Every request runs a real 184M-parameter DeBERTa-v3 encoder with eight
classification heads — the inference time below is measured, not canned. It
classifies rather than generating text, so there is no sampling and no
temperature: the same prompt always returns identical scores.
nvidia/prompt-task-and-complexity-classifier
0 characters
Try:
Prompt complexity score
— / 1.0
Task type
—
Contributing dimensions — each 0 to 1, weighted into the score above