How Learning Rate Decay Wastes Your Best Data in Curriculum-Based LLM Pretraining

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Luo, Kairong, Sun, Zhenbo, Wen, Haodong, Shi, Xinyu, Cui, Jiarui, Dang, Chenyi, Lyu, Kaifeng, Chen, Wenguang
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910220104499200
author Luo, Kairong
Sun, Zhenbo
Wen, Haodong
Shi, Xinyu
Cui, Jiarui
Dang, Chenyi
Lyu, Kaifeng
Chen, Wenguang
author_facet Luo, Kairong
Sun, Zhenbo
Wen, Haodong
Shi, Xinyu
Cui, Jiarui
Dang, Chenyi
Lyu, Kaifeng
Chen, Wenguang
contents Due to the scarcity of high-quality data, large language models (LLMs) are often trained on mixtures of data with varying quality levels, even after sophisticated data curation. A natural approach to better leverage high-quality data is curriculum-based pretraining, where the model is trained on data sorted in ascending order of quality as determined by a quality metric. However, prior studies have reported limited improvements from such curriculum-based pretraining strategies. This work identifies a critical factor constraining these methods: the incompatibility between the ascending data quality order and the decaying learning rate (LR) schedule. We find that while curriculum-based training substantially outperforms random shuffling when using a constant LR, its advantage diminishes under standard LR decay schedules. Our experiments show this incompatibility can be mitigated by two simple strategies: (1) employing a more moderate LR decay schedule, where the final LR is only moderately smaller than the peak LR, and (2) replacing LR decay with model averaging, i.e., computing a weighted average of the final few checkpoints. By combining these strategies, we improve the average score on a suite of standard benchmarks by 1.64% over random shuffling, without additional data refinement. Validated on 1.5B-parameter models trained over 30B tokens with various data-quality metrics, our findings call for a re-evaluation of curriculum-based LLM pretraining and underscore the potential of co-designing data curricula with optimization methods.
format Preprint
id arxiv_https___arxiv_org_abs_2511_18903
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle How Learning Rate Decay Wastes Your Best Data in Curriculum-Based LLM Pretraining
Luo, Kairong
Sun, Zhenbo
Wen, Haodong
Shi, Xinyu
Cui, Jiarui
Dang, Chenyi
Lyu, Kaifeng
Chen, Wenguang
Machine Learning
Artificial Intelligence
Computation and Language
Due to the scarcity of high-quality data, large language models (LLMs) are often trained on mixtures of data with varying quality levels, even after sophisticated data curation. A natural approach to better leverage high-quality data is curriculum-based pretraining, where the model is trained on data sorted in ascending order of quality as determined by a quality metric. However, prior studies have reported limited improvements from such curriculum-based pretraining strategies. This work identifies a critical factor constraining these methods: the incompatibility between the ascending data quality order and the decaying learning rate (LR) schedule. We find that while curriculum-based training substantially outperforms random shuffling when using a constant LR, its advantage diminishes under standard LR decay schedules. Our experiments show this incompatibility can be mitigated by two simple strategies: (1) employing a more moderate LR decay schedule, where the final LR is only moderately smaller than the peak LR, and (2) replacing LR decay with model averaging, i.e., computing a weighted average of the final few checkpoints. By combining these strategies, we improve the average score on a suite of standard benchmarks by 1.64% over random shuffling, without additional data refinement. Validated on 1.5B-parameter models trained over 30B tokens with various data-quality metrics, our findings call for a re-evaluation of curriculum-based LLM pretraining and underscore the potential of co-designing data curricula with optimization methods.
title How Learning Rate Decay Wastes Your Best Data in Curriculum-Based LLM Pretraining
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2511.18903