Preference
UltraFeedback
Released model completions with GPT-4 ratings for instruction following, honesty, truthfulness and helpfulness.
254,495items
743subjects
MIT, as declared in the pinned upstream dataset card. Original attribution and card are retained.license
preferencedomain
textmodality
item-level responses released
Saturation status: Yes
Response matrix
Loading response matrix…
lowhighUnobserved
Scale: Grades: 1, 2, 3, 4, 5