AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

D4: Improving LLM Pretraining via Document De-Duplication and Diversification

Kushal Tirumala, Daniel Simig, Armen Aghajanyan, Ari Morcos

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 14 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
apply D4 selection (SemDeDup then SSL Prototypes) at R=0.25 for pretrainingOPT-6.7B · validation perplexity, 100B tokenstraining efficiency: 20% fewer update steps to same perplexityImproved
apply D4 selection at R=0.25OPT-6.7B · 16 NLP tasks, 0-shot averageaccuracy: 2% improvementImproved
repeat D4-selected data for 2 epochs instead of single epoch on random new dataOPT-1.3B · Instruction + Answers perplexityperplexity: 14.19 -> 13.85Improved
repeat randomly selected data for 2 epochs instead of single epoch on all new dataOPT-1.3B · Instruction + Answers perplexityperplexity: 14.19 -> 14.37Hurt
repeat D4-selected data for 2 epochs instead of single epoch on random new dataOPT-1.3B · Non-web snapshot perplexityperplexity: 16.27 -> 16.10Improved
apply data selection (SemDeDup, SSL Prototypes or D4)OPT-1.3B · web snapshot validation sets (C4, CC-dedup, CommonCrawl)perplexityHurt
add re-clustering after SemDeDup before SSL Prototypes (ablation)D4 · Instruction + Answers and Non-web perplexityperplexityImproved
combine SemDeDup with SSL Prototypes (D4) instead of either alone (ablation)data selection · Instruct OPT validation perplexity, 1.3BperplexityImproved
vary number of K-Means clusters (ablation)SSL Prototypes · 125M OPT on C4perplexity changeNo clear effect
increase model scale from 125M to 6.7BD4 · efficiency gainefficiency gain: est. 20% for LLama-65B, 22% for OPT-175BImproved

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