$ llamacpp-1650ti start
llamacpp-1650ti: starting /home/xbill/models/gemma-4-E2B-it-qat-q4_0/gemma-4-E2B_q4_0-it.gguf on 127.0.0.1:8080 (-ngl 99)
llamacpp-1650ti: pid 88166, log /home/xbill/gemma4-dev/local-llamacpp-1650ti-2b-q4_0/run/llama-server.log
llamacpp-1650ti: waiting up to 180s for http://127.0.0.1:8080/health
     0s  VRAM 3 MiB, 0 %
llamacpp-1650ti: healthy after 2s -- http://127.0.0.1:8080
llamacpp-1650ti: device=gpu · pid=88166 · -ngl 99 · mapped: ggml-cuda, libcublas, libcuda, libcudart · /home/xbill/llama.cpp/build/bin/llama-server

$ llamacpp-1650ti stop
llamacpp-1650ti: sent SIGTERM to pid 88166; VRAM is released on exit