Swiss AI Model Launch¶

Make it easy to launch models 🚀

A CLI for launching LLMs on HPC clusters via SLURM or FirecREST. Public serving endpoint: https://serving.swissai.svc.cscs.ch/.

Quickstart¶

pip install git+https://github.com/swiss-ai/model-launch.git
sml init
sml

That’s it — the second command sml init sets up credentials, the third launches a model interactively.

Launching Apertus-8B with sml

Where to start¶

New here? → Getting Started
Setting up credentials? → Initialization
Just want a script to run? → browse examples/ on GitHub
Sizing questions? → How to size a model
Hooking up Claude Desktop? → MCP Server
Always-on hosting / general questions? → FAQ

What SML is and isn’t¶

SML is a thin orchestrator that submits SLURM jobs to bring up sglang or vLLM with the right model and arguments. It hands you back a live TUI of logs and job state until the model is healthy.

It is not a model server itself, not a long-running deployment manager (use Kubernetes for always-on serving — see FAQ), and not a public traffic gateway (that’s serving-api).

See Architecture for how the pieces fit together.