llama.cpp

Add --continue-after-failure support to llama-bench for resilient benchmark sweeps

#22854

Open