oBERT-12-downstream-dense-squadv1
This model is obtained with The Optimal BERT Surgeon: Scalable and Accurate Second-Order Pruning for Large Language Models.
It corresponds to the model presented in the Table 3 - 12 Layers - 0% Sparsity
, and it represents an upper bound for performance of the corresponding pruned models:
- 80% unstructured:
neuralmagic/oBERT-12-downstream-pruned-unstructured-80-squadv1
- 80% block-4:
neuralmagic/oBERT-12-downstream-pruned-block4-80-squadv1
- 90% unstructured:
neuralmagic/oBERT-12-downstream-pruned-unstructured-90-squadv1
- 90% block-4:
neuralmagic/oBERT-12-downstream-pruned-block4-90-squadv1
SQuADv1 dev-set:
EM = 82.71
F1 = 89.48
BibTeX entry and citation info
@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}
}