File size: 4,637 Bytes
b1e7fea
 
 
322968f
 
 
b1e7fea
322968f
 
b1e7fea
 
 
 
 
322968f
b1e7fea
 
 
 
 
 
 
 
 
 
 
554c817
 
 
b1e7fea
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bfa4dbb
 
b1e7fea
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
322968f
 
 
 
 
 
 
 
 
 
 
 
b1e7fea
 
 
 
 
 
 
 
1353d86
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b1e7fea
 
 
 
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
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
---
language: es
thumbnail: https://i.imgur.com/uxAvBfh.png
tags:
 - Spanish
 - Electra

datasets:
 -large_spanish_corpus

---

## ELECTRICIDAD: The Spanish Electra [Imgur](https://imgur.com/uxAvBfh)

**Electricidad-base-discriminator** (uncased) is a ```base``` Electra like model (discriminator in this case) trained on a [Large Spanish Corpus](https://github.com/josecannete/spanish-corpora) (aka BETO's corpus)

As mentioned in the original [paper](https://openreview.net/pdf?id=r1xMH1BtvB):
**ELECTRA** is a new method for self-supervised language representation learning. It can be used to pre-train transformer networks using relatively little compute. ELECTRA models are trained to distinguish "real" input tokens vs "fake" input tokens generated by another neural network, similar to the discriminator of a [GAN](https://arxiv.org/pdf/1406.2661.pdf). At small scale, ELECTRA achieves strong results even when trained on a single GPU. At large scale, ELECTRA achieves state-of-the-art results on the [SQuAD 2.0](https://rajpurkar.github.io/SQuAD-explorer/) dataset.

For a detailed description and experimental results, please refer the paper [ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators](https://openreview.net/pdf?id=r1xMH1BtvB).


## Model details ⚙

|Name| # Value|
|-----|--------|
|Layers| 12 |
|Hidden | 768 |
|Params| 110M |

## Evaluation metrics (for discriminator) 🧾

|Metric | # Score |
|-------|---------|
|Accuracy| 0.985|
|Precision| 0.726|
|AUC | 0.922|



## Fast example of usage 🚀

```python
from transformers import ElectraForPreTraining, ElectraTokenizerFast
import torch

discriminator = ElectraForPreTraining.from_pretrained("mrm8488/electricidad-base-discriminator")
tokenizer = ElectraTokenizerFast.from_pretrained("mrm8488/electricidad-base-discriminator")

sentence = "El rápido zorro marrón salta sobre el perro perezoso"
fake_sentence = "El rápido zorro marrón amar sobre el perro perezoso"

fake_tokens = tokenizer.tokenize(fake_sentence)
fake_inputs = tokenizer.encode(fake_sentence, return_tensors="pt")
discriminator_outputs = discriminator(fake_inputs)
predictions = torch.round((torch.sign(discriminator_outputs[0]) + 1) / 2)

[print("%7s" % token, end="") for token in fake_tokens]

[print("%7s" % prediction, end="") for prediction in predictions.tolist()]

# Output:
'''
el rapido  zorro  marro    ##n   amar  sobre     el  perro   pere ##zoso    0.0    0.0    0.0    0.0    0.0    0.0    1.0    1.0    0.0    0.0    0.0    0.0    0.0[None, None, None, None, None, None, None, None, None, None, None, None, None
'''
```
As you can see there are **1s** in the places where the model detected a fake token. So, it works! 🎉


### Some models fine-tuned on a downstream task 🛠️

[Question Answering](https://huggingface.co/mrm8488/electricidad-base-finetuned-squadv1-es)

[POS](https://huggingface.co/mrm8488/electricidad-base-finetuned-pos)

[NER](https://huggingface.co/mrm8488/electricidad-base-finetuned-ner)


### Spanish LM model comparison 📊
| Dataset     | Metric   | RoBERTa-b | RoBERTa-l | BETO   | mBERT  | BERTIN | Electricidad-b |
|-------------|----------|-----------|-----------|--------|--------|--------|---------|
| UD-POS      | F1       | 0.9907    | 0.9901    | 0.9900 | 0.9886 | 0.9904 | 0.9818  |
| Conll-NER   | F1       | 0.8851    | 0.8772    | 0.8759 | 0.8691 | 0.8627 | 0.7954  |
| Capitel-POS | F1       | 0.9846    | 0.9851    | 0.9836 | 0.9839 | 0.9826 | 0.9816  |
| Capitel-NER | F1       | 0.8959    | 0.8998    | 0.8771 | 0.8810 | 0.8741 | 0.8035  |
| STS         | Combined | 0.8423    | 0.8420    | 0.8216 | 0.8249 | 0.7822 | 0.8065  |
| MLDoc       | Accuracy | 0.9595    | 0.9600    | 0.9650 | 0.9560 | 0.9673 | 0.9490  |
| PAWS-X      | F1       | 0.9035    | 0.9000    | 0.8915 | 0.9020 | 0.8820 | **0.9045**  |
| XNLI        | Accuracy | 0.8016    | 0.7958    | 0.8130 | 0.7876 | 0.7864 | 0.7878  |



## Acknowledgments

I thank [🤗/transformers team](https://github.com/huggingface/transformers) for allowing me to train the model (specially to [Julien Chaumond](https://twitter.com/julien_c)).


## Citation
If you want to cite this model you can use this:

```bibtex
@misc{mromero2020electricidad-base-discriminator,
  title={Spanish Electra by Manuel Romero},
  author={Romero, Manuel},
  publisher={Hugging Face},
  journal={Hugging Face Hub},
  howpublished={\url{https://huggingface.co/mrm8488/electricidad-base-discriminator/}},
  year={2020}
}
```



> Created by [Manuel Romero/@mrm8488](https://twitter.com/mrm8488)

> Made with <span style="color: #e25555;">&hearts;</span> in Spain