AI Paper Atlas

Research map / Optimization / Optimizers

Winner-Take-All Column Row Sampling for Memory Efficient Adaptation of Language Model

Zirui Liu, Guanchu Wang, Shaochen Zhong, Zhaozhuo Xu, Daochen Zha, Ruixiang Tang, Zhimeng Jiang, Kaixiong Zhou, Vipin Chaudhary, Shuai Xu, Xia Hu

NeurIPS 2023 · Optimization › Optimizers · approach: Memory-efficient LLM optimizers · 2 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 WTA-CRS@0.3 to full fine-tuningT5-Base · GLUE averageaverage score: 85.7 -> 85.4No clear effect
add WTA-CRS@0.3 to LoRAT5-3B · GLUE averageaverage score: 89.5 -> 89.8Improved
add WTA-CRS@0.3 to LoRAT5-Base · peak memory (GB)peak memory: 13.84 -> 6.50Improved
add WTA-CRS@0.3 to full fine-tuningT5-Large · peak memory (GB)peak memory: 45.85 -> 21.58Improved
lower column-row budget from k=0.3|D| to k=0.1|D| (ablation)WTA-CRS · T5-Base/Large memorypeak memory (GB): 8.44 -> 7.30Improved
lower column-row budget from k=0.3|D| to k=0.1|D| (ablation)WTA-CRS · GLUE averageaccuracy: about 1% dropHurt
use WTA-CRS@0.1 on larger model (ablation)T5-3B · GLUE averageaccuracy: 0.4% dropNo clear effect
compare LST with LoRA and WTA-CRSLST on T5-Base · GLUE averageaverage score: 85.7 -> 83.8Hurt
replace WTA-CRS with deterministic top-k selection (ablation)WTA-CRS · T5-Base SST2/MNLI/QQPvalidation accuracyHurt
replace WTA-CRS with plain CRS sampling (ablation)WTA-CRS · T5-Base SST2/MNLI/QQPvalidation accuracyHurt
combine WTA-CRS with LoRA for larger batchesT5-3B · maximum batch sizebatch size enlargement: 4.8x to 6.4xImproved
use WTA-CRS@0.1 with larger batch sizeT5-Large · training throughputthroughput: 1.08xImproved

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