Fine-tuning a language model for question-answering
My objective is to fine-tune a model on data about Manchester United's (MU's) 2021/22 season. I want to be able to prompt the fine-tuned model with questions such as "How can MU improve?", or "What are MU's biggest weaknesses?"
What data should I train the model on? Should it just be a long string of text which details MU's season, a list of question-answer pairs - e.g. "prompt": "What could Manchester United do to better manage unexpected changes such as player loans or COVID-19 disruptions?", "completion": "To better manage unexpected changes, Manchester United could invest in contingency planning, including having backup players ready for loan situations, and ensuring strict health protocols to minimize COVID-19 disruptions. Rigorous planning and flexibility can help the team adapt to changing circumstances. - a combination of both, or should I structure my data in another way?
