llama.cpp

6dbc1174 - ggml-cpu: add AVX2 optimization for nvfp4 dot product and use UE4M3 LUT (#23961)