This Deepseek "Minny" was a somewhat naive attempt at pushing the capabilities of a 1.5B model, without retraining.