Training is your data in column vector with each column as your feature, and groups represent your label(class of that row i,e one entity feature set) for each data.. Number of rows in Training and Groups should be same.
Training with two features: groups
feature1 feature 2
5 2
3 5
Label vector:
1
2..
I hope it helps.. I have two features in my feature matrix and for each row i have one label. Set your data accordingly.
Better is to use Libsvm for multiclass.. It is faster and efficient. Just take care about inversion in passing parameters. libSvm : For training use this comment. It will save your trained model and you can use this for testing any point of time.