v0.2.0b1: testing with the simplest code

[jonny7737]

The simplest code:

Define:

• config_list
• llm_config

Instantiate:

• assistant agent => AssistantAgent
• user agent => UserProxyAgent

Call:

• initiate_chat

Testing several prompts on several LLMs locally (no openai)

• autogen caching is a problem
• How to disable caching?

[sonichi]

add "seed": None in the llm_config
https://microsoft.github.io/autogen/docs/Installation#migration-guide-to-v02

[jonny7737]

Thanks.

The doc was unclear to me that it was referring to llm_config.

[jonny7737]

.cache/41 is still created and cache is still active.

This:
llm_config = { "model": "aaa", "temperature": 0.1, "timeout": 180, "config_list": config_list, "seed": None, }
Does not disable autogen caching.

My verification is to delete .cache, run my app, the chat is normal but .cache/41 is created. immediately run my app again and the chat runs hair on fire to the end.

[sonichi]

That's weird. When I use assistant = AssistantAgent("assistant", llm_config={"config_list": config_list, "seed": None}) the cache is disabled. Could you verify the pyautogen version and share the code to reproduce it?

[jonny7737]

I'll see if I can craft a minimal piece of code that will reproduce the issue.
FYI - Not seeing any other issues with v0.2.0b1

Feel free to close this ticket if you like. I can create a new one with the reproducing code when I have it ready.

[sonichi]

Thank for testing and sharing the results! There is one bug in v0.2.0b1 which is fixed in #551 . The caching problem seems unrelated, but maybe worth trying.

[jonny7737]

The issue was caused by me not autogen.

Thanks for the help and sorry for the trouble.