Training

Monitor and trigger training runs.

Cloud Pro+ Feature

Training requires a Cloud Pro or Team plan.

Check Training Status

stats = zubbl.get_training_stats()
if stats:
    print(f"Available: {stats.is_available}")
    print(f"Buffered: {stats.total_trajectories_buffered}")
    print(f"Total runs: {stats.training_runs}")
    if stats.last_training_metrics:
        print(f"Last metrics: {stats.last_training_metrics}")

Trigger Training

result = zubbl.trigger_training()
print(f"Training triggered: {result}")

Training runs automatically when the buffer reaches the threshold (default: 50 trajectories). A weekly Kaggle notebook also runs DPO/SFT training on collected trajectories and reports results back to the API.