AdaEva: Accelerating LLM-Driven Algorithm Design with Adaptive Partial Evaluation

Large Language Models (LLMs) are increasingly used for automated algorithm design. However the computational cost of evaluating the generated algorithms can be excessive.