AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Magneto: A Foundation Transformer

Hongyu Wang, Shuming Ma, Shaohan Huang, Li Dong, Wenhui Wang, Zhiliang Peng, Yu Wu, Payal Bajaj, Saksham Singhal, Alon Benhaim, Barun Patra, Zhun Liu and 3 more

ICML 2023 · Architectures and efficiency › Attention and transformers · approach: Transformer block and depth design

Read the paper

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
Sub-LN plus DeepNet-style initialization instead of Pre-LNGPT-style causal LM (72L) · Zero-shot, one-shot, four-shot in-context learningAvg. accuracy: +1.7 zero-shot vs Pre-LN (65.8 vs 64.1); +0.4 vs Normformer zero-shotImproved
Sub-LN plus DeepNet-style initializationBERT-base masked LM · GLUE devAvg. score: 86.3 vs 85.7 Post-LN, 84.6 Pre-LNImproved
Sub-LN plus DeepNet-style initialization18L-18L encoder-decoder NMT · OPUS-100BLEU: +0.5 over Pre-LN (31.0 vs 30.5)Improved
Post-LN instead of Sub-LN in 18L-18L NMTencoder-decoder NMT · OPUS-100BLEU: did not convergeHurt
Sub-LN plus DeepNet-style initialization for BEiT-style ViTViT-Base and ViT-Large (masked image modeling) · ImageNet-1k val top-1 and ADE20k mIoUtop-1 accuracy / mIoU: +0.4 top-1 at 12L, +0.6 at 24L; +0.8 mIoU at 12L, +0.4 at 24LImproved
Sub-LN plus DeepNet-style initializationTransformer Transducer on LibriSpeech · LibriSpeech 960h without LM, 18L and 36LWER (lower is better): test-other 6.16 vs 6.62 (18L); 5.56 vs 6.04 (36L)Improved
Sub-LN plus DeepNet-style initializationBEiT-3-style vision-language pretraining (24L base) · VQA test-dev and NLVR2 test-PVQA score / accuracy: +0.63 VQA test-dev; +0.54 NLVR2 test-PImproved
Higher learning rate enabled by Sub-LN+DeepNet init (1e-3 vs diverged Pre-LN and Normformer) at 24L (ablation)GPT-style causal LM 24L · Zero-shot avg accuracyAvg. accuracy: MAGNETO 61.4 at lr 1e-3; Pre-LN and Normformer divergedImproved
Sub-LN plus DeepNet-style initialization with larger learning rate (36L LibriSpeech)Transformer Transducer 36L · LibriSpeech training stabilitypeak learning rate: MAGNETO up to 3e-3 vs Transformer 1.5e-3Improved

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