Fine-Tuning
Fine-tuned models learn from your specifications, codebase patterns, and corrections. They produce output that matches your team's style better than general-purpose models.
When to Fine-Tune
Fine-tuning helps when:
- You have consistent team conventions the general model doesn't follow
- You make frequent manual corrections to generated code
- You have a large codebase with unique patterns
- Your domain uses specialized terminology
Skip fine-tuning if you're just starting with Aexol — build a Knowledge Base first.
Creating a Model
- Go to
/studio/fine-tuning - Create a new fine-tuning job
- Select training data:
.aexolspecs, generated artifacts, manual corrections - Configure training parameters
- Start training
Training Data Quality
Good: diverse examples, consistent style, corrected output showing "right" answers
Bad: duplicates, inconsistent style, uncorrected errors
Using Fine-Tuned Models
Once trained, the model appears in:
- Studio — generate and import panels
- MCP — reference by name in
remote_start_inference - Agent — select as preferred model
Model Management
- Versioning — train new versions as your codebase evolves
- Team sharing — models shared across team members
- Compare — side-by-side output at
/studio/fine-tuning
Next Steps
- Code Generation — Using models
- Knowledge Base — Lighter-weight context alternative
- Refinement — Improving output