Pairwise preference conversion of openai/coval.
Rows are generated from CoVal comparison rankings where one candidate response is
strictly preferred over another candidate at least 2 rank groups away.
Adjacent preferences and ties are excluded. For example, A>B>C=D contributes
A over C and A over D, but excludes A over B, B over C, and the
tie between C and D.
prompt: chat messages from the original prompt.
chosen: assistant response messages… See the full description on the dataset page:
https://huggingface.co/datasets/sumuks/coval-ha.
