Train the smallest LM you can that fits in 16MB. Best model wins! - View it on GitHub