Local LLM stack

A self-hosted vLLM, LiteLLM, Open WebUI, monitoring, and fine-tuning lab stack for an NVIDIA Spark host.

Status: active · GitHub


What it is

A reproducible local assistant stack for running OpenAI-compatible model endpoints on a private NVIDIA Spark host. It wires vLLM backends through LiteLLM, exposes Open WebUI, and includes Prometheus/Grafana monitoring plus a small LoRA/QLoRA-oriented fine-tuning lab.

Why it matters

The project turns local inference into an inspectable systems surface: model aliases, routing, health checks, request tracing, load tests, throughput evals, vision-serving experiments, and deployment runbooks live beside the compose stack.

GitHub