RAG-agent max_consecutive_auto_reply doesn't work as expected

[thinkall]

I am using RAG-agent and it works pretty ok, but it gives me ~10 answers instead of just one or two, and it does not explicitly write anything like "I do not have enough information". It just spits out proper answers, only many times. I set the setting max_consecutive_auto_reply=1 but this does not change anything.

Find my code here: https://github.com/wiiiktor/resume/tree/main/data/autogen-bug There is a python file with a very small dataset. In the meantine, I updated tokenizers (PyCharm told me so) and error is different now, but still a similar tune: RuntimeError: Failed to import transformers.models.nystromformer because of the following error (look up to see its traceback):
maximum recursion depth exceeded while calling a Python object
name = "tokenizers"
version = "0.14.0" -> this is what I have now

-- origin