examples/aflow/humaneval/optimized/round_8/graph.py

import evoagentx.workflow.operators as operator
import examples.aflow.humaneval.optimized.round_8.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)  # Added Test operator for validation
        self.sc_ensemble = operator.ScEnsemble(self.llm)  # Added ScEnsemble operator for better solution selection

    async def __call__(self, problem: str, entry_point: str):
        """
        Implementation of the workflow
        Custom operator to generate code and validate it with tests, then select the best solution through ensemble.
        """
        solution = await self.custom_code_generate(problem=problem, entry_point=entry_point, instruction=prompt_custom.GENERATE_PYTHON_CODE_PROMPT)
        validation = await self.test(problem=problem, solution=solution['response'], entry_point=entry_point, benchmark=self.benchmark)  # Testing the solution
        
        if validation['result']:
            return solution['response']
        else:
            # Reflect on the errors if tests fail and generate another solution
            modified_solution = await self.custom(input=problem + f" with problems: {validation['solution']}", instruction=prompt_custom.GENERATE_PYTHON_CODE_PROMPT)
            # Generate multiple solutions for ensemble
            ensemble_solutions = [solution['response'], modified_solution['response']]
            final_solution = await self.sc_ensemble(solutions=ensemble_solutions, problem=problem)  # Select the best solution through ensemble
            return final_solution['response']  # Return the selected solution