agents-optimize

You can optimize NeMo agents by applying automated prompt tuning, switchyard routing splits, and model swaps based on evaluation benchmarks to improve accuracy and reduce latency. This process validates all proposed changes against a baseline to prevent quality regressions.

Yes, you can test smaller, faster models or implement routing splits within NeMo to maintain response quality while reducing compute overhead. The optimization process evaluates performance metrics against a baseline to ensure no quality regression occurs before deployment.

Evaluation-driven promotion validates proposed agent configuration changes against an evaluation baseline before deployment. This mechanism orchestrates prompt tuning and routing splits to ensure automated optimization prevents quality regressions in production agents.

Yes, executing agent lifecycle management and validation workflows requires integration with the NeMo CLI and files service. This integration allows the optimizer to manage agent configurations, evaluate performance metrics, and persist optimization snapshots.

The best way to automate hyperparameter tuning for deployed agents is using an orchestration skill that suggests routing splits and model swaps based on evaluation benchmarks. This approach manages agent configurations and persists optimization snapshots automatically.

Routing splits allow you to balance accuracy and latency by directing traffic across multiple models, whereas a direct model swap is an all-or-nothing change. The optimizer evaluates both strategies against benchmarks to ensure compute overhead is reduced without causing quality regressions.