shannonai / kfolden-ood-detection Goto Github PK

Dear Authors,
Could it be possible to have access to the baseline models on which you perform Mahlanobist and MSP ?
Kindest regards,
Pierre

google drive permission denied

Thank you for providing the valuable work and benchmark!
I have a question for reproducibility:

• When I re-run the implementation of vanilla BERT followed your script, BERT trained with AGNews-FL achieved 91.25 (ID Acc.). However, Table 3 in the paper, the ID classification accuracy for the vanilla BERT is reported by 85.58. I didn't modify the code, and I ran the script as it is. Which result is correct? Or is the dataset you uploaded to Google Drive different from when you wrote the paper?

Best,