Open nancycsecon opened 3 weeks ago
Right now this isn't supported in the OAI client but it looks like it might be supported for Anthropic and Mistral. You would need to modify this: https://github.com/microsoft/autogen/blob/08fa1b6d080a99daccaebd0390fc3c7cd91f05e6/autogen/oai/client.py#L294
So that it adds a "tool_choice" parameter.
Hey @nancycsecon, in relation to @spullara's point, can you let us know what provider/model you are using?
If you had an example that may help, it could possibly also help to do further prompt tweaking.
Can you add a new hook to process_last_received_message
, or process_all_messages_before_reply
?
Or register a new reply function that calls the database and then returns the result of generate_oai_reply
?
It seems like there's little point trying to keep it as a tool at this point, and the above would be more appropriate for your use case
Describe the issue
I registered a tool function to an agent which searches from a database. Now I want the tool function be called every time the agent speaks. Now the agent sometimes bypass calling the tool and would hallucinate answers. Is there any workaround to fix this?
Steps to reproduce
No response
Screenshots and logs
No response
Additional Information
No response