Please add function calling support
Currently running the 0528-UD-Q2-K-XL version on llama.cpp, attempting to call function calling results in an error. The command llama-server -m ... --jinja --chat-template-file /XXX/llama-cpp-deepseek-r1.jinja returns incorrect function calling responses. How should the correct Jinja file be configured?
There seems to be some issues with the model where tool calling doesn't work as intended we are investigating and will let the Deepseek team know as well
Could you please explain how to properly run the deepseek r1 tool calling functionality based on llama.cpp? My commands always fail to execute successfully, as shown below:
llama.cpp/llama-server -m /models/gguf/UD-IQ2_XXS/DeepSeek-R1-0528-UD-IQ2_XXS-00001-of-00005.gguf --n-gpu-layers 99 --host 0.0.0.0 --port 9996 --parallel 2 --ctx-size 16384 --cache-type-k q4_0 --threads 24 -a DeepSeek-R1-0528-UD-IQ2_XXS --jinja --temp 0.6 --top_p 0.95 --min_p 0.01 --seed 3407