Merge pull request #1587 from didiforgithub/main

Add Features to AFLOW.

Author: geekan

metagpt/actions/action_node.py

@@ -541,22 +541,22 @@ async def code_fill(
         result = {field_name: extracted_code}
         return result
 
-    async def single_fill(self, context: str) -> Dict[str, str]:
+    async def single_fill(self, context: str, images: Optional[Union[str, list[str]]] = None) -> Dict[str, str]:
         field_name = self.get_field_name()
         prompt = context
-        content = await self.llm.aask(prompt)
+        content = await self.llm.aask(prompt, images=images)
         result = {field_name: content}
         return result
 
-    async def xml_fill(self, context: str) -> Dict[str, Any]:
+    async def xml_fill(self, context: str, images: Optional[Union[str, list[str]]] = None) -> Dict[str, Any]:
         """
         Fill context with XML tags and convert according to field types, including string, integer, boolean, list and dict types
         """
         field_names = self.get_field_names()
         field_types = self.get_field_types()
 
         extracted_data: Dict[str, Any] = {}
-        content = await self.llm.aask(context)
+        content = await self.llm.aask(context, images=images)
 
         for field_name in field_names:
             pattern = rf"<{field_name}>(.*?)</{field_name}>"
@@ -635,12 +635,12 @@ async def fill(
 
         elif mode == FillMode.XML_FILL.value:
             context = self.xml_compile(context=self.context)
-            result = await self.xml_fill(context)
+            result = await self.xml_fill(context, images=images)
             self.instruct_content = self.create_class()(**result)
             return self
 
         elif mode == FillMode.SINGLE_FILL.value:
-            result = await self.single_fill(context)
+            result = await self.single_fill(context, images=images)
             self.instruct_content = self.create_class()(**result)
             return self
 

metagpt/ext/aflow/scripts/interface.py

@@ -0,0 +1,98 @@
+# -*- coding: utf-8 -*-
+# @Date    : 2024-03-21
+# @Author  : Your Name
+# @Desc    : Interface for AFLOW
+
+import asyncio
+import importlib.util
+import sys
+from pathlib import Path
+from typing import Optional, Tuple
+
+from metagpt.configs.models_config import ModelsConfig
+from metagpt.ext.aflow.scripts.evaluator import DatasetType
+from metagpt.ext.aflow.scripts.optimizer_utils.data_utils import DataUtils
+from metagpt.logs import logger
+
+
+def load_best_round(dataset: str, optimized_path: str = "metagpt/ext/aflow/scripts/optimized") -> int:
+    """加载最佳表现的轮次"""
+    data_utils = DataUtils(f"{optimized_path}/{dataset}")
+
+    # 使用get_top_rounds获取得分最高的轮次
+    top_rounds = data_utils.get_top_rounds(sample=2, mode="Graph")
+    if not top_rounds[1]:
+        return 1
+
+    return top_rounds[1]["round"]
+
+
+def load_workflow_class(graph_path: str):
+    """动态加载工作流类"""
+    spec = importlib.util.spec_from_file_location("workflow_module", graph_path)
+    module = importlib.util.module_from_spec(spec)
+    sys.modules["workflow_module"] = module
+    spec.loader.exec_module(module)
+    return module.Workflow
+
+
+async def aflow_inference(
+    dataset: DatasetType,
+    question: str,
+    entry_point: Optional[str] = None,
+    round: Optional[int] = None,
+    llm_name: str = "gpt-4o-mini",
+    optimized_path: str = "metagpt/ext/aflow/scripts/optimized",
+) -> Tuple[str, float]:
+    """AFLOW推理接口
+
+    Args:
+        dataset: 数据集名称
+        question: 输入问题
+        round: 指定使用的轮次，如果为None则使用最佳轮次
+        llm_name: 使用的LLM模型名称
+        optimized_path: 优化结果保存路径
+
+    Returns:
+        (答案, 成本)的元组
+    """
+    # 如果没有指定轮次，使用最佳轮次
+    if round is None:
+        round = load_best_round(dataset, optimized_path)
+
+    logger.info(f"Using round {round} for inference")
+
+    # 构建工作流路径并加载
+    graph_path = Path(optimized_path) / dataset / "workflows" / f"round_{round}" / "graph.py"
+    if not graph_path.exists():
+        raise FileNotFoundError(f"Workflow file not found: {graph_path}")
+
+    # 动态加载工作流类
+    WorkflowClass = load_workflow_class(str(graph_path))
+
+    # 创建工作流实例
+    llm_config = ModelsConfig.default().get(llm_name)
+    workflow = WorkflowClass(
+        name=f"{dataset}_workflow",
+        llm_config=llm_config,
+        dataset=dataset,
+    )
+
+    # 执行推理
+    if dataset in ["MBPP", "HumanEval"]:
+        # 代码类任务需要额外的entry_point参数
+        answer, cost = await workflow(question, entry_point=entry_point)
+    else:
+        answer, cost = await workflow(question)
+
+    return answer, cost
+
+
+if __name__ == "__main__":
+    asyncio.run(
+        aflow_inference(
+            dataset="MBPP",
+            question="write a function named add_two_numbers to calculate the sum of two numbers",
+            entry_point="add_two_numbers",
+        )
+    )