In the Current Population Survey (CPS), replication methods are used to calculate variances of survey estimates. Since these are often noisy, generalized variance functions (GVFs) are used to produce published estimates of variance that are more stable over time. Recently, the calculation of GVF model parameters has been reconfigured in the CPS. Rather than cluster series and create interdependencies among variance estimates, the parameters for each series are calculated individually, based only on their own histories. Instead of an iterative refitting, a single model is constructed for each historical series, smoothing out the noisiness of replicate variances while retaining seasonality. This paper details these changes to the GVF model framework and presents the resultant improvements in CPS published variance estimates.