How-toUpdated 2026-07-01 · 9 min read · by RTXsparks Lab
Serving Qwen2.5-VL 72B with llama.cpp on RTX Spark
Production llama.cpp deployment of Qwen2.5-VL 72B on RTX Spark. Configs, systemd, TLS, observability.
Install
Install llama.cpp via pip/docker/apt as appropriate. Verify GPU visibility with nvidia-smi.
- # llama.cpp install
- brew install llama.cpp
Config
Optimized llama.cpp config for Qwen2.5-VL 72B at Q4_K_M: batch, KV cache dtype, and parallelism knobs.
Systemd unit
[Unit] Description=llama.cpp for Qwen2.5-VL 72B — [Service] ExecStart=/usr/bin/llama-cpp serve qwen2-5-vl-72b --port 8000 — Restart=always.
nginx + TLS
Terminate TLS at nginx with a Let's Encrypt cert, rate-limit at 20 req/s per IP, and proxy to the local llama.cpp port.
Observability
Scrape llama.cpp Prometheus metrics into Grafana; watch tokens_per_second, batch_size, and kv_cache_usage.
Frequently asked questions
Is llama.cpp the fastest for Qwen2.5-VL 72B?
It's the easiest, not always the fastest.
Can I run multiple models with one instance?
One process per model is standard.