llama.cpp

Use model->gguf_kv for loading the template instead of using the C API.

#10868

Merged