OLMo2-0425-1B fine-tuned on different subsets of the Alpaca dataset. Each subset was partitioned based on the perplexity wrt pretrained model.