llama.cpp

sched : offload_op also requires supports_op

#7977

Merged