Description
Prompt tuning is trial and error: fix one case, break another, never knowing why. GEPA uses AI-powered reflective optimization to improve prompts and code, having models analyze failures and propose better versions.
Combining evolutionary search with a Pareto frontier, it beats RL tuning with far fewer evaluations and ships in DSPy.
Reflection:Learns from failures.
Evolution:Pareto search.
Efficient:Few evaluations.
DSPy:Integrated.
Combining evolutionary search with a Pareto frontier, it beats RL tuning with far fewer evaluations and ships in DSPy.
Features
Reflection:Learns from failures.
Evolution:Pareto search.
Efficient:Few evaluations.
DSPy:Integrated.
