Fine-Tuning: What is happening under the hood?

Anyone familiar with what is happening under the hood with fine-tuning of OpenAI’s GPT models?

For example, are some weights frozen?