Run your first model

Follow Installation, then choose a model that fits your available memory.

Open the Web UI

snowllm webui

Opens http://127.0.0.1:8001. For headless use, add --no-browser.

Download and load a recipe, then open Chat. The inference API stays at http://127.0.0.1:8000.

Model storage

Downloads go to ~/models by default. Set the model directory in Settings before downloading if you want to use a different drive.

List the available recipes:

snowllm recipes

For example, the Qwen3.6 35B FP8 recipe uses about 34.9 GiB for resident weights, plus workspaces and caches:

snowllm pull qwen3.6-35b-a3b-fp8
snowllm qwen3.6-35b-a3b-fp8

Choose a smaller supported model if needed.

Stop direct serving with Ctrl+C before reusing the port.

With snowllm webui running:

snowllm load qwen3.6-35b-a3b-fp8 --max-model-len 32k
snowllm status
snowllm unload

Wait for snowllm status to report ready. See Model lifecycle.

See Integrations for API addresses, requests and Pi setup.