Average relative error of the template stars between the MAP estimates and the literature values sorted by their spectral type. The average error is calculated as the rms of the relative errors of temperature and gravity, both in log scale (Eq. (9)). The pink area indicates the ltr uncertainty of the literature value. We only present template stars whose literature temperature value is within the network training range and whose literature gravity value is presented. The colour codes are the same as in Fig. 3.