I am writing a c++ extension for pytorch where I need to access individual elements of a tensor, and maybe convert it to a standard c++ type such as float. How can I do this?

I think this was raised as Issue #5 but I cannot find any example or documentation that mentions this.