| Albert v2 finetuned on SQuAD v1. | |
| Trained using the [nn_pruning](https://github.com/huggingface/nn_pruning/tree/main/examples/question_answering) script, with pruning disabled. | |
| [Original results](https://github.com/google-research/albert) are F1=90.2, EM=83.2, we improved them to: | |
| ```{ | |
| "exact_match": 83.74645222327341, | |
| "f1": 90.78776054621733 | |
| }``` |