llama.cpp

0710d5f0 - ggml : add GGML_SCHED_NO_REALLOC option to disable reallocations in ggml_backend_sched