ax-python-agent-optimize

To optimize Python Ax agents, you apply evaluator-driven workflows like AxGEPA and BootstrapFewShot. This process tunes agents and reusable components by generating optimization workflows while managing and persisting optimizer artifacts for later reuse.

GEPA is an evaluator-driven optimization technique used in axllm to tune Python Ax agents. It works by applying AxGEPA workflows that leverage evaluator callbacks to compare outputs and iteratively improve agent performance based on bounded dataset experiments.

You build evaluator and judge callbacks for Ax agents to compare outputs during optimization runs. These evaluators are integrated into the optimization workflow, allowing the system to assess agent performance and persist optimizer artifacts for reproducible testing.

Yes, you need explicit budgets and dataset rows to run axllm optimization experiments. Bounded experiments require these constraints to keep runs controlled, ensuring reproducible testing when tuning Python Ax agents with workflows like AxGEPA and BootstrapFewShot.

Yes, you can persist optimizer artifacts for later reuse in axllm. The optimization process explicitly manages these artifacts, ensuring that evaluator-driven optimization workflows remain reproducible and aligned with the generated axllm package API shapes.

Bounded dataset experiments in axllm require explicit budgets and package examples to function correctly. Optimization runs are constrained by these dataset rows, meaning experiments must be carefully scoped to ensure reproducible testing within defined limits.