When should you consider fine-tuning a foundation model?
Why D is the answer
When prompting cannot reach the required quality or format
Consider fine-tuning when prompting cannot reach the required quality or format.
In detail
- Fine-tune when prompting cannot reach the required quality or format.
- Tuning costs more than prompting.
Use case
An illustrative scenario, not a real incident.
A team tried prompting and still cannot reach a strict format, so fine-tuning becomes worthwhile.
See it in different ways
stateDiagram-v2 [*] --> Prompting Prompting --> Rag: lacks current facts Prompting --> FineTuning: quality or format is not reached Prompting --> Done: meets the need Rag --> Done FineTuning --> Done
flowchart TD
Q{"Can prompting reach the quality?"}
Q -->|"Yes"| A0["Stop at prompting"]
Q -->|"No"| A1["Consider fine-tuning"]Key point: Fine-tune when prompting falls short.
Why the other options are wrong
- A. Before you have tried any prompts on your task
- Try prompts first.
- B. Whenever the model lacks recent facts
- Recent facts need retrieval.
- C. For every new app, as a default step
- It should not be a default step.