Download dataflow_agent/workflow/wf_operator_qa.py from pzp5700/Paper2Any: direct link, hf CLI and curl.
- Browser
- Download file 12.6 kB
-
https://huggingface.co/spaces/pzp5700/Paper2Any/resolve/main/dataflow_agent/workflow/wf_operator_qa.py
- Command line
-
hf download hf://spaces/pzp5700/Paper2Any/dataflow_agent/workflow/wf_operator_qa.py
-
curl -L -o wf_operator_qa.py https://huggingface.co/spaces/pzp5700/Paper2Any/resolve/main/dataflow_agent/workflow/wf_operator_qa.py
12.6 kB
| """ | |
| OperatorQA Workflow - 算子问答工作流 | |
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | |
| 生成时间: 2025-12-01 | |
| 本工作流实现了基于 Agentic RAG 的算子问答功能: | |
| 1. 前置工具:获取用户查询、对话历史 | |
| 2. 后置工具:RAG 检索、获取算子信息/源码/参数(LLM 自主调用) | |
| 3. 支持多轮对话 | |
| """ | |
| from __future__ import annotations | |
| import json | |
| from typing import Any, Dict, List, Optional | |
| from pydantic import BaseModel, Field | |
| from langchain.tools import tool | |
| from dataflow_agent.state import MainState | |
| from dataflow_agent.graphbuilder.graph_builder import GenericGraphBuilder | |
| from dataflow_agent.workflow.registry import register | |
| from dataflow_agent.toolkits.tool_manager import get_tool_manager | |
| from dataflow_agent.agentroles.data_agents.operator_qa_agent import ( | |
| OperatorQAAgent, | |
| OperatorRAGService, | |
| create_operator_qa_agent, | |
| ) | |
| from dataflow_agent.toolkits.filetool.filetools import ( | |
| read_file_content, | |
| list_directory_content, | |
| ) | |
| from dataflow_agent.logger import get_logger | |
| log = get_logger(__name__) | |
| # ============================================================================== | |
| # Workflow Factory | |
| # ============================================================================== | |
| def create_operator_qa_graph() -> GenericGraphBuilder: | |
| """ | |
| Workflow factory: dfa run --wf operator_qa | |
| 构建算子问答工作流图,支持: | |
| 1. RAG 检索相关算子 | |
| 2. 获取算子详细信息 | |
| 3. 获取算子源码(后置工具) | |
| 4. 多轮对话 | |
| """ | |
| builder = GenericGraphBuilder( | |
| state_model=MainState, | |
| entry_point="operator_qa_node" | |
| ) | |
| # 创建共享的 RAG 服务实例 | |
| rag_service = OperatorRAGService() | |
| # 创建共享的消息历史管理器(用于多轮对话) | |
| from dataflow_agent.graphbuilder.message_history import AdvancedMessageHistory | |
| shared_message_history = AdvancedMessageHistory() | |
| # ========================================================================== | |
| # 前置工具 (Pre-Tools) - 每次自动执行 | |
| # ========================================================================== | |
| def get_user_query(state: MainState) -> str: | |
| """获取用户查询""" | |
| return state.request.target or "" | |
| # 注意:对话历史由 BaseAgent 的 AdvancedMessageHistory 自动管理, | |
| # 不再通过 pre-tool 嵌入 prompt | |
| # ========================================================================== | |
| # 后置工具 (Post-Tools) - LLM 自主决定是否调用 | |
| # ========================================================================== | |
| class SearchOperatorsInput(BaseModel): | |
| """搜索算子的输入参数""" | |
| query: str = Field(description="搜索查询,描述需要的算子功能,如 '过滤文本' '数据清洗'") | |
| top_k: int = Field(default=5, description="返回结果数量,默认5个") | |
| def search_operators(query: str, top_k: int = 5) -> str: | |
| """ | |
| 根据功能描述搜索相关算子。这是主要的算子检索工具。 | |
| 当用户询问某类功能的算子、或需要查找算子时,首先使用此工具。 | |
| 如果对话历史中已有相关算子信息,可以不调用此工具直接回答。 | |
| """ | |
| result = rag_service.search_and_get_info(query, top_k=top_k) | |
| return json.dumps(result, ensure_ascii=False, indent=2) | |
| class GetOperatorInfoInput(BaseModel): | |
| """获取算子详细信息的输入参数""" | |
| operator_name: str = Field(description="要获取信息的算子名称,如 'PromptedFilter'") | |
| def get_operator_info(operator_name: str) -> str: | |
| """ | |
| 获取指定算子的详细描述信息。 | |
| 当用户询问某个特定算子的功能、用途时使用此工具。 | |
| """ | |
| info = rag_service.get_operator_info([operator_name]) | |
| return info | |
| class GetOperatorSourceInput(BaseModel): | |
| """获取算子源码的输入参数""" | |
| operator_name: str = Field(description="要获取源码的算子名称,如 'PromptedFilter'") | |
| def get_operator_source_code(operator_name: str) -> str: | |
| """ | |
| 获取指定算子的完整源代码。 | |
| 当用户询问算子的具体实现细节、或需要了解算子内部逻辑时使用此工具。 | |
| """ | |
| return rag_service.get_operator_source(operator_name) | |
| class GetOperatorParamsInput(BaseModel): | |
| """获取算子参数的输入参数""" | |
| operator_name: str = Field(description="要获取参数信息的算子名称") | |
| def get_operator_parameters(operator_name: str) -> str: | |
| """ | |
| 获取指定算子的参数详情,包括 __init__ 和 run 方法的参数。 | |
| 当用户询问算子如何配置、参数含义时使用此工具。 | |
| """ | |
| params = rag_service.get_operator_params(operator_name) | |
| return json.dumps(params, ensure_ascii=False, indent=2) | |
| # ========================================================================== | |
| # 文件操作工具 (File Tools) - LLM 自主决定是否调用 | |
| # ========================================================================== | |
| class ReadFileInput(BaseModel): | |
| """读取文件内容的输入参数""" | |
| file_path: str = Field(description="文件路径,可以是相对路径(相对于项目根目录)或绝对路径") | |
| start_line: int = Field(default=None, description="起始行号(从1开始,可选)。不指定则从第1行开始") | |
| end_line: int = Field(default=None, description="结束行号(包含,可选)。不指定则读取到文件末尾") | |
| def read_text_file(file_path: str, start_line: int = None, end_line: int = None) -> str: | |
| """ | |
| 读取文本文件内容。 | |
| 支持读取项目内的任意文本文件,可指定读取的行范围。 | |
| 出于安全考虑,只能读取项目根目录内的文件。 | |
| 当用户需要查看某个文件的内容、或需要了解项目中某个文件的具体实现时使用此工具。 | |
| Examples: | |
| >>> read_text_file("README.md") # 读取整个文件 | |
| >>> read_text_file("src/main.py", start_line=10, end_line=20) # 读取第10-20行 | |
| """ | |
| result = read_file_content(file_path, start_line, end_line) | |
| return json.dumps(result, ensure_ascii=False, indent=2) | |
| class ListDirectoryInput(BaseModel): | |
| """查看目录内容的输入参数""" | |
| dir_path: str = Field(description="目录路径,可以是相对路径(相对于项目根目录)或绝对路径") | |
| show_hidden: bool = Field(default=False, description="是否显示隐藏文件(以.开头的文件),默认 False") | |
| recursive: bool = Field(default=False, description="是否递归显示子目录内容,默认 False") | |
| def list_directory(dir_path: str, show_hidden: bool = False, recursive: bool = False) -> str: | |
| """ | |
| 查看目录内容。 | |
| 列出指定目录下的文件和子目录,支持 Windows 和 Linux 系统。 | |
| 出于安全考虑,只能查看项目根目录内的目录。 | |
| 当用户需要了解项目结构、查看某个目录下有哪些文件时使用此工具。 | |
| Examples: | |
| >>> list_directory(".") # 列出项目根目录 | |
| >>> list_directory("src", show_hidden=True) # 列出 src 目录,包含隐藏文件 | |
| >>> list_directory("dataflow_agent", recursive=True) # 递归列出目录 | |
| """ | |
| result = list_directory_content(dir_path, show_hidden, recursive) | |
| return json.dumps(result, ensure_ascii=False, indent=2) | |
| # ========================================================================== | |
| # 节点定义 (Nodes) | |
| # ========================================================================== | |
| # DATAFLOW_LOG_LEVEL=DEBUG python /mnt/DataFlow/lz/proj/agentgroup/ziyi/DataFlow-Agent/script/run_dfa_operator_qa.py --query "GeneralFilter算子有什么作用?" | |
| async def operator_qa_node(state: MainState) -> MainState: | |
| """ | |
| 算子问答节点 | |
| 使用 OperatorQAAgent 处理用户查询,支持: | |
| - LLM 自主决定是否调用 RAG 检索 | |
| - 多轮对话(由 BaseAgent 通过 messages 数组管理) | |
| - 工具调用(检索算子、获取源码、参数等) | |
| """ | |
| # 多轮对话:如果有历史消息,追加新的用户问题 | |
| # 第一轮时 state.messages 为空,由 build_messages 构建 system + user | |
| # 后续轮次手动追加用户问题,避免 build_messages 不被调用导致新问题丢失 | |
| if state.messages: | |
| user_query = state.request.target or "" | |
| if user_query: | |
| from langchain_core.messages import HumanMessage | |
| state.messages = state.messages + [HumanMessage(content=user_query)] | |
| log.debug(f"追加用户问题到历史,当前共 {len(state.messages)} 条消息") | |
| tm = get_tool_manager() | |
| # 创建 Agent(使用共享的消息历史管理器) | |
| agent = create_operator_qa_agent( | |
| tool_manager=tm, | |
| rag_service=rag_service, | |
| model_name=state.request.model or "gpt-4o", | |
| temperature=0.1, | |
| max_tokens=4096, | |
| parser_type="json", | |
| tool_mode="auto", | |
| message_history=shared_message_history, # 共享消息历史 | |
| ) | |
| # 执行 Agent(use_agent=True 启用工具调用) | |
| state = await agent.execute(state, use_agent=True) | |
| # 记录结果 | |
| result = state.agent_results.get("operator_qa", {}) | |
| log.info(f"OperatorQA 执行结果: {result}") | |
| return state | |
| # ========================================================================== | |
| # 图结构 (Graph Structure) | |
| # ========================================================================== | |
| nodes = { | |
| "operator_qa_node": operator_qa_node, | |
| "_end_": lambda state: state, | |
| } | |
| edges = [ | |
| ("operator_qa_node", "_end_"), | |
| ] | |
| # 关键:将节点 role 映射为 "operator_qa",与 Agent 的 role_name 和前置工具的 role 保持一致 | |
| builder.add_nodes(nodes, role_mapping={"operator_qa_node": "operator_qa"}).add_edges(edges) | |
| return builder | |
| # ============================================================================== | |
| # 便捷执行函数 | |
| # ============================================================================== | |
| async def run_operator_qa( | |
| query: str, | |
| chat_api_url: str = "http://123.129.219.111:3000/v1/", | |
| api_key: Optional[str] = None, | |
| model: str = "gpt-4o", | |
| ) -> Dict[str, Any]: | |
| """ | |
| 执行算子问答 | |
| Args: | |
| query: 用户查询 | |
| chat_api_url: Chat API 地址 | |
| api_key: API 密钥 | |
| model: 模型名称 | |
| Returns: | |
| 问答结果字典 | |
| """ | |
| import os | |
| from dataflow_agent.state import DFRequest | |
| # 构建请求 | |
| req = DFRequest( | |
| language="zh", | |
| chat_api_url=chat_api_url, | |
| api_key=api_key or os.getenv("DF_API_KEY", ""), | |
| model=model, | |
| target=query, | |
| ) | |
| # 构建状态 | |
| state = MainState(request=req, messages=[]) | |
| # 执行工作流 | |
| graph_builder = create_operator_qa_graph() | |
| graph = graph_builder.build() | |
| final_state = await graph.ainvoke(state) | |
| # 提取结果 | |
| result = final_state.get("agent_results", {}).get("operator_qa", {}) | |
| return { | |
| "answer": result.get("results", {}).get("answer", ""), | |
| "related_operators": result.get("results", {}).get("related_operators", []), | |
| "code_snippet": result.get("results", {}).get("code_snippet", ""), | |
| } | |