How-toUpdated 2026-07-01 · 9 min read · by RTXsparks Lab
Serving Llama 3.2 90B Vision with vLLM on RTX Spark
Production vLLM deployment of Llama 3.2 90B Vision on RTX Spark. Configs, systemd, TLS, observability.
Install
Install vLLM via pip/docker/apt as appropriate. Verify GPU visibility with nvidia-smi.
- # vLLM install
- pip install vllm
Config
Optimized vLLM config for Llama 3.2 90B Vision at Q4_K_S: batch, KV cache dtype, and parallelism knobs.
Systemd unit
[Unit] Description=vLLM for Llama 3.2 90B Vision — [Service] ExecStart=/usr/bin/vllm serve llama-3-2-90b-vision --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 vLLM port.
Observability
Scrape vLLM Prometheus metrics into Grafana; watch tokens_per_second, batch_size, and kv_cache_usage.
Frequently asked questions
Is vLLM the fastest for Llama 3.2 90B Vision?
Yes for throughput; TensorRT-LLM edges it on p50 latency.
Can I run multiple models with one instance?
Yes with --served-model-name and multi-LoRA.