Fine Tuning

Fine-tuning allows you to adapt foundation models to your specific data and use cases. Orbit supports fine-tuning workflows for supported LLMs and embedding models.

Key Concepts

• Data preparation: Clean and format your training data for optimal results.
• Training jobs: Launch fine-tuning jobs using Orbit's orchestration or directly via cloud providers.
• Model versioning: Track and manage different versions of your fine-tuned models.

Example: Fine-Tuning Workflow

1. Prepare your dataset in JSONL or CSV format.
2. Use the Orbit CLI or UI to start a fine-tuning job.
3. Monitor training progress and review evaluation metrics.
4. Deploy the fine-tuned model as an Orbit Agent or API endpoint.

Best Practices

• Start with a small dataset to validate your pipeline.
• Use validation and test splits to avoid overfitting.
• Document model changes and evaluation results.

See the Orbit documentation for supported models and detailed fine-tuning guides.