David
commited on
Commit
•
6864ed8
1
Parent(s):
aea99f0
Update README.md
Browse files
README.md
CHANGED
@@ -7,7 +7,7 @@ tags:
|
|
7 |
- tag2
|
8 |
license: apache-2.0
|
9 |
datasets:
|
10 |
-
-
|
11 |
metrics:
|
12 |
- metric1
|
13 |
- metric2
|
@@ -20,16 +20,20 @@ We release a `small` and `medium` version with the following configuration:
|
|
20 |
|
21 |
| Model | Layers | Embedding/Hidden Size | Params | Vocab Size | Max Sequence Length | Cased |
|
22 |
| --- | --- | --- | --- | --- | --- | --- |
|
23 |
-
| SELECTRA small | 12 | 256 | 22M | 50k | 512 | True |
|
24 |
| SELECTRA medium | 12 | 384 | 41M | 50k | 512 | True |
|
25 |
|
|
|
|
|
26 |
## Usage
|
27 |
|
|
|
|
|
28 |
```python
|
29 |
from transformers import ElectraForPreTraining, ElectraTokenizerFast
|
30 |
|
31 |
-
discriminator = ElectraForPreTraining.from_pretrained("
|
32 |
-
tokenizer = ElectraTokenizerFast.from_pretrained("
|
33 |
|
34 |
sentence_with_fake_token = "Estamos desayunando pan rosa con tomate y aceite de oliva."
|
35 |
|
@@ -55,10 +59,20 @@ We fine-tune our models on 4 different down-stream tasks:
|
|
55 |
- [CoNLL2002 - POS](https://huggingface.co/datasets/conll2002)
|
56 |
- [CoNLL2002 - NER](https://huggingface.co/datasets/conll2002)
|
57 |
|
58 |
-
We provide the mean and standard deviation of 5 fine-tuning runs.
|
59 |
-
|
60 |
-
|
61 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
|
63 |
## Training
|
64 |
|
@@ -66,4 +80,10 @@ We provide the mean and standard deviation of 5 fine-tuning runs.
|
|
66 |
|
67 |
## Motivation
|
68 |
|
69 |
-
Despite the abundance of excelent Spanish language models (BETO, bertin, etc) we felt there was still a lack of distilled or compact models with comparable metrics to their bigger siblings.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
- tag2
|
8 |
license: apache-2.0
|
9 |
datasets:
|
10 |
+
- oscar
|
11 |
metrics:
|
12 |
- metric1
|
13 |
- metric2
|
|
|
20 |
|
21 |
| Model | Layers | Embedding/Hidden Size | Params | Vocab Size | Max Sequence Length | Cased |
|
22 |
| --- | --- | --- | --- | --- | --- | --- |
|
23 |
+
| **SELECTRA small** | **12** | **256** | **22M** | **50k** | **512** | **True** |
|
24 |
| SELECTRA medium | 12 | 384 | 41M | 50k | 512 | True |
|
25 |
|
26 |
+
Selectra small is about 5 times smaller than BETO but achieves comparable results (see Metrics section below).
|
27 |
+
|
28 |
## Usage
|
29 |
|
30 |
+
|
31 |
+
|
32 |
```python
|
33 |
from transformers import ElectraForPreTraining, ElectraTokenizerFast
|
34 |
|
35 |
+
discriminator = ElectraForPreTraining.from_pretrained("Recognai/selectra_small")
|
36 |
+
tokenizer = ElectraTokenizerFast.from_pretrained("Recognai/selectra_small")
|
37 |
|
38 |
sentence_with_fake_token = "Estamos desayunando pan rosa con tomate y aceite de oliva."
|
39 |
|
|
|
59 |
- [CoNLL2002 - POS](https://huggingface.co/datasets/conll2002)
|
60 |
- [CoNLL2002 - NER](https://huggingface.co/datasets/conll2002)
|
61 |
|
62 |
+
We provide the mean and standard deviation of 5 fine-tuning runs.
|
63 |
+
|
64 |
+
The metrics
|
65 |
+
|
66 |
+
|
67 |
+
| Model | CoNLL2002 - POS (acc) | CoNLL2002 - NER (f1) | PAWS-X (acc) | XNLI (acc) | Params |
|
68 |
+
| --- | --- | --- | --- | --- | --- |
|
69 |
+
| SELECTRA small | 0.9653 +- 0.0007 | 0.863 +- 0.004 | 0.896 +- 0.002 | 0.784 +- 0.002 | 22M |
|
70 |
+
| SELECTRA medium | 0.9677 +- 0.0004 | 0.870 +- 0.003 | 0.896 +- 0.002 | 0.804 +- 0.002 | 41M |
|
71 |
+
| [mBERT](https://huggingface.co/bert-base-multilingual-cased) | 0.9689 | 0.8616 | 0.8895 | 0.7606 | 178M |
|
72 |
+
| [BETO](https://huggingface.co/dccuchile/bert-base-spanish-wwm-cased) | 0.9693 | 0.8596 | 0.8720 | 0.8012 | 110M |
|
73 |
+
| [BSC-BNE](https://huggingface.co/BSC-TeMU/roberta-base-bne) | 0.9706 | 0.8764 | 0.8815 | 0.7771 | 125M |
|
74 |
+
| [Bertin](https://huggingface.co/bertin-project/bertin-roberta-base-spanish/tree/v1-512) | 0.9697 | 0.8707 | 0.8965 | 0.7843 | 125M |
|
75 |
+
|
76 |
|
77 |
## Training
|
78 |
|
|
|
80 |
|
81 |
## Motivation
|
82 |
|
83 |
+
Despite the abundance of excelent Spanish language models (BETO, bertin, etc) we felt there was still a lack of distilled or compact models with comparable metrics to their bigger siblings.
|
84 |
+
|
85 |
+
## Acknowledgment
|
86 |
+
|
87 |
+
This research was supported by the use of the Google TPU Research Cloud (TRC).
|
88 |
+
|
89 |
+
## Authors
|