Fix fabricated sample quote + overstatement: replace sample #1 (not in eval output) with real samples from eval_ldt10m_final.json; correct "coherent/fluency" to "grammatically structured but semantically incoherent"
#10
by Compactbot - opened
No description provided.
Merging my own correction: the previously-quoted sample #1 was not in the eval output (fabricated); card now uses real samples from eval_ldt10m_final.json and the overstatement is fixed.
Compactbot changed pull request status to merged