Beyond Demographics: Fine-tuning Large Language Models to Predict Individuals' Subjective Text Perceptions