Part 4/8:
Fine-Tuning Data Format and Usage
Consistency in data formatting is critical. The dataset used for training follows a specific prompt-completion structure. If your dataset uses a particular format, you should stick to it for all training.
Shapiro highlights that for certain applications, such as question generation, having a fixed output format regardless of varying input structures can be advantageous. Fine-tuning is particularly powerful when you want consistent results, as shown in examples where the AI generates storylines or summaries.
Practical Examples: Generating Stories
Shapiro demonstrates generating stories based on different prompts. For example:
- Scenario: A plot set on Mars in 2180, involving colonization and mysterious underground devices.