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

• URL: http://arxiv.org/abs/2511.18903v1
• Date: Mon, 24 Nov 2025 09:03:49 GMT
• Title: How Learning Rate Decay Wastes Your Best Data in Curriculum-Based LLM Pretraining
• Authors: Kairong Luo, Zhenbo Sun, Haodong Wen, Xinyu Shi, Jiarui Cui, Chenyi Dang, Kaifeng Lyu, Wenguang Chen,
• Abstract summary: 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.<n>This work identifies a critical factor constraining these methods: the incompatibility between the ascending data quality order and the decaying learning rate schedule.
• Score: 22.50461083222824
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: 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.

Related papers

• DAJ: Data-Reweighted LLM Judge for Test-Time Scaling in Code Generation [30.131052926559956]
  We propose DAJ, a reasoning-based LLM judge trained with rewards under a bi-level data-reweighted learning framework.<n>Our approach automatically emphasizes hard problems, in-distribution samples, and trajectory-aligned data, without relying on hand-crafted verifiables.
  arXiv  Detail & Related papers  (2026-01-29T19:04:24Z)
• Nonparametric LLM Evaluation from Preference Data [86.96268870461472]
  We propose a nonparametric statistical framework, DMLEval, for comparing and ranking large language models (LLMs) from preference data.<n>Our framework provides practitioners with powerful, state-of-the-art methods for comparing or ranking LLMs.
  arXiv  Detail & Related papers  (2026-01-29T15:00:07Z)
• Improving Pretraining Data Using Perplexity Correlations [56.41097718862742]
  We present a framework that selects high-quality pretraining data without any LLM training of our own.<n>We build a new statistical framework for data selection centered around estimates of perplexity-benchmark correlations.<n>Our approach outperforms DSIR on every benchmark, while matching the best data selector found in DataComp-LM.
  arXiv  Detail & Related papers  (2024-09-09T17:23:29Z)
• Entropy Law: The Story Behind Data Compression and LLM Performance [115.70395740286422]
  We find that model performance is negatively correlated to the compression ratio of training data, which usually yields a lower training loss. Based on the findings of the entropy law, we propose a quite efficient and universal data selection method. We also present an interesting application of entropy law that can detect potential performance risks at the beginning of model training.
  arXiv  Detail & Related papers  (2024-07-09T08:14:29Z)
• Uncertainty Aware Learning for Language Model Alignment [97.36361196793929]
  We propose uncertainty-aware learning (UAL) to improve the model alignment of different task scenarios. We implement UAL in a simple fashion -- adaptively setting the label smoothing value of training according to the uncertainty of individual samples. Experiments on widely used benchmarks demonstrate that our UAL significantly and consistently outperforms standard supervised fine-tuning.
  arXiv  Detail & Related papers  (2024-06-07T11:37:45Z)
• Simple and Scalable Strategies to Continually Pre-train Large Language Models [20.643648785602462]
  Large language models (LLMs) are routinely pre-trained on billions of tokens, only to start the process over again once new data becomes available. We show that a simple and scalable combination of learning rate re-warming, LR re-decaying, and replay of previous data is sufficient to match the performance of fully re-training from scratch.
  arXiv  Detail & Related papers  (2024-03-13T17:58:57Z)
• Take the Bull by the Horns: Hard Sample-Reweighted Continual Training Improves LLM Generalization [165.98557106089777]
  A key challenge is to enhance the capabilities of large language models (LLMs) amid a looming shortage of high-quality training data. Our study starts from an empirical strategy for the light continual training of LLMs using their original pre-training data sets. We then formalize this strategy into a principled framework of Instance-Reweighted Distributionally Robust Optimization.
  arXiv  Detail & Related papers  (2024-02-22T04:10:57Z)
• How to Train Data-Efficient LLMs [56.41105687693619]
  We study data-efficient approaches for pre-training language models (LLMs) We find that Ask-LLM and Density sampling are the best methods in their respective categories. In our comparison of 19 samplers, involving hundreds of evaluation tasks and pre-training runs, we find that Ask-LLM and Density are the best methods in their respective categories.
  arXiv  Detail & Related papers  (2024-02-15T02:27:57Z)
• When Less is More: Investigating Data Pruning for Pretraining LLMs at Scale [12.94829977468838]
  Large volumes of text data have contributed significantly to the development of large language models. To date, efforts to prune datasets down to a higher quality subset have relied on hand-crafteds encoded as rule-based filters. We take a wider view and explore scalable estimates of data quality that can be used to measure the quality of pretraining data.
  arXiv  Detail & Related papers  (2023-09-08T19:34:05Z)
• Ranking & Reweighting Improves Group Distributional Robustness [14.021069321266516]
  We propose a ranking-based training method called Discounted Rank Upweighting (DRU) to learn models that exhibit strong OOD performance on the test data. Results on several synthetic and real-world datasets highlight the superior ability of our group-ranking-based (akin to soft-minimax) approach in selecting and learning models that are robust to group distributional shifts.
  arXiv  Detail & Related papers  (2023-05-09T20:37:16Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.