TuCo: Measuring the Contribution of Fine-Tuning to Individual Responses of LLMs