Over fine-tune of quantised LLaMa 3.2 1B on this dataset (12 epochs).

As with other overtrained models, temperature above 0.5-0.7 and high values of sampling parameters will result in the model producing garbage or the dataset source material regardless of current context.