We describe asaga, an asynchronous parallel version of the incremental gradient algorithm Saga that enjoys fast linear convergence rates. We highlight a subtle but important technical issue present in a large fra