Optimizing ising for spin I>1/2

[zeta]

I am trying in vain to use simulated anneal to minimize an ising system with spin 7/2. This is a model for a pattern recognition task. The system is roughly 40^2 and the spin -7/2<I<7/2
ie., 2I+1 = 8 states per node. It's big.

The problem is that I can't work out what a thermal fluctuation corresponds to in this case or in other words how does one move between neighboring states? Take the TS problem, you simply rearrange coordinates. What's the analogous move here?
cheers

 

[zeta]

strictly speaking this is called the Heisenberg model, b/c I>1/2