JayZeeDesign / microsoft-autogen-experiments

207 stars 72 forks source link

Endless loop content_agent #3

Open MikaMcFly opened 11 months ago

MikaMcFly commented 11 months ago

Hi, first thanks a lot for sharing the code base. I run the code myself and it kind of works. But there a two main issues:

1) The scraped content is not saved to a file + the content itself nether. Would it be possible to add a vector store or at least a saving to directory function? Also the content generated, is also not saved as a text file. When prompting to do so, it fails in execution. The might be a coder agent necessary to write a litte script to transform the out into a .txt file.

2) When giving the final enter to close the project it starts a never ending loop of saying thank you and asking for additional help. I have to kill the terminal to end this. Is there an easier way to end the conversation?

Culpable commented 6 months ago

Issue 2

I don't seem to be having the issue you described in (2), although that may be because I adjusted the parent UserProxyAgent to human_input_mode="ALWAYS".

Issue 1

Did you manage to resolve (1)?

I'd love to be able to save the output into a file.

I notice in the the primary_assistant in AutoGen Studio, they use the line:

If you want the user to save the code in a file before executing it, put # filename: inside the code block as the first line.

I'm not sure if there is any logic within AutoGen to pick up on said filename, but I notice within the UI it will display the image and code produced within the chat.

I'd love to be able to better understand how this works so I can use it with my own Autogen group chat.

Other Resources

I'm having some difficulty understanding exactly how Autogen works, not from a high-level perspective. I've read the paper and understand how all that works, but some functionality such as:

  1. How the group chat manager selects the correct agent to use.
  2. Whether each intermediary agent is being passed the entire chat history or just the last message.
  3. How things such as file names are being displayed.

This is a bit confusing to me.

If anyone can point me towards a tutorial that delves deeply into this, specifically in terms of using the library, I would be extremely thankful. I've looked through some of the getting-started guides on YouTube, and although they cover it at a surface level, I really can't find an in-depth explanation of how everything in Autogen ties together.