anikifoss

/

DeepSeek-V3.1-Terminus-HQ4_K

like 2

Currently llama.cpp does not return <think> token in response. If you know how to fix that, please share in the "Community" section!

I was researching the reasoning formatting issues of Kimi-K2-Thinking and found this llama.cpp param:

-sp,   --special                        special tokens output enabled (default: false)

@unsloth mentions it here:
https://docs.unsloth.ai/models/kimi-k2-thinking-how-to-run-locally

I did not have a chance to test it yet, but I am quite positive it might help :).