This is a common problem. It says "unstable" when one of your parameters is theoretically infinite. This is the case when your model can perfectly estimate the response value, or when there are more parameters that can be estimated with the data (this is the case when you have "sparse" data, where sparse means that there are no or very few repeated covariates).
One solution is to reduce the number of variables or to change continuous variables to categorical ones. There is no information on which variables should be eliminated or which should be changed to categorical ones, because all of them are related at the same time. In general, the model created can be used to group the observations, but should not be used to estimate the parameters.
FAQ # 1232
