Training API / Overview
Turn accepted outcomes into adapters
Collect project-scoped generation or reranking samples, create jobs through controlled tuning workflows, and monitor each job through completion.Training lifecycle
- Submit accepted samples with stable sample IDs.
- Let a Tune node create queued jobs from eligible project data.
- Dispatch queued jobs to the training service in bounded batches.
- Monitor queued, submitted, training, completed, and failed states.
- Retry only jobs in a retryable state and inspect artifact metadata before promotion.