Model Fine-Tuning User Guide
Prepare datasets, choose base models, create fine-tuning jobs, manage adapters, evaluate outputs, and promote tuned models.
Who This Guide Is For
- ML engineers
- Data scientists
- LLM application teams
Where To Go
| Page |
Use It For |
/model-fine-tuning |
Fine-tuning dashboard. |
/model-fine-tuning/new |
Create a fine-tuning job. |
/model-fine-tuning/base-models |
Select base models. |
/model-fine-tuning/datasets |
Fine-tuning datasets. |
/model-fine-tuning/data |
Data preparation. |
/model-fine-tuning/jobs |
Fine-tuning jobs. |
/model-fine-tuning/adapters |
Adapter artifacts. |
/model-fine-tuning/evaluation |
Evaluate tuned outputs. |
/model-fine-tuning/settings |
Fine-tuning settings. |
Core Concepts
| Concept |
Meaning |
| Base model |
The foundation model selected for tuning. |
| Dataset |
Instruction, preference, classification, or task-specific training data. |
| Adapter |
A lightweight fine-tuned artifact such as LoRA where supported. |
| Job |
The training execution for a fine-tuning configuration. |
| Evaluation |
The validation stage before registration or deployment. |
Common Workflows
Fine-tune a model
- Select base model.
- Choose or upload dataset.
- Validate schema and examples.
- Configure training method and resources.
- Launch job.
- Review metrics and artifacts.
- Evaluate tuned output.
- Register the approved artifact in ModelOps.
Manage adapters
- Open Adapters.
- Review model, dataset, and training metadata.
- Compare evaluation results.
- Archive stale adapters.
- Promote selected adapters for deployment.
Best Practices
- Validate dataset quality before spending GPU time.
- Keep base model, dataset version, training config, and evaluation result attached to every tuned artifact.
- Use held-out evaluation sets for promotion decisions.
- Register only approved tuned artifacts in ModelOps.