Training on SQUAD2 gives less results on Evaluation set (Research Paper shows better results)

[bhadreshpsavani]

I tried Training MPNet on SQUAD2 data below is the result I was getting on Evalutionset

I used this script

(exact, 50.07159100480081)                                                                                  
(f1, 50.07159100480081)
(total, 11873)
(HasAns_exact, 0.0)
(HasAns_f1, 0.0)
(HasAns_total, 5928)
(NoAns_exact, 100.0)
(NoAns_f1, 100.0)
(NoAns_total, 5945)
(best_exact, 50.07159100480081)
(best_exact_thresh, 0.0)
(best_f1, 50.07159100480081)
(best_f1_thresh, 0.0)

[StillKeepTry]

We recommend you to use huggingface version for squad fine-tuning.