Learning Less Is More: Premature Upper-Layer Attention Specialization Hurts Language Model Pretraining

Add code
May 11, 2026

Share this with someone who'll enjoy it:

View paper onarxiv icon

Share this with someone who'll enjoy it: