obs-llamacpp-server-readme-metrics

The /metrics and /slots documentation. --metrics enable prometheus compatible metrics endpoint (default: disabled); --slots expose slots monitoring endpoint (default: enabled). The /slots JSON per-slot timings object is the only place per-request decode timing exists. Confirms there is no OTel: an in-repo search for 'opentelemetry' in ggml-org/llama.cpp returns 0 code hits.