summary_method="reflection_with_llm" is not working

[appydam]

What happened?

this is not working, it is supposed to store summary of llms but its simply not working. I also see one more open issue of 3 months old, and its still not resolved.
Please fix this now, if this can't work, then autogen is useless totally.

What did you expect to happen?

Please fix this bug

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

Ok, here is the code snippet

llm_config = {
    "model": "mlx-community/Llama-3.2-3B-Instruct-4bit",
    "base_url": "http://localhost:1234/v1",
    "api_key": "lm-studio"
}

from autogen import ConversableAgent

onboarding_personal_information_agent = ConversableAgent(
    name="Onboarding_Personal_Information_Agent",
    system_message='''You are a helpful customer onboarding agent,
    you work for a phone provider called ACME.
    Your job is to gather the customer's name and location.
    Do not ask for any other information, only ask about the customer's name and location.
    After the customer gives you their name and location, repeat them 
    and thank the user, and ask the user to answer with TERMINATE to move on to describing their issue.
    ''',
    llm_config=llm_config,
    human_input_mode="NEVER",
    is_termination_msg=lambda msg: "terminate" in msg.get("content").lower()
)


onboarding_issue_agent = ConversableAgent(
    name="Onboarding_Issue_Agent",
    system_message='''You are a helpful customer onboarding agent,
    you work for a phone provider called ACME,
    you are here to help new customers get started with our product.
    Your job is to gather the product the customer use and the issue they currently 
    have with the product,
    Do not ask for other information.
    After the customer describes their issue, repeat it and add
    "Please answer with 'TERMINATE' if I have correctly understood your issue." ''',
    llm_config=llm_config,
    human_input_mode="NEVER",
    is_termination_msg=lambda msg: "terminate" in msg.get("content").lower()
)


customer_engagement_agent = ConversableAgent(
    name="Customer_Engagement_Agent",
    system_message='''You are a helpful customer service agent.
    Your job is to gather customer's preferences on news topics.
    You are here to provide fun and useful information to the customer based on the user's
    personal information and topic preferences.
    This could include fun facts, jokes, or interesting stories.
    Make sure to make it engaging and fun!
    Return 'TERMINATE' when you are done.''',
    llm_config=llm_config,
    human_input_mode="NEVER",
    is_termination_msg=lambda msg: "terminate" in msg.get("content").lower(),
)

customer_proxy_agent = ConversableAgent(
    name="customer_proxy_agent",
    llm_config=False,
    code_execution_config=False,
    human_input_mode="ALWAYS",
)

chats = [] # This is going to be our list of chats


chats.append(
    {
        "sender": onboarding_personal_information_agent,
        "recipient": customer_proxy_agent,
        "message": 
            "Hello, I'm here to help you solve any issue you have with our products. " 
            "Could you tell me your name?",
        "summary_method": "reflection_with_llm",
        "summary_args": {
        "summary_prompt" : "Return the customer information "
                             "into a JSON object only: "
                             "{'name': '', 'location': ''}",
        },
        "clear_history" : True
    }
)


chats.append(
    {
        "sender": onboarding_issue_agent,
        "recipient": customer_proxy_agent,
        "message": 
                "Great! Could you tell me what issue you're "
                "currently having and with which product?",
        "summary_method": "reflection_with_llm",
        "clear_history" : False
    }
)



chats.append(
        {
        "sender": customer_proxy_agent,
        "recipient": customer_engagement_agent,
        "message": "While we're waiting for a human agent to take over and help you solve "
        "your issue, can you tell me more about how you use our products or some "
        "topics interesting for you?",
        "max_turns": 2,
        "summary_method": "reflection_with_llm",
    }
)
chats

from autogen import initiate_chats

chat_results = initiate_chats(chats)

import pprint

for chat_result in chat_results:
    #pprint.pprint(chat_result.chat_history) # We could also get the whole chat history with this command
    pprint.pprint(chat_result.summary)

AutoGen version

0.6.1

Which package was this bug in

Core

Model used

llama-3.2-3b-instruct

Python version

No response

Operating system

mac

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

No response

[ekzhu]

@appydam could you please show error trace or output and explain why it is not working for you?

this is not working, it is supposed to store summary of llms but its simply not working

Would be great to explain how.