aiqbench analysis
This research introduces Power-Law Entropy Search (PLES), a method that drastically reduces the computational cost of estimating optimal hyperparameter scaling laws for large language models. By leveraging multi-fidelity Bayesian optimization, PLES identifies informative small-scale experiments to converge on accurate configurations using less than 10% of the resources required by traditional grid searches. This advancement addresses a critical bottleneck in training efficiency, allowing practitioners to predict production-scale settings without prohibitive experimentation costs.
Efficiently Estimating Optimal Hyperparameter Scaling Laws through Power-Law Entropy Search
arXiv:2609.01431v1 Announce Type: new Abstract: Optimal hyperparameter scaling laws describe how the best hyperparameters for large language model (LLM) training change with model and data scale, enabling practitioners to predict optimal configurations at production scales without expensive large-scale tuning. However, estimating these scaling laws conventionally requires exhaustive grid searches over thousands of training runs, consuming enormous computational resources. We introduce Power-Law Entropy Search (PLES), a computational cost-aware acquisition function built on multi-fidelity Bayesian optimization that efficiently estimates optimal hyperparameter scaling laws through adaptive experimentation. A key innovation in PLES is that it searches for candidates that reduce the overall uncertainty of a scaling law estimate, instead of optimizing a single objective function. At each iteration, PLES selects the candidate configuration that maximally reduces the uncertainty of the scaling law estimates per unit computational cost, naturally favoring informative small-scale experiments. We evaluate PLES on synthetic benchmarks, surrogate models fitted to real LLM training data, and actual LLM pre-training runs. Across all settings, PLES converges to accurate optimal hyperparameter scaling laws using less than one-tenth of the computational budget required by conventional grid search and other baselines.
Abstract:Optimal hyperparameter scaling laws describe how the best hyperparameters for large language model (LLM) training change with model and data scale, enabling practitioners to predict optimal configurations at production scales without expensive large-scale tuning. However, estimating these scaling laws conventionally requires exhaustive grid searches over thousands of training runs, consuming enormous computational resources. We introduce Power-Law Entropy Search (PLES), a computational cost-aware acquisition function built on multi-fidelity Bayesian optimization that efficiently estimates optimal hyperparameter scaling laws through adaptive experimentation. A key innovation in PLES is that it searches for candidates that reduce the overall uncertainty of a scaling law estimate, instead of optimizing a single objective function. At each iteration, PLES selects the candidate configuration that maximally reduces the uncertainty of the scaling law estimates per unit computational cost, naturally favoring informative small-scale experiments. We evaluate PLES on synthetic benchmarks, surrogate models fitted to real LLM training data, and actual LLM pre-training runs. Across all settings, PLES converges to accurate optimal hyperparameter scaling laws using less than one-tenth of the computational budget required by conventional grid search and other baselines.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2609.01431 [cs.LG]
(or arXiv:2609.01431v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.01431 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 277 words
Related coverage
Expenditure Horizon: Measuring Optimization Ability, with an Application to NanoGPT
METR · Jul 21, 2026
PIRAMID: Progress and Plans
LessWrong · Jul 27, 2026
Train-to-Test (T²) Scaling Laws: Why Reasoning Models Should Be Overtrained
Snorkel AI Blog · Aug 18, 2026
Debating RSI, the US-China Gap, and Jaggedness with JS Denain of Epoch AI
Interconnects · Sep 22, 2026