Skip to content

First login

Sign in

Open one of the URLs the installer printed (http://<host>:8090/ redirects to /dgxmm/), sign in as admin with the printed password (also saved in secrets/admin_password) and choose a new one of at least 12 characters. Every other action is refused until you have changed it.

Delete the saved first password

secrets/admin_password is not needed after the first password change. Delete it (see security hardening).

Forgot the password later? See troubleshooting.

Deploy your first model

Deploy a model from Deploy model, or find one in Catalog first and download it. The wizard shows a memory check before anything starts (see deploying a model). After Deploy the deployment moves through Starting → Loading → Ready; the first vLLM start of a model can take several minutes.

Call your first model

Every ready deployment is one model on the LiteLLM endpoint, selected by its model name (shown on the deployment's API access card). Create a virtual key on the LiteLLM page (admins) or on the deployment's LiteLLM tab, then:

export LITELLM_API_KEY=sk-...                   # the virtual key, shown once when it is created
curl http://<host>:4000/v1/models -H "Authorization: Bearer $LITELLM_API_KEY"
curl http://<host>:4000/v1/chat/completions \
  -H "Authorization: Bearer $LITELLM_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"model": "default-chat", "messages": [{"role": "user", "content": "Hello!"}]}'

http://<host>:4000 is the bundled LiteLLM; with your own LiteLLM, use its public URL. Any OpenAI SDK works the same way (base_url="http://<host>:4000/v1"). Python snippets and direct engine endpoints are in calling your models.

Next steps