Curriculum pretraining with ascending data quality outperforms random order under constant learning rate but loses most benefit under standard decay; moderate decay or final-checkpoint averaging recovers a 1.64% average benchmark gain on 1.5B models trained for 30B tokens.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
How Learning Rate Decay Wastes Your Best Data in Curriculum-Based LLM Pretraining
Curriculum pretraining with ascending data quality outperforms random order under constant learning rate but loses most benefit under standard decay; moderate decay or final-checkpoint averaging recovers a 1.64% average benchmark gain on 1.5B models trained for 30B tokens.