MagenticOneGroupChat always reset the participants for each task

[iaghidro]

What happened?

With MagenticOneCoderAgent the state doesn't load in any way (tried with and without saving it to a json file). I did the same thing with a round robin group chat and it worked as expected. I confirmed in discord that this should be working.

What did you expect to happen?

I expected the state to load as described in the docs which works for the round robin group chat.

How can we reproduce it (as minimally and precisely as possible)?

I'm sharing 3 files:

1. the python code
   code.txt
2. the team state .json file
   team_state.json
3. the console logs for the run
   logs.txt

This should be enough to reproduce quickly and easily.

AutoGen version

0.4.1

Which package was this bug in

AgentChat

Model used

gpt-4o

Python version

3.12

Operating system

mac

Any additional info you think would be helpful for fixing this bug

This should not affect anything, but I run a conda env and have followed the docs for installation.

[ekzhu]

I think the MagenticOneGroupChat state looks fine. What you saw should be due to the implementation of MagenticOneGroupChat, it always start a new task with reset of its memory and the agents' memory. cc @husseinmozannar

[iaghidro]

Is that intentional to reset the state each time? If so how do I work around it? I thought saving and reloading the state is supposed to work for cases like this

[ekzhu]

Is that intentional to reset the state each time? If so how do I work around it? I thought saving and reloading the state is supposed to work for cases like this

Well, it is by design for the magentic one team. Not sure if skipping resetting makes sense for this team. @afourney can better answer this.