup (no LN) · WMT-17 En-DeBLEU: 28.4 -> 27.7Hurt

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