With langchain v0.2.2 use ChatTongyi(streaming=True) occurred error ' KeyError: 'request' ' about langchain

[MannaYang]

The function tool code :

def get_current_time():
    weekdays = ["Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday"]
    time_zone = pytz.timezone('Asia/Shanghai')
    date_time = datetime.now(time_zone)
    time_format = date_time.strftime("%Y-%m-%d %H:%M:%S")
    week_day = date_time.weekday()
    return f"CurrentTime: {time_format} {weekdays[week_day]}"

[wangyaoyong-wyy]

I also encountered this issue while using ChatTongyi, and after breakpoint source code feedback, Multiple tool_calls are not supported in message This feature will be supported in the future， Perhaps Tongyi does not currently support multi model scheduling?

[MannaYang]

I also encountered this issue while using ChatTongyi, and after breakpoint source code feedback, Multiple tool_calls are not supported in message This feature will be supported in the future， Perhaps Tongyi does not currently support multi model scheduling?

Maybe you're right. I tried to use create_react_agent.invoke() , like AgentExecutor on_llm_new_token callback , but it doesn't work.

[wangyaoyong-wyy]

This issue may occur when using create_tool_calling-agent. According to the error code displayed on the dashscope official website, it is due to passing incorrect parameters that this issue occurs. If you switch to create_json_chat_agent, this issue will not occur

prompt = hub.pull("hwchase17/react-chat-json")
agent = create_json_chat_agent(chat, tools, prompt)
# agent = create_tool_calling_agent(chat,tools,prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True, handle_parsing_errors=True)
resp =agent_executor.invoke({"input": "what is LangChain latest version?"})
print(resp)

[MannaYang]

This issue may occur when using create_tool_calling-agent. According to the error code displayed on the dashscope official website, it is due to passing incorrect parameters that this issue occurs. If you switch to create_json_chat_agent, this issue will not occur

prompt = hub.pull("hwchase17/react-chat-json")
agent = create_json_chat_agent(chat, tools, prompt)
# agent = create_tool_calling_agent(chat,tools,prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True, handle_parsing_errors=True)
resp =agent_executor.invoke({"input": "what is LangChain latest version?"})
print(resp)

With langgraph.prebuilt.create_react_agent() is return CompiledGraph object.
Currently ChatTongyi() calling stream directly and carrying the tool still reports errors("Multiple tool_calls are not supported in message").
So I want to pass config={“callbacks”: “Streamingxxxxxx”} via invoke to get the chunk generated by on_llm_new_token , but it doesn't work.