Thank you for proposing an interesting token-level data selection method. I encountered some issues while preparing the TyDiQA dataset for evaluation (For research purposes only):
- Dataset Download Issue: I'm unable to download the TyDiQA dataset from the following links (see screenshot attached):
Could you provide a new download link or share the dataset with me? My email: daishaojie96@gmail.com
- Metrics Clarification: Regarding Table 1 in the paper, for the results on HellaSwag, LogiQA, and ARC_challenge, are the reported metrics accuracy (acc) or accuracy normalized (acc_norm)?
Looking forward to your reply. Thank you!
Thank you for proposing an interesting token-level data selection method. I encountered some issues while preparing the TyDiQA dataset for evaluation (For research purposes only):
Could you provide a new download link or share the dataset with me? My email: daishaojie96@gmail.com
Looking forward to your reply. Thank you!