Improving large language models with concept-aware fine-tuning