language: en
license: apache-2.0
tags:
- text-classification
datasets:
- sst-2
metrics:
- accuracy
bert-base-uncased model fine-tuned on SST-2
This model was created using the nn_pruning python library: the linear layers contains 37% of the original weights.
The model contains 51% of the original weights overall (the embeddings account for a significant part of the model, and they are not pruned by this method).
In terms of perfomance, its accuracy is 91.17.
Fine-Pruning details
This model was fine-tuned from the HuggingFace model checkpoint on task, and distilled from the model textattack/bert-base-uncased-SST-2. This model is case-insensitive: it does not make a difference between english and English.
A side-effect of the block pruning method is that some of the attention heads are completely removed: 88 heads were removed on a total of 144 (61.1%). Here is a detailed view on how the remaining heads are distributed in the network after pruning.
Details of the SST-2 dataset
| Dataset | Split | # samples |
|---|---|---|
| SST-2 | train | 67K |
| SST-2 | eval | 872 |
Results
Pytorch model file size: 351MB (original BERT: 420MB)
| Metric | # Value | # Original (Table 2) | Variation |
|---|---|---|---|
| accuracy | 91.17 | 92.7 | -1.53 |