It's hard to tell from the screenshot alone.  Just to clarify, the idea for the panel is to fit pre-edge peaks to known lineshapes (Gaussian, Lorentzian, Voigt, etc).  The interface could probably be used for general-purpose curve-fitting of many kinds of spectra, but right now it definitely skews to pre-edge peak fitting. 

Because of that emphasis, the idea is to first press the "Fit Baseline" button -- this will do a crude fit of a Lorentzian + Line that are meant to represent any residual background and the main absorption edge.  The fit for this baseline is done over the energy range defined by "Fit Energy Range" but ignoring the "Pre-edge Peak Range", which is meant to be the portion of the XAS spectra where the pre-edge peaks are. That is, this baseline fit should

look something like Figure 5.4.2 at https://xraypy.github.io/xraylarch/xasviewer/index.html#pre-edge-peak-fitting.  In that figure, the black dots show the ends of the Pre-edge Peak Range that is ignored in this baseline fit.   For sure, the initial guess for this range may not be very good.  But it is easy to refine this ignored range.  It is also not actually critical to get this baseline fit perfect - it's just a place to start the fits of the peaks themselves, and the remaining fits will refine the baseline as well as the peaks over the full energy range.     After a baseline fit is done, you can add peaks and then use the "Fit Model", which fits over the whole "Fit Energy Range",  ignoring the "Pre-edge Peak Range".

From your screenshot, I don't see much of a pre-edge and not many obvious peaks.  The result is that the actual fit energy range for the baseline (that is 

"Fit Energy Range" - "Pre-edge Peak Range") is very short.  I can believe the fit will fail under those conditions.  But I sort of think that failure really comes because your data doesn't have pre-edge peaks.  

Maybe you are trying to do something more complicated?