Controls diversity (0.1 = focused, 1.0 = all options)
Maximum response length (1000-10000)
Maximum wait time for LLM response
Maximum wait time for Canvas API calls
Maximum wait time for file downloads
How AnZar asks a model to stop thinking. Auto recognizes known model families and sends their parameter; unknown models are called without any extra field.
Advanced Options
Used only when Thinking Models is set to "Custom". A JSON object whose fields are merged into every grading request, for endpoints with their own thinking parameter, e.g. {"novathink": {"level": "off"}}. Requests that the server rejects for naming one of these fields are retried once without them.
Controls how the LLM behaves when grading. This is sent as the system message to the LLM before each grading request.