llama.cpp

convert : force patch_embd weights to F16 or F32 to avoid broken GGUFs

#15367

Merged