autogen icon indicating copy to clipboard operation
autogen copied to clipboard

[Issue]: It just terminates the session

Open nothingness6 opened this issue 2 years ago • 6 comments

Describe the issue

Whatever I type, it just terminates the session.

Steps to reproduce

  1. Changed to a different model, Mistral Instruct v0.1-> Mistral Instruct v0.2.
  2. Changed to a different model, Mistral Instruct v0.2 -> Mistral v0.1

Screenshots and logs

autogen

Modified files:  0
Request reply None sender_type agent silent False sender userproxy
[33muserproxy[0m (to LMstudio_assistant):

Active Connections:  ['3e4e4342-b114-4435-9722-8a649bf193a1']
Hello
Current message connection id:  3e4e4342-b114-4435-9722-8a649bf193a1

--------------------------------------------------------------------------------
Request reply None sender_type agent silent False sender LMstudio_assistant
[33mLMstudio_assistant[0m (to userproxy):
Active Connections:  ['3e4e4342-b114-4435-9722-8a649bf193a1']

Current message connection id:  3e4e4342-b114-4435-9722-8a649bf193a1
# filename

--------------------------------------------------------------------------------
Request reply None sender_type agent silent False sender userproxy
[33muserproxy[0m (to LMstudio_assistant):

TERMINATE

Active Connections:  ['3e4e4342-b114-4435-9722-8a649bf193a1']
--------------------------------------------------------------------------------
Current message connection id:  3e4e4342-b114-4435-9722-8a649bf193a1
Modified files:  0
Request reply None sender_type agent silent False sender userproxy
[33muserproxy[0m (to LMstudio_assistant):

Active Connections:  ['3e4e4342-b114-4435-9722-8a649bf193a1']
hello
Current message connection id:  3e4e4342-b114-4435-9722-8a649bf193a1

--------------------------------------------------------------------------------
Request reply None sender_type agent silent False sender LMstudio_assistant
[33mLMstudio_assistant[0m (to userproxy):
Active Connections:  ['3e4e4342-b114-4435-9722-8a649bf193a1']

Current message connection id:  3e4e4342-b114-4435-9722-8a649bf193a1
TERMIN

--------------------------------------------------------------------------------
Request reply None sender_type agent silent False sender userproxy
[33muserproxy[0m (to LMstudio_assistant):

TERMINATE
Active Connections:  ['3e4e4342-b114-4435-9722-8a649bf193a1']

--------------------------------------------------------------------------------
Current message connection id:  3e4e4342-b114-4435-9722-8a649bf193a1
Modified files:  0

Additional Information

AutoGen Version: v0.0.56 Operating System: Windows 10 64 bit Python Version: 3.10

nothingness6 avatar Apr 01 '24 10:04 nothingness6

Have you tried to update the system messages of the agent? Different models should have different system messages. The default one is tailored to OpenAI's models.

ekzhu avatar Apr 02 '24 07:04 ekzhu

Have you tried to update the system messages of the agent? Different models should have different system messages. The default one is tailored to OpenAI's models.

Great! I figured it out. Thanks for your help! Btw, it shows the response with a single or a couple words on the line instead of a full sentence. Plus, why does it show the history instead of showing just a response? Please take a look at this: autogen autogen-02

nothingness6 avatar Apr 02 '24 10:04 nothingness6

It is showing a chat history because under-the-hood a message from the human user triggers a two-agent chat workflow, and the summary of the two-agent chat conversation is the final response to the human user.

ekzhu avatar Apr 10 '24 07:04 ekzhu

same error here, i use lm studio, and i always do like this getting stuck in clicles saying random things, ho an i fix, i already changed the system message of the agent, ca u give me urs if it works?

Giova200 avatar Apr 16 '24 12:04 Giova200

@jackgerrits @victordibia This issue involves LMStudio and AGS.

sonichi avatar Apr 16 '24 14:04 sonichi

@Giova200 , thanks for mentioning this. I think this is related to #2050. Essentially, some models/endpoints require max_tokens to be explicitly which seems to be the case with lmstudio. I am working on exposing that parameter correctly in the Agent UI. In the mean time, consider the work around in the #2050 (setting the default max tokens directly in code)

victordibia avatar Apr 16 '24 21:04 victordibia