Home › Key concepts › Prompt, retrieve or fine-tune?
Key concept · Walk (intermediate) · the same on every platform
Prompt, retrieve or fine-tune?
Three ways to adapt a model, in order of cost.
What it is
Prompting changes the instructions, which is free and instant. Retrieval (RAG) gives the model the right documents at run time, for knowledge. Fine-tuning changes the model's weights with examples, for style, format or a narrow skill.
Why it matters
Teams often jump to fine-tuning when a better prompt or better retrieval would fix it.
An example
Need answers from today's policy documents: retrieval. Need replies always in the company's tone and format: maybe fine-tuning.
The common mistake
Fine-tuning before you have an evaluation set to show it helped.
Where you meet it on each platform
The idea stays the same; only the product name changes.
| Job | AWS | Azure | Google Cloud | Databricks | Open source |
|---|---|---|---|---|---|
| Fine-tuningteach a model your style or task with examples | Bedrock custom models / SageMaker | Foundry fine-tuning | Vertex AI tuning | Mosaic AI fine-tuning | Hugging Face TRL or Unsloth |
| Vector searchfind passages by meaning, not exact words | OpenSearch Serverless / Aurora pgvector | Azure AI Search | Vertex AI Vector Search / AlloyDB | Vector Search | pgvector or Qdrant |
| Prompts and structured outputask clearly and get JSON back you can check | Bedrock Converse API | Foundry structured outputs | Gemini controlled generation | ai_query + structured output | Pydantic + JSON mode |
Related ideas
Where it shows up
Part of the Walk level on these paths: LLM Engineering (adapting models).