English

Tokens-per-Parameter Coverage Is Critical for Robust LLM Scaling Law Extrapolation

Machine Learning 2026-05-14 v2

Abstract

Neural scaling laws approximate a language model's loss as a power-law function of parameter count NN and token count DD. Following Chinchilla-style compute-optimal training, many studies fit scaling laws from runs performed under a fixed tokens-per-parameter (TPP) ratio kk and set D=kND = kN. We show that this collinear design, combined with the empirically common near-equality of the exponents governing NN and DD, induces an inherent ill-conditioning in the Gauss-Newton least-squares problem: the condition number of the design grows as the inverse square of the gap between the NN and DD-exponents. The scale coefficients become practically unidentifiable, with confidence intervals inflating by an order of magnitude or more, yielding a ``sloppy'' model whose extrapolations degrade sharply off the training ray. We prove this for four scaling-law formalisms and derive a closed-form TPP-diversity threshold that is necessary and sufficient for well-conditioned estimation. Empirically, non-collinear designs outperform collinear ones on held-out splits with a 97.3\% win rate across four laws, five corpora, multiple floating point precision modes. We further show the degeneracy is rooted in Jacobian geometry and is not an artifact of the loss function: any smooth estimation objective whose curvature involves the Jacobian inherits the same ill-conditioning.

Keywords

Cite

@article{arxiv.2605.08541,
  title  = {Tokens-per-Parameter Coverage Is Critical for Robust LLM Scaling Law Extrapolation},
  author = {Joshua Shay Kricheli and Alexander Lawrence Reid and Soumajyoti Sarkar and Venkata Gandikota and Paulo Shakarian},
  journal= {arXiv preprint arXiv:2605.08541},
  year   = {2026}
}
R2 v1 2026-07-01T12:59:15.105Z