How much data and how long are you talking about? If it fits in memory, then the slowness is likely due to other coding errors causing a bottleneck, not the SVM training. (Unless you wrote that as well.)