fix(ar_eg): drop corrupt audio clip id=1689 (test)
#2
by AdnanElAssadi - opened
FleursT2ARetrieval corpus row ar_eg/test id=1689 (idx 274) has a corrupt audio frame: torchcodec raises 'Could not receive frame from decoder: Invalid data found when processing input', which crashes audio-side encoding for any model whose pipeline force-decodes the array. Verified the other 77,809 clips across all 102 subsets decode cleanly. Query<->doc is self-mapped by id, so dropping this row removes it from corpus, queries, and qrels with no orphans (ar_eg test 428->427).
Yeah, I recived this sometimes also on other datasets
Samoed changed pull request status to merged