Salvato in:
Dettagli Bibliografici
Autori principali: Li, Binghui, Chen, Fengling, Huang, Zixun, Wang, Lean, Wu, Lei
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2509.19189
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908835018440704
author Li, Binghui
Chen, Fengling
Huang, Zixun
Wang, Lean
Wu, Lei
author_facet Li, Binghui
Chen, Fengling
Huang, Zixun
Wang, Lean
Wu, Lei
contents Scaling laws have emerged as a unifying lens for understanding and guiding the training of large language models (LLMs). However, existing studies predominantly focus on the final-step loss, leaving open whether the entire loss dynamics obey similar laws and, crucially, how the learning rate schedule (LRS) shapes them. We address these gaps in a controlled theoretical setting by analyzing stochastic gradient descent (SGD) on a power-law kernel regression model. The key insight is a novel intrinsic-time viewpoint, which captures the training progress more faithfully than iteration count. We then establish a Functional Scaling Law (FSL) that captures the full loss trajectory under arbitrary LRSs, with the schedule's influence entering through a simple convolutional functional. We further instantiate the theory for three representative LRSs -- constant, exponential decay, and warmup-stable-decay (WSD) -- and derive explicit scaling relations in both data- and compute-limited regimes. These comparisons explain key empirical phenomena: (i) higher-capacity models are more data- and compute-efficient; (ii) learning-rate decay improves training efficiency; and (iii) WSD-type schedules outperform pure decay. Finally, experiments on LLMs ranging from 0.1B to 1B parameters demonstrate the practical relevance of FSL as a surrogate model for fitting and predicting loss trajectories in large-scale pre-training.
format Preprint
id arxiv_https___arxiv_org_abs_2509_19189
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Functional Scaling Laws in Kernel Regression: Loss Dynamics and Learning Rate Schedules
Li, Binghui
Chen, Fengling
Huang, Zixun
Wang, Lean
Wu, Lei
Machine Learning
Scaling laws have emerged as a unifying lens for understanding and guiding the training of large language models (LLMs). However, existing studies predominantly focus on the final-step loss, leaving open whether the entire loss dynamics obey similar laws and, crucially, how the learning rate schedule (LRS) shapes them. We address these gaps in a controlled theoretical setting by analyzing stochastic gradient descent (SGD) on a power-law kernel regression model. The key insight is a novel intrinsic-time viewpoint, which captures the training progress more faithfully than iteration count. We then establish a Functional Scaling Law (FSL) that captures the full loss trajectory under arbitrary LRSs, with the schedule's influence entering through a simple convolutional functional. We further instantiate the theory for three representative LRSs -- constant, exponential decay, and warmup-stable-decay (WSD) -- and derive explicit scaling relations in both data- and compute-limited regimes. These comparisons explain key empirical phenomena: (i) higher-capacity models are more data- and compute-efficient; (ii) learning-rate decay improves training efficiency; and (iii) WSD-type schedules outperform pure decay. Finally, experiments on LLMs ranging from 0.1B to 1B parameters demonstrate the practical relevance of FSL as a surrogate model for fitting and predicting loss trajectories in large-scale pre-training.
title Functional Scaling Laws in Kernel Regression: Loss Dynamics and Learning Rate Schedules
topic Machine Learning
url https://arxiv.org/abs/2509.19189