That ModelDB entry actually contains several models. One is very simple (defined by demofig2.hoc)--uses two compartments to represent the entire cell. This is probably not the model that you mean.
The other models are based on anatomical data from real cells. However, none of that data includes any data from axons, so the model authors made up their own myelinated axon--including initial segment and axon hillock--and attached that to the "real" anatomy (see demofig1.hoc). You can modify the properties of that axon in any way you like, but be prepared to justify what you do to whomever reviews whatever publication you generate based on your computational experiments.
I want to . . . modify the angle at branching point.
Then you'll need to use the 3-D specification of geometry, not the stylized (L,diam) approach that the model authors used. You'll find descriptions of these approaches in the Conceptual Overview of Sections
(Python https://www.neuron.yale.edu/neuron/stat ... f-sections
, hoc https://www.neuron.yale.edu/neuron/stat ... f-sections
I should mention that shape plots of the model cells are going to look strange for two reasons. First, the model setup code accounts for the contribution of spines to cell surface area by distorting the lengths and diameters of the sections. The amount of distortion varies from branch to branch, so you end up with peculiar looking models. Is there a way to avoid that? Yes. Use a model that accounts for the contribution of spines to cell surface area without stretching the model cell's branches.
Second, unless you are very careful, your model axon will come off of the soma at an angle that looks quite unnatural. Is there a way to fix that? Yes, several, but they are tedious, and which one to choose depends on your intended use of the model.
Finally, because of those two reasons, it would be inadvisable to use any of these models if you are interested in studying either extracellular stimulation or generation of local field potential by cellular activity. NeuroMorpho.org contains many neuronal morphologies that include axonal reconstructions, and at least some of them may be suitable for use in computational modeling.