GLM: Continuous variable with few states as factor or numeric?

Question

I have a basic question. I am running binomial GLMs, with numeric predictors. Some of these predictors have very few unique values - some have 2, some 3 and some have 4. All