Hi.
Why you guys did not considered for deepseek-ai/deepseek-r1 the reasoning extraction only, as they do in the official model from deepseek.com?
something like:
reasoning_content = response.choices[0].message.reasoning_content
For the reasoning part
in NVIDIA NIM module is only possible to extract the content:
content = response.choices[0].message.content
Where the reasoning is inside surrounded by the “<think>” tag.
Any plans on this?
Thank you