# oBERT-12-upstream-pruned-unstructured-97-finetuned-squadv1-v2

This model is obtained with [The Optimal BERT Surgeon: Scalable and Accurate Second-Order Pruning for Large Language Models](https://arxiv.org/abs/2203.07259).


It corresponds to the model presented in the `Table 2 - oBERT - SQuADv1 97%` (in the upcoming updated version of the paper).

```
Pruning method: oBERT upstream unstructured + sparse-transfer to downstream
Paper: https://arxiv.org/abs/2203.07259
Dataset: SQuADv1
Sparsity: 97%
Number of layers: 12
```

The dev-set performance reported in the paper is averaged over four seeds, and we release the best model (marked with `(*)`):

```
| oBERT 97%     | F1    | EM    |
| ------------- | ----- | ----- |
| seed=42       | 84.92 | 76.94 |
| seed=3407     | 84.87 | 76.71 |
| seed=123      | 84.95 | 77.06 |
| seed=12345 (*)| 84.95 | 76.90 |
| ------------- | ----- | ----- |
| mean          | 84.92 | 76.90 |
| stdev         | 0.037 | 0.145 |
```

Code: _coming soon_

## BibTeX entry and citation info
```bibtex
@article{kurtic2022optimal,
  title={The Optimal BERT Surgeon: Scalable and Accurate Second-Order Pruning for Large Language Models},
  author={Kurtic, Eldar and Campos, Daniel and Nguyen, Tuan and Frantar, Elias and Kurtz, Mark and Fineran, Benjamin and Goin, Michael and Alistarh, Dan},
  journal={arXiv preprint arXiv:2203.07259},
  year={2022}
}
```