llama.cpp

ggml : speed up model loading

#29598

Merged