You are not really telling us if these large value are just a byproduct
of bad steps in the
function minimizer or intrinsically necessary in the model. If they just
occur from
bad steps you can use the mfexp function to bound the exp. Otherwise one
would really
need to see the code to understand this better.
Dave