How is one-vs-rest implemented in scikit-learn?

Question

I am trying to understand how scikit-learn handles the multi-class classification problem for support vector machines. I am focused at the moment on the \'one-vs-rest\' case.