llama.cpp

f8f071fa - convert : handle pre-quantized models (#14810)

convert : handle pre-quantized models (#14810) * convert : begin handling pre-quantized models * convert : fix conversion from FP8 for Deepseek-V3.1-Base