Use Cases
Where fine-tuning performs best and how to structure your data.
Fine-tuning on documents (style-adherent generation)
Fine-tuning on documents is ideal when you want a consistent voice, structure, and formatting. The model learns your tone and patterns, not just facts. This is a strong fit for:
- Character roleplay
- LinkedIn post generator
- Blog writer
- Email assistant
Research shows fine-tuning can meaningfully improve performance in these style-adherent settings.
What to upload
- Examples that reflect the exact tone and structure you want
- Enough variety to cover edge cases (short, long, formal, casual)
Fine-tuning on CSVs (labeling and classification)
Use CSVs when each row is an input paired with a single label. This is ideal for classification and tagging tasks where the output is short and consistent. Strong fits include:
- Customer support request type labeling (billing, bug, feature request, account access)
- Fraud detection (legit, suspicious, fraudulent)
- Transaction categorization (travel, software, payroll, refunds)
- Content categorization for routing (sales, legal, HR, security)
What to upload
- One row per example
- Clear input columns and a target/label column
- Balanced classes when possible (avoid all examples being the same label)