langgraph/tutorials/multi_agent/multi-agent-collaboration/ #511
giscus[bot]
bot
Replies: 25 comments 25 replies
-
|
awesome but how can we change to start to legal Team collaboration for searching and retrieving text ? |
Beta Was this translation helpful? Give feedback.
-
|
Could you share more about what you're struggling with? Are you asking whether the design pattern permits other agent personas and tools beyond the ones shown in the example? |
Beta Was this translation helpful? Give feedback.
-
|
This tutorial cannot run on gpt-4o, will give bugs " keyerror: chart_generator". But works fine when use gpt-4-1106-preview. |
Beta Was this translation helpful? Give feedback.
-
|
Neither gpt-4o, nor gpt-4-turbo, gpt-4-1106-preview works. |
Beta Was this translation helpful? Give feedback.
-
|
Did you resolve it? |
Beta Was this translation helpful? Give feedback.
-
|
Looks like gpt-3.5-turbo also causes the "KeyError: 'chart_generator'" and gpt-4 triggers rate limit issues: I'm too lazy to add code to handle rate limits but interestingly it works fine for "gpt-4-1106-preview" which is enough to prove out the concept. |
Beta Was this translation helpful? Give feedback.
-
|
me2："KeyError: 'chart_generator'" |
Beta Was this translation helpful? Give feedback.
-
|
I'm using gpt-4-turbo and it works. But, when I tried swapping out the 'tavily_tool' for a duckduckgo search tool. Everything works perfectly retrieving information from the internet but then encounters the same error when generating the plot: KeyError: 'chart_generator' |
Beta Was this translation helpful? Give feedback.
-
|
really annoying，isn't It |
Beta Was this translation helpful? Give feedback.
-
|
For gpt-4o, change the following code to make it work: ...
research_agent = create_agent(
llm,
[tavily_tool],
system_message="You should provide accurate data for use, "
"and source code shouldn't be the final answer",
)
...
chart_agent = create_agent(
llm,
[python_repl],
system_message="Run the python code to display the chart.",
) |
Beta Was this translation helpful? Give feedback.
-
do you know how to make this work with opensource models? |
Beta Was this translation helpful? Give feedback.
-
all you need to do is change those two prompts? |
Beta Was this translation helpful? Give feedback.
-
|
Yes, the key point lies in the prompt. The previous prompt caused errors under gpt-4o because it thought it needed to call a tool named chat_generator(which doesn't exist), and it also believed that showing the source code to the user would be sufficient to conclude. The modifications to these two prompts were made to correct these issues. Other open-source LLM may also exhibit different behaviors, requiring appropriate adjustments to the prompts. |
Beta Was this translation helpful? Give feedback.
-
|
thank u for sharing,it's very insightful. :) |
Beta Was this translation helpful? Give feedback.
-
|
Has anyone encountered this error : |
Beta Was this translation helpful? Give feedback.
-
I am trying to set up a react sql agent to work together with a chart_generator agent.
|
Beta Was this translation helpful? Give feedback.
-
|
Your pipeline has a complicated yet effective automated process. But your sample invoke shows how it can get trapped in a local infinity loop between two Nodes/Agents. Beside using hard limits such as revision_number or tried_number to escape from such loop, do you have any architectural idea to solve such issues? |
Beta Was this translation helpful? Give feedback.
-
|
I've broken the graph to the to only SQLAgent, sql_tool_node and the end state. Here everything works as intended, so I am assuming the problem is in the router with passing of the state to the chart_generator. Does the chart_generator have to have access to the sql_tool_node in order to call py_repl_node to perform the plotting? And what does the router do? Here is my router: |
Beta Was this translation helpful? Give feedback.
-
|
Got error： ValueError: status_code: 400 |
Beta Was this translation helpful? Give feedback.
-
I was fine before, using the same model, and I've been getting this error lately, and I'm not sure why |
Beta Was this translation helpful? Give feedback.
-
|
You should format the json response that you display. It's really hard to follow. |
Beta Was this translation helpful? Give feedback.
-
|
I tried to customize this example with my own tools, but I see below error. Any suggestions, please? |
Beta Was this translation helpful? Give feedback.
-
|
You will need to use the gpt-4-1106-preview model or use this as a reference to change to the gpt-4o model and prompt. |
Beta Was this translation helpful? Give feedback.
-
|
Check on the State model. Have it already defined sender key |
Beta Was this translation helpful? Give feedback.
-
|
I am getting the following error when. running multi-agent-collobration.ipynb. Can someone help? Error code: 400 - {'error': {'message': "An assistant message with 'tool_calls' must be followed by tool messages responding to each 'tool_call_id'. The following tool_call_ids did not have response messages: call_NEv19stcIQYKEzIodrGshzRv", 'type': 'invalid_request_error', 'param': 'messages', 'code': None}} |
Beta Was this translation helpful? Give feedback.
-
|
Same error |
Beta Was this translation helpful? Give feedback.
-
|
For me it seemed to be a timing error and LLM creativity error. With 4o, the agent was going a lot faster, but when I went to gpt-4-1106-preview, the agent went a lot slower. For a quick hack, I just added a time delay to the call_tool edge: from typing import Literal def router(state) -> Literal["call_tool", "end", "continue"]: I also had the issue mentioned above where the LLM was thinking there was tool called chart_generator, so I made sure I loaded the LLM with temperature of 0: llm = ChatOpenAI(model="gpt-4o", temperature=0) Those seemed to help me. I think we need an intelligent tool that can figure out all these API call rate limit errors for these agents. |
Beta Was this translation helpful? Give feedback.
-
|
Oh this is actually helpful let me try thanks a bunch |
Beta Was this translation helpful? Give feedback.
-
|
I tried using the agent-executor and it works do you think agent executor delays it on purpose |
Beta Was this translation helpful? Give feedback.
-
|
I'm running into an error with this sample code and need some help figuring out how to fix it. ''' ''' |
Beta Was this translation helpful? Give feedback.
-
|
use a better LLM. The arguments in the tool call are wrong and therefore you're getting the error. The code is fine |
Beta Was this translation helpful? Give feedback.
-
|
Your llm might be passing parameters incorrectly so you can use an abstract syntax tree format in python to conform the parameters in the function. you can do this by using a dictionary with a fallback. Include in your prompt, "write it in dictionary format with the following keys." |
Beta Was this translation helpful? Give feedback.
-
|
Hi folks: thanks for the tutorial. Here if I have a tool that gives the result as a pandas df (for e.g. dates and stock prices) and another that plots it (using matplotlib and not a REPL) then how do I "pass" the data (which is a df or lists) between the 2 tools? |
Beta Was this translation helpful? Give feedback.
-
|
Hi folks: thanks for the tutorial. I replaced the told above with the following - 1 that gives the dates and prices as a pandas df and another that plots it (using matplotlib and not a REPL). Will the data pass seamlessly between the 2? Getting the following output. Is it a tool call issue or a data format (json vs pandas) issue? |
Beta Was this translation helpful? Give feedback.
-
|
Taking out the data from the errors messages to give a clearer picture: |
Beta Was this translation helpful? Give feedback.
-
|
Hey guys I've been getting this error |
Beta Was this translation helpful? Give feedback.
-
|
Used the same code |
Beta Was this translation helpful? Give feedback.
-
|
use openai gpt4. Dont waste time on other LLM. why u wan to drive toyota, when u can drive Lamborghini. |
Beta Was this translation helpful? Give feedback.
-
|
Can we use LlamaCpp models to create an agent? |
Beta Was this translation helpful? Give feedback.
-
|
Hey I have a question about state. So we have messages and sender which is fine. Although if we have so many iterations and messages are stacking. So let's say after 14 iterations we have a messages that contains 14 elements. This is a dangerous scenerio in 2 cases.
Does state take care of it or we do have to control it in some sense ? |
Beta Was this translation helpful? Give feedback.
-
|
Everything except the AI message content is displayed. |
Beta Was this translation helpful? Give feedback.
-
|
In LangGraph, the string &npsb (which means space) is attached to the edge string, making it difficult to read. |
Beta Was this translation helpful? Give feedback.
-
|
If you run the following program in jupyterlab, the graph will be displayed nicely. from IPython.display import Image, display If you convert it to a python script, from PIL import Image as PilImage When you do this, the string   is attached to the edge, making it difficult to see. |
Beta Was this translation helpful? Give feedback.
-
|
I removed the   string in graph_mermaid.py. （Solved.） .venv/lib/python3.12/site-packages/langchain_core/runnables/graph_mermaid.py edge_data = "  ↓ edge_data = " |
Beta Was this translation helpful? Give feedback.
-
File "/home/flyang/anaconda3/envs/LLMs/lib/python3.11/site-packages/openai/_base_client.py", line 1040, in _request |
Beta Was this translation helpful? Give feedback.
-
|
Can I run this with Gemini 1.5 flash or Ollama? |
Beta Was this translation helpful? Give feedback.
-
|
yupp |
Beta Was this translation helpful? Give feedback.
-
|
Yes of course ! |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the tutorial; however, why is there an additional incomplete image here? |
Beta Was this translation helpful? Give feedback.
-
|
Arent the edges from researcher to charter missing? |
Beta Was this translation helpful? Give feedback.
and others
-
langgraph/tutorials/multi_agent/multi-agent-collaboration/
Build language agents as graphs
https://langchain-ai.github.io/langgraph/tutorials/multi_agent/multi-agent-collaboration/
Beta Was this translation helpful? Give feedback.
All reactions