AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Searching for Efficient Transformers for Language Modeling

David So, Wojciech Mańke, Hanxiao Liu, Zihang Dai, Noam Shazeer, Quoc V Le

NeurIPS 2021 · Architectures and efficiency › Attention and transformers · approach: Efficient language transformers · 31 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 squared ReLU activations and depthwise convolution after Q, K, and V projectionsTransformer (Primer-EZ) · LM1B, Tensor2Tensor TPUv2Speedup: 2.34 +/- 0.04Improved
Add squared ReLU activations and depthwise convolution after Q, K, and V projectionsTransformer (Primer-EZ) · C4, 110M parametersPPLX: 20.61 -> 19.83Improved
Add Primer-EZ modifications to Switch TransformerSwitch Transformer · C4, 550M parametersPPLX: 17.16 -> 16.56Improved
Add squared ReLUSynthesizer · C4, 145M parametersPPLX: 20.35 -> 19.55Improved
Replace ReLU with squared ReLU (ablation)Transformer feed-forward block · Autoregressive language modelingperplexityImproved
Add 3x1 depthwise convolution after each Q, K, and V projection (ablation)Transformer self-attention · Autoregressive language modelingperplexityImproved
Replace the Q/K projections with a shared Q and K depthwise representation (ablation)Primer · Autoregressive language modelingperformanceHurt
Apply the second normalization after the feed-forward transformation (ablation)Primer · Autoregressive language modelingperformanceMixed
Use custom normalization with x(x - μ) (ablation)Primer · Autoregressive language modelingperformanceMixed
Increase the feed-forward upward projection to 12X while controlling for parameters (ablation)Transformer · Small models (~35M parameters)perplexityImproved
Increase the feed-forward upward projection to 12X while controlling for parameters (ablation)Transformer · Larger modelsperformanceNo clear effect
Add per-channel learnable scalars after attention softmax (ablation)Primer · Fixed-length language modelingperplexityImproved

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