LangChain工具集成报错求助：序列化失败与Arxiv工具绑定400错误

大家好，我在做LangChain工具集成的时候碰到了两个棘手的问题，折腾了好久都没搞定，想请各位大佬帮忙看看！

遇到的错误情况 #

首先是用TavilySearchResults工具时，出现了Pydantic序列化失败的问题：

Failed to use model_dump to serialize <class 'langchain_community.tools.tavily_search.tool.TavilySearchResults'> to JSON: PydanticSerializationError(Unable to serialize unknown type: <class 'pydantic._internal._model_construction.ModelMetaclass'>)

后来换成ArxivQueryRun工具，又碰到了LLM绑定工具时的400请求错误：

BadRequestError: Error code: 400 - {'error': {'message': "'tools.0.type' : property 'type' is missing", 'type': 'invalid_request_error'}}

相关代码片段 #

Arxiv工具初始化代码

from langchain_community.utilities import ArxivAPIWrapper
from langchain_community.tools import ArxivQueryRun

# Initialize the Arxiv wrapper and tool
arxiv_wrapper = ArxivAPIWrapper(top_k_results=1, doc_content_chars_max=300)
arxiv_tool = ArxivQueryRun(api_wrapper=arxiv_wrapper)

tools = [tavily_search]  # 这里替换成arxiv_tool也会报错
llm_with_tools = llm.bind(tools=tools)

状态图Graph代码

arcus_builder = StateGraph(ARCUSState)
arcus_builder.add_node("generate_co_info", generate_co_info)
arcus_builder.add_node("search_web", search_web)
arcus_builder.add_node("search_wiki", search_wiki)
arcus_builder.add_node("save_result", save_result)
arcus_builder.add_node("write_research", write_research)
arcus_builder.add_node("use_case_generator", use_case_generator)
arcus_builder.add_node("resource_collector", resource_collector)
tool_node = ToolNode(tools=tools)
arcus_builder.add_node("tools", tool_node)
arcus_builder.add_node("write_report", write_report)

arcus_builder.add_edge(START, "generate_co_info")
arcus_builder.add_edge("generate_co_info", "search_web")
arcus_builder.add_edge("generate_co_info", "search_wiki")
arcus_builder.add_edge("search_web", "save_result")
arcus_builder.add_edge("search_wiki", "save_result")
arcus_builder.add_edge("save_result", "write_research")
arcus_builder.add_edge("write_research", "use_case_generator")
arcus_builder.add_edge("use_case_generator", "resource_collector")
arcus_builder.add_conditional_edges(
    "resource_collector",
    tools_condition_custom,
    {
        "tools": "tools",
        "write_report": "write_report",
    }
)
arcus_builder.add_edge("tools", "resource_collector")
arcus_builder.add_edge("write_report", END)


memory = MemorySaver()
arcus_graph = arcus_builder.compile(checkpointer=memory)

display(Image(arcus_graph.get_graph(xray=1).draw_mermaid_png()))

调用代码

company_name = "Tesla"
messages = [HumanMessage(f" Company Name: {company_name}")]
thread = {"configurable": {"thread_id": "1"}}
research = arcus_graph.invoke({"messages": messages}, thread)

补充说明 #

目前不使用工具的时候，整个状态图流程是完全正常跑通的，但只要一加入工具（不管是Tavily还是Arxiv）就会出现对应的错误。我现在的需求是让绑定了工具的LLM能在状态图里正常工作，解决工具的序列化和绑定问题。

备注：内容来源于stack exchange，提问作者Vaibhav Prajapati