Select one or more online hosts. The model list and the job stay on that selection.
Leave this blank. The daemon caps the reply. A number here cannot exceed 25 tokens per second for the job duration. The model's context window is shown after the job runs.