search

microsoft / autogen

A programming framework for agentic AI. Discord: https://aka.ms/autogen-dc. Roadmap: https://aka.ms/autogen-roadmap
https://microsoft.github.io/autogen/
Creative Commons Attribution 4.0 International
28.72k stars 4.19k forks source link

[Issue]: GroupChat keeps calling all the available agent and doesn't stop even after output from one agent is generated #1464

Open saurabhgssingh opened 5 months ago

saurabhgssingh commented 5 months ago

Describe the issue

When I have multiple agents configured in a group chat. If the question is very specific to one agent [say text to sql], the group chat manager invokes other agents as well even though they are not required.

image

Steps to reproduce

groupchat = GroupChat( agents=[user_proxy,coding_assistant,simulator_worker_agent,text_2_sql_agent], messages=[], max_round=24 ) manager = GroupChatManager(groupchat=groupchat, llm_config=llm_config)

Screenshots and logs

No response

Additional Information

AutoGen Version: 0.2.9 Python 3.11

ekzhu commented 5 months ago

Can you show your llm_config for the group chat manager? Which model are you using?

One thing we are working on is graph-based group chat, which allows user to add constraints to the speaker transition: #857

saurabhgssingh commented 5 months ago

[{ "model" : "minerva-llm", "api_key": "key", "base_url": "https://endpoint.openai.azure.com/", "api_type": "azure", "api_version": "2023-07-01-preview" }]

groupchat = GroupChat( agents=[user_proxy,fetch_schema_agent,simulator_agent], messages=[], max_round=24 ) manager = GroupChatManager(groupchat=groupchat, llm_config=llm_config)

We are using azure gpt4 turbo model

sonichi commented 5 months ago

How are the text2sql and rag agent defined? Is the "description" field used to describe their roles?