question about get_fantasy_model functions for exact GP in GPytorch

[cagoc]

I am curious about the time complexity about get_fantasy_model for exact GPs. In GPytorch document 1.13 - Advanced Usage - SVGP Model Updating, it costs O((q^2)N) time with q new data. However, I find that in the paper Efficient Nonmyopic Bayesian Optimization via One-Shot Multi-Step Trees and OVC paper, the corresponding complexity is O(Nrq), where r is the rank of R, where R is the cholesky decomposition of the training data covariance. Could you provide the corresponding paper about how get_fantasy_model for exact GPs in GPytorch achieves O(Nq^2) complexity? Thank you for your help!