How do you use prompt walk-around function calling with this SDK to improve Agent performance on o3-mini?

[jackien1]

o3-mini is really bad at function calling compared with gpt-4o but using prompt function calling makes it much better.

https://gorilla.cs.berkeley.edu/leaderboard.html

[rm-openai]

Would be open to adding something - any thoughts on a good interface?

[jackien1]

Not sure but anything simple but would be great (I'm dying from gpt-4o costs).

[rm-openai]

I'm actually not exactly sure what you mean by "prompt walk-around function calling" so a pointer/example would be helpful!

[jackien1]

I'm actually not exactly sure what you mean by "prompt walk-around function calling" so a pointer/example would be helpful!

They talk about it here with the code:
https://gorilla.cs.berkeley.edu/blogs/8_berkeley_function_calling_leaderboard.html

Also not super familiar with the inner-workings of the library but been reading up on using prompts rather than straight function-calling mode to improve performance—can also support more models with native FC support.

[jackien1]

Nevermind gpt-4.1 is goated!