AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Quantizable Transformers: Removing Outliers by Helping Attention Heads Do Nothing

Yelysei Bondarenko, Markus Nagel, Tijmen Blankevoort

NeurIPS 2023 · Architectures and efficiency › Attention and transformers · approach: Transformer expressivity and learnability · 11 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 clipped softmax with gamma<0BERT-base · BERT Wikipedia valW8A8 perplexity: 1294 -> 4.52Improved
add gated attentionBERT-base · BERT Wikipedia valW8A8 perplexity: 1294 -> 4.65Improved
add gated attentionOPT-125m · Wikipedia valW8A8 perplexity: 21.18 -> 16.02Improved
add clipped softmaxOPT-125m · Wikipedia valW8A8 perplexity: 21.18 -> 37.20Hurt
add clipped softmaxViT-S/16 · ImageNet-1K valW8A8 accuracy: 69.24 -> 79.77Improved
add gated attentionOPT-1.3B · Wikipedia valW8A8 perplexity: 989.6 -> 29.95Improved
use gamma<0 in clipped softmax (clip at zero) (ablation)BERT-base · BERT Wikipedia valW8A8 perplexity: 1294 -> 4.55Improved
use zeta>1 only (clip at one) (ablation)BERT-base · BERT Wikipedia valW8A8 perplexity: 1294 -> 1469No clear effect
add zeta>1 on top of gamma<0 (ablation)BERT-base · BERT Wikipedia valW8A8 perplexity: 4.55 -> 4.56No clear effect
initialize gate bias for very low open probability (ablation)BERT-6L and ViT · pretrainingFP and quantized performanceHurt
add LayerNorm after patch embeddings (ablation)ViT-S/16 vanilla · ImageNet-1K valW8A8 accuracy: 69.24 -> 79.62Improved
add weight decay on LayerNorm gamma (ablation)OPT-125m vanilla · Wikipedia valW8A8 perplexity: 21.18 -> 39.46Mixed

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