AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

LiteTransformerSearch: Training-free Neural Architecture Search for Efficient Language Models

Mojan Javaheripi, Gustavo Henrique de Rosa, Subhabrata Mukherjee, Shital Shah, Tomasz Lukasz Religa, Caio Cesar Teodoro Mendes, Sebastien Bubeck, Farinaz Koushanfar, Debadeepta Dey

NeurIPS 2022 · Architectures and efficiency › Neural architecture search · approach: Bi-level differentiable NAS · 5 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
use decoder parameter count instead of partial training (500 steps) as ranking proxyLTS NAS · 1200 Transformer-XL models, WT103SRC: 0.92 -> 0.98Improved
use decoder parameter count instead of partial training (5,000 steps)LTS NAS · 1200 Transformer-XL models, WT103SRC: 0.96 -> 0.98Improved
use decoder parameter count instead of total parameter count as proxyparameter-count proxy · Random architectures, WT103SRCImproved
use decoder parameter count instead of total parameter count on homogeneous modelsparameter-count proxy · 100 homogeneous Transformer-XL modelsSRC: 0.95Improved
change adaptive input embedding factor k in {1,2,4} (ablation)Transformer · WikiText-103validation perplexityNo clear effect
vary width-to-depth aspect ratio in [1,630] at fixed decoder params (ablation)GPT-2 backbone · WikiText-103validation perplexity: <12.8%No clear effect
vary aspect ratio in [8,323] at fixed decoder params (ablation)Transformer-XL backbone · WikiText-103validation perplexity: <7%No clear effect
vary topology at fixed decoder params (ablation)GPT-2 backbone · WikiText-103latency / peak memory: 2.8x latency, 5.5x memoryHurt
per-architecture learning rate sweep vs fixed 0.01 (ablation)GPT-2, 100 models · WikiText-103SRC with decoder params: 0.96 -> 0.93No clear effect
layer-scaled baseline replaced by LTS heterogeneous searchTransformer-XL · LM1B on TITAN Xplatency at same perplexity: 30.5Improved
LTS search vs layer scaling for GPT-2GPT-2 · LM1B on ARM CPUlatencyNo clear effect
relu_log_det proxy on homogeneous modelslow-cost proxy · 100 homogeneous modelsSRC: 0.42Hurt

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