Bad Request Error - Groupchat with DBRX Foundation Models

[Aljgutier]

What happened?

Python Code
This code is exactly taken from the following example

https://microsoft.github.io/autogen/0.2/docs/notebooks/agentchat_groupchat/

import autogen
user_proxy = autogen.UserProxyAgent(
    name="User_proxy",
    system_message="A human admin.",
    code_execution_config={
        "last_n_messages": 2,
        "work_dir": "groupchat",
        "use_docker": False,
    },  # Please set use_docker=True if docker is available to run the generated code. Using docker is safer than running the generated code directly.
    human_input_mode="TERMINATE",
)
coder = autogen.AssistantAgent(
    name="Coder",
    llm_config=llm_config,
)
pm = autogen.AssistantAgent(
    name="Product_manager",
    system_message="Creative in software product ideas.",
    llm_config=llm_config,
)
groupchat = autogen.GroupChat(agents=[user_proxy, coder, pm], messages=[], max_round=12)
manager = autogen.GroupChatManager(groupchat=groupchat, llm_config=llm_config)

**To Reproduce

user_proxy.initiate_chat(
    manager, message="Find a latest paper about gpt-4 on arxiv and find its potential applications in software."
)
# type exit to terminate the chat

Expected behavior
The agents should write code and lookup articles in arxiv

Note that his code works successfully when using OPEN AI models such as GPT 4o

However, when running with DBRX databricks-meta-llama-3-70b-instruct then a BadRequestError is generated

BadRequestError: Error code: 400 - {'error_code': 'BAD_REQUEST', 'message': "Bad request: rpc error: code = InvalidArgument desc = Chat message input must end with a 'user', 'assistant', or 'tool' role\n"}

the DBRX foundation models work fine in most autogen examples but not for GroupChat

Which packages was the bug in?

V0.2 (autogen-agetnchat==0.2.*)

AutoGen library version.

Python dev (main branch)

Other library version.

No response

Model used

databricks-meta-llama-3-70b-instruct

Model provider

None

Other model provider

No response

Python version

3.10

.NET version

None

Operating system

None

[Aljgutier]

Thank you for your attention on this issue. Please let me know if additional info is needed or how I can be helpful

[ekzhu]

@Aljgutier looks like a model compatibility issue. Could you try this with latest version of selector group chat?

We don’t officially support all open ai compatible model clients — we don’t have access to all model providers. So if you can debug this and find a fix, we may be able to patch it with your PR.

[Aljgutier]

Thank you, @ekzhu, for the follow-up. Indeed, the issue is a model compatibility issue.

I am willing to work on a fix but only within the 0.2 context and with DBRX models. From the 0.4 autogen_ext its not clear that DBRX is supported

• interestingly, two_agent chat continues to work okay with the DBRX foundation models
• 0.4 Autogen is not an option

I will need to pursue other avenues.

thanks

[ekzhu]

Why is v0.4 not an option?

Looks like DBRX is open ai compatible to some extent. I think once we address #5965 it will be supported.