Computational Elasticity: Quantifying the Relationship Between Inference Accuracy of Large Language Models and Computational Resources During Inference

Introduction / Main Floor: Computational Elasticity: Quantifying the Relationship Between Inference Accuracy of Large Language Models and Computational Resources During Inference

This article introduces a pre-registered pilot study that systematically analyzes how the inference accuracy of large language models on the GPQA Diamond benchmark scales with the increase of computational resources during inference using a parametric curve fitting method.