Accurate Prediction of Ligand-Protein Interaction Affinities with Fine-Tuned Small Language Models