Savepoint/loadpoint is an easy way to use an initial point for an NLP or advanced basis for an LP. The commands are:
The first time you run this the solution is saved in the GDX file. The next time around, this GDX is used to set as many levels and marginals that can be matched by name. For this very large MCP model it turned out to be useful to bring down solution times of related experiments.
Sorry for the ugly syntax. I would have preferred something more symmetric for reading and writing.