AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Should We Still Pretrain Encoders with Masked Language Modeling?

Hippolyte Gisserot-Boukhlef, Nicolas Boizard, Manuel Faysse, Duarte Miguel Alves, Emmanuel Malherbe, Andre Martins, CELINE HUDELOT, Pierre Colombo

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems · 16 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
pretrain with MLM instead of CLM610M encoder · QA averageF1: 42.09 -> 62.77Improved
pretrain with MLM instead of CLM610M encoder · SC averageaccuracy: 83.58 -> 87.00Improved
pretrain with MLM instead of CLM610M encoder · TC averageF1: 92.69 -> 92.21Hurt
pretrain with MLM instead of CLM610M encoder · IR averageNDCG@10: 76.20 -> 79.55Improved
pretrain with MLM instead of CLM1B encoder · IR averageNDCG@10: 78.68 -> 79.95Improved
use CLM instead of MLM early in training610M encoder · IR average at step 1KNDCG@10: 20.19 -> 29.94Improved
use CLM instead of MLM early in training610M encoder · TC average at step 1KF1: 83.15 -> 87.01Improved
train MLM to the end instead of CLM610M encoder · SC average at step 40Kaccuracy: 83.48 -> 86.40Improved
start with CLM then MLM (3K+9K) vs pure MLM (ablation)610M encoder · SC average, 12K stepsaccuracy: 83.04 -> 83.67Improved
start with CLM then MLM (5K+17K) vs pure MLM (ablation)610M encoder · SC average, 22K stepsaccuracy: 86.07 -> 86.31Improved
allocate 75% of steps to CLM (9K+3K) (ablation)610M encoder · SC average, 12K stepsaccuracy: 83.04 -> 82.32Hurt
pure CLM instead of pure MLM (ablation)610M encoder · SC average, 12K stepsaccuracy: 83.04 -> 80.86Hurt

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