Is Training Data Quality or Quantity More Impactful to Small Language Model Performance?