Table 1 Performance of the DISTIL-BERT-MRC model on squad 2.0, NewsQA, and natural questions datasets across different Epochs.
Model | Dataset | Epochs # | Train Acc % | Valid Acc % | Train Loss % | Valid Loss % |
|---|---|---|---|---|---|---|
Distil-BERT-MRC | SQuAD 2.0 | 1 | 76.10 | 68.85 | 1.18 | 1.41 |
| Â | 5 | 91.30 | 89.20 | 0.49 | 0.60 | |
| Â | 10 | 93.50 | 91.80 | 0.33 | 0.44 | |
| Â | 15 | 94.10 | 93.25 | 0.27 | 0.38 | |
NewsQA | 1 | 78.00 | 74.45 | 1.15 | 1.28 | |
| Â | 5 | 91.80 | 90.05 | 0.46 | 0.56 | |
| Â | 10 | 93.80 | 92.80 | 0.29 | 0.39 | |
| Â | 15 | 94.40 | 94.52 | 0.25 | 0.36 | |
Natural Questions | 1 | 77.20 | 72.65 | 1.16 | 1.33 | |
| Â | 5 | 91.50 | 89.90 | 0.47 | 0.57 | |
| Â | Â | 10 | 93.90 | 92.20 | 0.31 | 0.41 |
| Â | Â | 15 | 94.20 | 92.58 | 0.28 | 0.37 |
