paper-with-me

홈 › Papers

DeepNet: Scaling Transformers to 1,000 Layers

2022-03-01 · Hongyu Wang, Shuming Ma, Li Dong, Shaohan Huang, Dongdong Zhang, Furu Wei

In this paper, we propose a simple yet effective method to stabilize extremely deep Transformers. Specifically, we introduce a new normalization function (DeepNorm) to modify the residual connection in Transformer, accompanying with theoretically derived initialization. In-depth theoretical analysis shows that model updates can be bounded in a stable way. The proposed method combines the best of two worlds, i.e., good performance of Post-LN and stable training of Pre-LN, making DeepNorm a preferred alternative. We successfully scale Transformers up to 1,000 layers (i.e., 2,500 attention and feed-forward network sublayers) without difficulty, which is one order of magnitude deeper than previous deep Transformers. Remarkably, on a multilingual benchmark with 7,482 translation directions, our 200-layer model with 3.2B parameters significantly outperforms the 48-layer state-of-the-art model with 12B parameters by 5 BLEU points, which indicates a promising scaling direction.

📄 PDF Abstract BibTeX arXiv:2203.00555

Code (6)

microsoft/unilm 공식 구현 pytorch
facebookresearch/xformers pytorch
labmlai/annotated_deep_learning_paper_implementations pytorch
lucidrains/RETRO-pytorch pytorch
microsoft/torchscale pytorch
tyy2022/usst_autosimultrans2022 paddle

Tasks

Translation

Methods 이 논문이 사용한 방법론

Multi-Head Attention 설명 없음
Attention 설명 없음
Linear Layer A Linear Layer is a projection $\mathbf{XW + b}$.
Dense Connections Dense Connections, or Fully Connected Connections, are a type of layer in a deep neural network that use a linear operation where every input is connected to every output…
Softmax The Softmax output function transforms a previous layer's output into a vector of probabilities. It is commonly used for multiclass classification. Given an input vector $x$…
Absolute Position Encodings Absolute Position Encodings are a type of position embeddings for [Transformer-based models] where positional encodings are…
BPE Byte Pair Encoding, or BPE, is a subword segmentation algorithm that encodes rare and unknown words as sequences of subword units. The intuition is that various word…
Position-Wise Feed-Forward Layer 설명 없음

Similar Papers 제목 키워드 기반

Foundation Transformers

2022-10-12 · Hongyu Wang, Shuming Ma, Shaohan Huang, Li Dong 외

A big convergence of model architectures across language, vision, speech, and multimodal is emerging. However, under the same name "Transformers", the above areas use different implementations for better performance, e.g…

Language ModelingLanguage ModellingMachine TranslationSpeech Recognition

ADHDeepNet From Raw EEG to Diagnosis: Improving ADHD Diagnosis through Temporal-Spatial Processing, Adaptive Attention Mechanisms, and Explainability in Raw EEG Signals

2025-09-10 · Ali Amini, Mohammad Alijanpour, Behnam Latifi, Ali Motie Nasrabadi arxiv

Attention Deficit Hyperactivity Disorder (ADHD) is a common brain disorder in children that can persist into adulthood, affecting social, academic, and career life. Early diagnosis is crucial for managing these impacts o…

Data Augmentation

Diversity of Transformer Layers: One Aspect of Parameter Scaling Laws

2025-05-29 · Hidetaka Kamigaito, Ying Zhang, Jingun Kwon, Katsuhiko Hayashi 외

Transformers deliver outstanding performance across a wide range of tasks and are now a dominant backbone architecture for large language models (LLMs). Their task-solving performance is improved by increasing parameter …

Diversity

Nearest Class-Center Simplification through Intermediate Layers

2022-01-21 · Ido Ben-Shaul, Shai Dekel

Recent advances in theoretical Deep Learning have introduced geometric properties that occur during training, past the Interpolation Threshold -- where the training error reaches zero. We inquire into the phenomena coine…

Language ModelingLanguage Modelling

On the Residual Scaling of Looped Transformers: Stability and Transferability

2026-06-16 · Shaowen Wang, Bingrui Li, Ge Zhang, Wenhao Huang 외 arxiv

Looped (weight-tied) Transformers apply a shared residual block $N$ times ($h \leftarrow h + \varepsilon\,f(h)$, same $f$ at each step), increasing effective depth without adding parameters. Prior depth-scaling analyses …