Fast implementations of partial least squares models for high-dimensional regression and classification. The fastPLS software provides compiled implementations of PLS-SVD, a SIMPLS-family estimator, OPLS and kernel PLS, together with truncated singular value decomposition backends, discriminant classifiers, cross-validation utilities and optional CUDA or Apple Metal acceleration when the required system libraries are available. Compact latent prediction and memory-aware numerical routes support analyses with large predictor or multivariate-response matrices.