GpFlash benchmark data: timing and memory measurements for Gaussian process training and prediction

Timing and memory measurements behind the figures and tables of the paper "GpFlash: Maximizing Speed and Minimizing Memory for Gaussian Process Training and Prediction" by André Marchildon (arXiv:2610.16861). GpFlash is an open-source Gaussian process (GP) library (https://github.com/marchildon/gpflash). The data compare GpFlash v0.1.0 with the GP libraries scikit-learn, GPy, GPflow, GPyTorch and SMT for the marginal log likelihood and its gradient (wall time and peak memory) and for posterior evaluations (mean, variance and gradients). They also include four studies of the individual algorithmic steps used in GpFlash. All measurements were made on a MacBook Pro (Apple M3 Pro) and on one node of the Trillium supercomputer (two AMD EPYC 9655 processors), with 10 input dimensions and the Matérn 5/2 kernel. The archive holds 86 CSV files, one per case or library and machine, with every individual run (the paper reports medians). It also has an environment file per machine (hardware, BLAS library, thread settings) and a README describing each file, its columns, and how each library was called and configured.