examples/aflow/mbpp/optimized/round_11/graph.py

import evoagentx.workflow.operators as operator
import examples.aflow.mbpp.optimized.round_11.prompt as prompt_custom
from evoagentx.models.model_configs import LLMConfig
from evoagentx.benchmark.benchmark import Benchmark
from evoagentx.models.model_utils import create_llm_instance

class Workflow:

    def __init__(
        self,
        name: str,
        llm_config: LLMConfig,
        benchmark: Benchmark
    ):
        self.name = name
        self.llm = create_llm_instance(llm_config)
        self.benchmark = benchmark 
        self.custom = operator.Custom(self.llm)
        self.custom_code_generate = operator.CustomCodeGenerate(self.llm)
        self.test = operator.Test(self.llm)
        self.ensemble = operator.ScEnsemble(self.llm)
        self.alternative_fallback = operator.Custom(self.llm)

    async def __call__(self, problem: str, entry_point: str):
        """
        Implementation of the workflow
        Custom operator to generate anything you want.
        But when you want to get standard code, you should use custom_code_generate operator.
        """
        solution = await self.custom_code_generate(problem=problem, entry_point=entry_point, instruction=prompt_custom.GENERATE_PYTHON_CODE_PROMPT) 
        test_result = await self.test(problem=problem, solution=solution['response'], entry_point=entry_point, benchmark=self.benchmark)
        
        if not test_result['result']:
            unique_solutions = set()
            while len(unique_solutions) < 3:
                feedback = f"Last solution failed: {test_result['solution']}.\nPrevious errors: {test_result['error_logs']}."
                fallback_solution = await self.custom_code_generate(problem=problem, entry_point=entry_point, instruction=prompt_custom.GENERATE_PYTHON_CODE_WITH_FEEDBACK_PROMPT + feedback)
                
                if fallback_solution['response'] not in unique_solutions:
                    unique_solutions.add(fallback_solution['response'])
                    
            fallback_results = []
            for fallback in unique_solutions:
                fallback_test_result = await self.test(problem=problem, solution=fallback, entry_point=entry_point, benchmark=self.benchmark)
                fallback_results.append(fallback_test_result)
                
            valid_fallbacks = [res['solution'] for res in fallback_results if res['result']]
            
            if valid_fallbacks:
                final_fallback = await self.custom(input=problem + f" Verify this solution: {valid_fallbacks[0]}.", instruction=prompt_custom.VERIFY_SOLUTION_PROMPT)
                return final_fallback['response']
            
            alternative_solution = await self.alternative_fallback(input=problem, instruction=prompt_custom.GENERATE_PYTHON_CODE_PROMPT)
            ensemble_result = await self.ensemble(solutions=[solution['response']] + list(unique_solutions) + [alternative_solution['response']], problem=problem)
            return ensemble_result['response']  

        return test_result['solution']