File size: 2,465 Bytes
24ca65d fb2efbd 522f4c8 fb2efbd 553eabd 522f4c8 8d75976 568ad9d 553eabd 8d75976 72e3301 8d75976 553eabd 8d75976 ed45e01 8d75976 875f10a 8d75976 5dd2a91 8d75976 827b924 5dd2a91 df5385a 5dd2a91 9e31f0e 568ad9d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 |
---
language: ro
license: mit
datasets:
- oscar
- wikipedia
---
# Romanian DistilBERT
This repository contains the uncased Romanian DistilBERT (named Distil-BERT-base-ro in the paper). The teacher model used for distillation is: [dumitrescustefan/bert-base-romanian-cased-v1](https://huggingface.co/dumitrescustefan/bert-base-romanian-cased-v1).
The model was introduced in [this paper](https://arxiv.org/abs/2112.12650). The adjacent code can be found
[here](https://github.com/racai-ai/Romanian-DistilBERT).
## Usage
```python
from transformers import AutoTokenizer, AutoModel
# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained("racai/distilbert-base-romanian-cased")
model = AutoModel.from_pretrained("racai/distilbert-base-romanian-cased")
# tokenize a test sentence
input_ids = tokenizer.encode("Aceasta este o propoziție de test.", add_special_tokens=True, return_tensors="pt")
# run the tokens trough the model
outputs = model(input_ids)
print(outputs)
```
## Model Size
It is 35% smaller than its teacher `bert-base-romanian-cased-v1`.
| Model | Size (MB) | Params (Millions) |
|--------------------------------|:---------:|:----------------:|
| bert-base-romanian-cased-v1 | 477 | 124 |
| distilbert-base-romanian-cased | 312 | 81 |
## Evaluation
We evaluated the model in comparison with its teacher on 5 Romanian tasks:
- **UPOS**: Universal Part of Speech (F1-macro)
- **XPOS**: Extended Part of Speech (F1-macro)
- **NER**: Named Entity Recognition (F1-macro)
- **SAPN**: Sentiment Anlaysis - Positive vs Negative (Accuracy)
- **SAR**: Sentiment Analysis - Rating (F1-macro)
- **DI**: Dialect identification (F1-macro)
- **STS**: Semantic Textual Similarity (Pearson)
| Model | UPOS | XPOS | NER | SAPN | SAR | DI | STS |
|--------------------------------|:----:|:----:|:---:|:----:|:---:|:--:|:---:|
| bert-base-romanian-cased-v1 | 98.00 | 96.46 | 85.88 | 98.07 | 79.61 | 95.58 | 80.30 |
| distilbert-base-romanian-cased | 97.97 | 97.08 | 83.35 | 98.20 | 80.51 | 96.31 | 80.57 |
### BibTeX entry and citation info
```bibtex
@article{avram2021distilling,
title={Distilling the Knowledge of Romanian BERTs Using Multiple Teachers},
author={Andrei-Marius Avram and Darius Catrina and Dumitru-Clementin Cercel and Mihai Dascălu and Traian Rebedea and Vasile Păiş and Dan Tufiş},
journal={ArXiv},
year={2021},
volume={abs/2112.12650}
}
```
|