llama.cpp

17c77e1e - Allow remaining FP8 tensors to be dequantized during conversion