Merge pull request #28 from chatchat-space/dev

Dev

Author: yuehua-s

libs/chatchat-server/chatchat/server/agent/graphs_factory/__init__.py

@@ -1,6 +1,6 @@
-from .reflexion import reflexion
-from .article_generation import article_generation
 from .base_rag import BaseRagGraph
 from .base_graph import BaseAgentGraph
-from .text_to_sql import TextToSQLGraph
 from .plan_and_execute import PlanExecuteGraph
+from .reflexion import ReflexionGraph
+from .text_to_sql import TextToSQLGraph
+from .article_generation import article_generation

libs/chatchat-server/chatchat/server/agent/graphs_factory/base_graph.py

@@ -15,7 +15,7 @@
 class BaseAgentGraph(Graph):
     name = "base_agent"
     label = "agent"
-    title = "聊天机器人[纯净版]"
+    title = "聊天机器人"
 
     def __init__(self,
                  llm: ChatOpenAI,

libs/chatchat-server/chatchat/server/agent/graphs_factory/reflexion.py

@@ -16,7 +16,7 @@
 class TextToSQLGraph(Graph):
     name = "text_to_sql"
     label = "agent"
-    title = "数据库查询机器人"
+    title = "数据库查询机器人[Beta]"
 
     def __init__(self,
                  llm: ChatOpenAI,

libs/chatchat-server/chatchat/server/agent/graphs_factory/text_to_sql.py

@@ -1,10 +1,12 @@
 import requests
-from chatchat.server.pydantic_v1 import Field
+from pydantic import Field
+
 from .tools_registry import BaseToolOutput, regist_tool
 from chatchat.server.utils import get_tool_config
 
 BASE_URL = "https://restapi.amap.com/v5/place/text"
 
+
 def amap_poi_search_engine(keywords: str,types: str,config: dict):
     API_KEY = config["api_key"]
     params = {
@@ -19,10 +21,9 @@ def amap_poi_search_engine(keywords: str,types: str,config: dict):
         return {"error": "API request failed"}
 
 
-
 @regist_tool(title="高德地图POI搜索")
 def amap_poi_search(location: str = Field(description="'实际地名'或者'具体的地址',不能使用简称或者别称"),
-                types: str = Field(description="POI类型，比如商场、学校、医院等等")):
+                    types: str = Field(description="POI类型，比如商场、学校、医院等等")):
     """ A wrapper that uses Amap to search."""
     tool_config = get_tool_config("amap")
     return BaseToolOutput(amap_poi_search_engine(keywords=location,types=types,config=tool_config))

libs/chatchat-server/chatchat/server/agent/tools_factory/amap_poi_search.py

@@ -1,11 +1,12 @@
 import requests
-from chatchat.server.pydantic_v1 import Field
+from pydantic import Field
 from .tools_registry import BaseToolOutput, regist_tool
 from chatchat.server.utils import get_tool_config
 
 BASE_DISTRICT_URL = "https://restapi.amap.com/v3/config/district"
 BASE_WEATHER_URL = "https://restapi.amap.com/v3/weather/weatherInfo"
 
+
 def get_adcode(city: str, config: dict) -> str:
     """Get the adcode"""
     API_KEY = config["api_key"]
@@ -22,6 +23,7 @@ def get_adcode(city: str, config: dict) -> str:
     else:
         return None
 
+
 def get_weather(adcode: str, config: dict) -> dict:
     """Get  weather information."""
     API_KEY = config["api_key"]
@@ -36,6 +38,7 @@ def get_weather(adcode: str, config: dict) -> dict:
     else:
         return {"error": "API request failed"}
 
+
 @regist_tool(title="高德地图天气查询")
 def amap_weather(city: str = Field(description="城市名")):
     """A wrapper that uses Amap to get weather information."""

libs/chatchat-server/chatchat/server/agent/tools_factory/amap_weather.py

@@ -1,4 +1,4 @@
-from chatchat.server.pydantic_v1 import Field
+from pydantic import Field
 
 from .tools_registry import BaseToolOutput, regist_tool
 

libs/chatchat-server/chatchat/server/agent/tools_factory/calculate.py

@@ -1,6 +1,6 @@
 from langchain_community.tools.sql_database.tool import QuerySQLDataBaseTool
 from langchain_community.utilities.sql_database import SQLDatabase
-from chatchat.server.pydantic_v1 import Field
+from pydantic import Field
 from chatchat.server.utils import get_tool_config, build_logger
 from .tools_registry import BaseToolOutput, regist_tool
 

libs/chatchat-server/chatchat/server/agent/tools_factory/querysql.py

@@ -1,22 +1,29 @@
 from typing import Dict, List
-
+from pydantic import Field
 from langchain.docstore.document import Document
-from langchain.text_splitter import RecursiveCharacterTextSplitter
-from langchain_community.utilities.bing_search import BingSearchAPIWrapper
-from langchain_community.utilities.duckduckgo_search import DuckDuckGoSearchAPIWrapper
-from langchain_community.utilities.searx_search import SearxSearchWrapper
-from langchain_community.utilities.tavily_search import TavilySearchAPIWrapper
-from markdownify import markdownify
-from strsimpy.normalized_levenshtein import NormalizedLevenshtein
 
 from chatchat.settings import Settings
-from chatchat.server.pydantic_v1 import Field
 from chatchat.server.utils import get_tool_config
 
-from .tools_registry import BaseToolOutput, regist_tool, format_context
+from .tools_registry import BaseToolOutput, regist_tool
+
+
+def duckduckgo_search(text, top_k: int):
+    from langchain_community.utilities.duckduckgo_search import DuckDuckGoSearchAPIWrapper
+    search = DuckDuckGoSearchAPIWrapper()
+    return search.results(text, top_k)
+
+
+def tavily_search(text, config, top_k: int):
+    from langchain_community.utilities.tavily_search import TavilySearchAPIWrapper
+    search = TavilySearchAPIWrapper(
+        tavily_api_key=config["tavily_key"],
+    )
+    return search.results(text, top_k)
 
 
 def searx_search(text, config, top_k: int):
+    from langchain_community.utilities.searx_search import SearxSearchWrapper
     search = SearxSearchWrapper(
         searx_host=config["host"],
         engines=config["engines"],
@@ -27,31 +34,24 @@ def searx_search(text, config, top_k: int):
 
 
 def bing_search(text, config, top_k: int):
+    from langchain_community.utilities.bing_search import BingSearchAPIWrapper
     search = BingSearchAPIWrapper(
         bing_subscription_key=config["bing_key"],
         bing_search_url=config["bing_search_url"],
     )
     return search.results(text, top_k)
 
 
-def tavily_search(text, config, top_k: int):
-    search = TavilySearchAPIWrapper(
-        tavily_api_key=config["tavily_key"],
-    )
-    return search.results(text, top_k)
-
-
-def duckduckgo_search(text, config, top_k: int):
-    search = DuckDuckGoSearchAPIWrapper()
-    return search.results(text, top_k)
-
-
 def metaphor_search(
     text: str,
     config: dict,
-    top_k:int
+    top_k: int
 ) -> List[Dict]:
     from metaphor_python import Metaphor
+    from langchain.text_splitter import RecursiveCharacterTextSplitter
+    from strsimpy.normalized_levenshtein import NormalizedLevenshtein
+
+    from markdownify import markdownify
 
     client = Metaphor(config["metaphor_api_key"])
     search = client.search(text, num_results=top_k, use_autoprompt=True)
@@ -101,20 +101,6 @@ def metaphor_search(
 }
 
 
-def search_result2docs(search_results) -> List[Document]:
-    docs = []
-    for result in search_results:
-        doc = Document(
-            page_content=result["snippet"] if "snippet" in result.keys() else "",
-            metadata={
-                "source": result["link"] if "link" in result.keys() else "",
-                "filename": result["title"] if "title" in result.keys() else "",
-            },
-        )
-        docs.append(doc)
-    return docs
-
-
 def search_engine(query: str, top_k: int = 0, engine_name: str = "", config: dict = {}):
     config = config or get_tool_config("search_internet")
     if top_k <= 0:
@@ -124,12 +110,10 @@ def search_engine(query: str, top_k: int = 0, engine_name: str = "", config: dic
     results = search_engine_use(
         text=query, config=config["search_engine_config"][engine_name], top_k=top_k
     )
-    docs = [x for x in search_result2docs(results) if x.page_content and x.page_content.strip()]
-    return {"docs": docs, "search_engine": engine_name}
+    return results
 
 
 @regist_tool(title="互联网搜索")
 def search_internet(query: str = Field(description="query for Internet search")):
     """Use this tool to use bing search engine to search the internet and get information."""
-    # return BaseToolOutput(search_engine(query=query), format=format_context)
     return BaseToolOutput(search_engine(query=query))

libs/chatchat-server/chatchat/server/agent/tools_factory/search_internet.py

@@ -4,7 +4,6 @@
 from chatchat.server.agent.tools_factory.tools_registry import (
     BaseToolOutput,
     regist_tool,
-    format_context,
 )
 from chatchat.server.knowledge_base.kb_api import list_kbs
 from chatchat.server.knowledge_base.kb_doc_api import search_docs

libs/chatchat-server/chatchat/server/agent/tools_factory/search_local_knowledgebase.py

@@ -1,8 +1,6 @@
 # LangChain 的 Shell 工具
+from pydantic import Field
 from langchain_community.tools import ShellTool
-
-from chatchat.server.pydantic_v1 import Field
-
 from .tools_registry import BaseToolOutput, regist_tool
 
 

libs/chatchat-server/chatchat/server/agent/tools_factory/shell.py

@@ -6,9 +6,9 @@
 
 import openai
 from PIL import Image
+from pydantic import Field
 
 from chatchat.settings import Settings
-from chatchat.server.pydantic_v1 import Field
 from chatchat.server.utils import MsgType, get_tool_config, get_model_info
 
 from .tools_registry import BaseToolOutput, regist_tool

libs/chatchat-server/chatchat/server/agent/tools_factory/text2image.py

@@ -2,12 +2,12 @@
 from requests.auth import HTTPBasicAuth
 from urllib.parse import parse_qs
 from typing import Optional
+from pydantic import Field
 
 from langchain_core.prompts import ChatPromptTemplate
 from langchain_core.output_parsers import StrOutputParser
 from langchain_core.runnables import RunnablePassthrough
 
-from chatchat.server.pydantic_v1 import Field
 from chatchat.server.utils import (
     get_tool_config,
     get_default_llm,

libs/chatchat-server/chatchat/server/agent/tools_factory/text2promql.py

@@ -4,8 +4,8 @@
 from langchain_experimental.sql import SQLDatabaseChain, SQLDatabaseSequentialChain
 from sqlalchemy import event
 from sqlalchemy.exc import OperationalError
+from pydantic import Field
 
-from chatchat.server.pydantic_v1 import Field
 from chatchat.server.utils import get_tool_config
 
 from .tools_registry import BaseToolOutput, regist_tool

libs/chatchat-server/chatchat/server/agent/tools_factory/text2sql.py

@@ -165,3 +165,24 @@ def format_context(self: BaseToolOutput) -> str:
             context += doc + "\n\n"
 
     return context
+
+
+def format_context(self: BaseToolOutput) -> str:
+    """
+    将包含知识库输出的ToolOutput格式化为 LLM 需要的字符串
+    """
+    context = ""
+    docs = self.data["docs"]
+    source_documents = []
+
+    for inum, doc in enumerate(docs):
+        doc = DocumentWithVSId.parse_obj(doc)
+        source_documents.append(doc.page_content)
+
+    if len(source_documents) == 0:
+        context = "没有找到相关文档,请更换关键词重试"
+    else:
+        for doc in source_documents:
+            context += doc + "\n\n"
+
+    return context

libs/chatchat-server/chatchat/server/agent/tools_factory/tools_registry.py

@@ -3,8 +3,8 @@
 """
 import requests
 import re
+from pydantic import Field
 
-from chatchat.server.pydantic_v1 import Field
 from chatchat.server.agent.tools_factory.tools_registry import format_context
 from chatchat.server.utils import get_tool_config, build_logger
 

libs/chatchat-server/chatchat/server/agent/tools_factory/url_reader.py

@@ -2,8 +2,8 @@
 简单的单参数输入工具实现，用于查询现在天气的情况
 """
 import requests
+from pydantic import Field
 
-from chatchat.server.pydantic_v1 import Field
 from chatchat.server.utils import get_tool_config
 
 from .tools_registry import BaseToolOutput, regist_tool

libs/chatchat-server/chatchat/server/agent/tools_factory/weather_check.py

@@ -1,10 +1,7 @@
 # LangChain 的 WikipediaQueryRun 工具
+from pydantic import Field
 from langchain_community.tools import WikipediaQueryRun
 from langchain_community.utilities import WikipediaAPIWrapper
-from chatchat.server.pydantic_v1 import Field
-
-
-
 from .tools_registry import BaseToolOutput, regist_tool
 
 

libs/chatchat-server/chatchat/server/agent/tools_factory/wikipedia_search.py

@@ -1,8 +1,6 @@
 # Langchain 自带的 Wolfram Alpha API 封装
-
-from chatchat.server.pydantic_v1 import Field
+from pydantic import Field
 from chatchat.server.utils import get_tool_config
-
 from .tools_registry import BaseToolOutput, regist_tool
 
 

libs/chatchat-server/chatchat/server/agent/tools_factory/wolfram.py

@@ -256,6 +256,8 @@ async def handle_user_input(
                         response_last = response["content"]
                     elif "response" in response:  # plan_execute_agent
                         response_last = response["response"]
+                    elif "answer" in response:  # reflexion
+                        response_last = response["answer"]
 
                     with st.status(node, expanded=True) as status:
                         st.json(response, expanded=True)
@@ -356,7 +358,7 @@ def graph_agent_page():
 
             tools_list = list_tools()
             # tool_names = ["None"] + list(tools_list)
-            if selected_graph == "数据库查询机器人":
+            if selected_graph == "数据库查询机器人[Beta]":
                 selected_tools = st.multiselect(
                     label="选择工具",
                     options=["query_sql_data"],
@@ -394,7 +396,7 @@ def graph_agent_page():
         st.title("自媒体文章生成")
         with st.chat_message("assistant"):
             st.write("Hello 👋😊，我是自媒体文章生成 Agent，输入任意内容以启动工作流～")
-    elif selected_graph == "数据库查询机器人":
+    elif selected_graph == "数据库查询机器人[Beta]":
         st.title("数据库查询")
         with st.chat_message("assistant"):
             st.write("Hello 👋😊，我是数据库查询机器人，输入你想查询的内容～")
@@ -412,7 +414,7 @@ def graph_agent_page():
             st.rerun()
         if selected_graph == "article_generation":
             user_input = cols[2].chat_input("请你帮我生成一篇自媒体文章 (换行:Shift+Enter)")
-        elif selected_graph == "数据库查询机器人":
+        elif selected_graph == "数据库查询机器人[Beta]":
             user_input = cols[2].chat_input("请你帮忙调用工具, 查看组织`tcs_public`的成员有哪些？(换行:Shift+Enter)")
         else:
             user_input = cols[2].chat_input("尝试输入任何内容和我聊天呦 (换行:Shift+Enter)")
@@ -434,6 +436,7 @@ def graph_agent_page():
                               max_tokens=None,
                               temperature=st.session_state["temperature"],
                               stream=True)
+    logger.info(f"Loaded llm: {llm}")
 
     # 创建 langgraph 实例
     graph_class = get_graph_class_by_label_and_title(label="agent", title=selected_graph)
@@ -446,15 +449,15 @@ def graph_agent_page():
     graph = graph_class.get_graph()
     if not graph:
         raise ValueError(f"Graph '{selected_graph}' is not registered.")
+    st.toast(f"已加载工作流: {selected_graph}")
 
     # langgraph 配置文件
     graph_config = {
         "configurable": {
             "thread_id": st.session_state["conversation_id"]
         },
     }
-
-    logger.info(f"graph: '{selected_graph}', configurable: '{graph_config}'")
+    logger.info(f"Loaded graph: '{selected_graph}', configurable: '{graph_config}'")
 
     # 绘制流程图并缓存
     graph_flow_image_name = f"{selected_graph}_flow_image"