Enable AFlow Optimization for All Evaluators

[RuishanFang]

Description:
The AFlowOptimizer has been integrated into our framework to enable the automatic optimization of agent workflows. This feature is currently only functional with humaneval_evaluator. To extend this capability across all supported benchmarks, we must make each evaluator compatible with the optimizer's requirements.

Proposed Evaluators to Extend

• aime_evaluator.py
• bbh_evaluator.py
• drop_evaluator.py
• gaia_evaluator.py
• gsm8k_evaluator.py
• hotpotqa_evaluator.py
• ifeval_evaluator.py
• math_evaluator.py
• mbpp_evaluator.py
• mmlu_pro_evaluator.py
• swebench_evaluator.py

Implementation Considerations:

• Implement the async_evaluate Method
• Define and Load Datasets for Optimization and Testing

References:

• https://github.com/FoundationAgents/MetaGPT/tree/main/metagpt/ext/aflow
• https://github.com/FoundationAgents/AFlow