[Feature Request] Enable Function Calling Support in Reasoning Models (Deepseek R1, OpenAI O1, etc) for Agentic Tasks
Required prerequisites
- [x] I have searched the Issue Tracker and Discussions that this hasn't already been reported. (+1 or comment there if it has.)
- [x] Consider asking first in a Discussion.
Motivation
I would like to propose an enhancement to our current reasoning models (e.g., r1, o1, etc.). Currently, these models do not support function calling, which limits their application in agentic tasks that require dynamic interaction with external functions.
Integrating a framework to enable function calling could allow the models to trigger and handle external functions as part of their reasoning process. This enhancement would broaden the range of tasks our agents can perform and improve their overall autonomy.
Could we explore the possibility of integrating an existing function calling framework or developing an abstraction layer that provides this capability?
Some references:
- https://github.com/deepseek-ai/DeepSeek-R1/issues/9
- https://www.boundaryml.com/blog/deepseek-r1-function-calling
- https://github.com/BoundaryML/baml
- https://github.com/justinlietz93/agent_tools
Solution
No response
Alternatives
No response
Additional context
No response
thanks for raising this issue @Aaron617 , for now we can use o3 mini do tool call task, for other reasoning models we can also explore adding support for this, but need sometime
I am very interested in this issues.