What do you prepare to fine-tune a model in Foundry?
Why D is the answer
Training data with example conversations
Fine-tuning is a supervised job, so it starts with training data of example conversations.
In detail
- Fine-tuning needs training data with example conversations.
- Keep some examples for validation.
Use case
An illustrative scenario, not a real incident.
A team prepares training data with example conversations showing the behavior it wants.
See it in different ways
flowchart LR A["Better prompt"] --> B["Add RAG for knowledge"] B --> C["Fine-tune for style or format"]
stateDiagram-v2 [*] --> CheckRetrieval CheckRetrieval --> TuneChunks: wrong chunks CheckRetrieval --> TunePrompt: right chunks, poor answers TunePrompt --> FineTune: style still not met TuneChunks --> [*] FineTune --> [*]
flowchart LR S0["Collect examples"] --> S1["Format as conversations"] --> S2["Fine-tune"] --> S3["Evaluate"]
Key point: Fine-tuning needs example conversations.
Why the other options are wrong
- A. A Bicep template for the model weights
- Bicep does not hold weights.
- B. A list of keywords for the content filter
- Filter keywords are not training data.
- C. A second copy of the system prompt only
- A prompt alone does not train anything.