Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,112 @@
|
|
1 |
---
|
2 |
-
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
language:
|
3 |
+
- pt
|
4 |
+
thumbnail: "Portuguese BERT for the Legal Domain"
|
5 |
+
tags:
|
6 |
+
- sentence-transformers
|
7 |
+
- transformers
|
8 |
---
|
9 |
+
|
10 |
+
# stjiris/bert-large-portuguese-cased-legal-mlm (Legal BERTimbau)
|
11 |
+
This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 1024 dimensional dense vector space and can be used for tasks like clustering or semantic search.
|
12 |
+
stjiris/bert-large-portuguese-cased-legal-mlm derives from [BERTimbau](https://huggingface.co/neuralmind/bert-large-portuguese-cased) large.
|
13 |
+
|
14 |
+
It was trained using the MLM technique with a learning rate 1e-5 [Legal Sentences from +-30000 documents](https://huggingface.co/datasets/stjiris/portuguese-legal-sentences-v1.0) 5300 training steps (best performance for our semantic search system implementation)
|
15 |
+
|
16 |
+
|
17 |
+
## Usage (Sentence-Transformers)
|
18 |
+
Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:
|
19 |
+
```
|
20 |
+
pip install -U sentence-transformers
|
21 |
+
```
|
22 |
+
Then you can use the model like this:
|
23 |
+
```python
|
24 |
+
from sentence_transformers import SentenceTransformer
|
25 |
+
sentences = ["Isto é um exemplo", "Isto é um outro exemplo"]
|
26 |
+
|
27 |
+
model = SentenceTransformer('stjiris/bert-large-portuguese-cased-legal-mlm')
|
28 |
+
embeddings = model.encode(sentences)
|
29 |
+
print(embeddings)
|
30 |
+
```
|
31 |
+
## Usage (HuggingFace Transformers)
|
32 |
+
```python
|
33 |
+
from transformers import AutoTokenizer, AutoModel
|
34 |
+
import torch
|
35 |
+
|
36 |
+
|
37 |
+
#Mean Pooling - Take attention mask into account for correct averaging
|
38 |
+
def mean_pooling(model_output, attention_mask):
|
39 |
+
token_embeddings = model_output[0] #First element of model_output contains all token embeddings
|
40 |
+
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
|
41 |
+
return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
|
42 |
+
|
43 |
+
# Sentences we want sentence embeddings for
|
44 |
+
sentences = ['This is an example sentence', 'Each sentence is converted']
|
45 |
+
|
46 |
+
# Load model from HuggingFace Hub
|
47 |
+
tokenizer = AutoTokenizer.from_pretrained('stjiris/bert-large-portuguese-cased-legal-mlm')
|
48 |
+
model = AutoModel.from_pretrained('stjiris/bert-large-portuguese-cased-legal-mlm')
|
49 |
+
|
50 |
+
# Tokenize sentences
|
51 |
+
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
|
52 |
+
|
53 |
+
# Compute token embeddings
|
54 |
+
with torch.no_grad():
|
55 |
+
model_output = model(**encoded_input)
|
56 |
+
# Perform pooling. In this case, mean pooling.
|
57 |
+
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
|
58 |
+
print("Sentence embeddings:")
|
59 |
+
print(sentence_embeddings)
|
60 |
+
```
|
61 |
+
|
62 |
+
|
63 |
+
## Full Model Architecture
|
64 |
+
```
|
65 |
+
SentenceTransformer(
|
66 |
+
(0): Transformer({'max_seq_length': 514, 'do_lower_case': False}) with Transformer model: BertModel
|
67 |
+
(1): Pooling({'word_embedding_dimension': 1028, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False})
|
68 |
+
)
|
69 |
+
```
|
70 |
+
## Citing & Authors
|
71 |
+
|
72 |
+
If you use this work, please cite:
|
73 |
+
|
74 |
+
```bibtex
|
75 |
+
@inproceedings{MeloSemantic,
|
76 |
+
author = {Melo, Rui and Santos, Professor Pedro Alexandre and Dias, Professor Jo{\~ a}o},
|
77 |
+
title = {A {Semantic} {Search} {System} for {Supremo} {Tribunal} de {Justi}{\c c}a},
|
78 |
+
}
|
79 |
+
|
80 |
+
@inproceedings{souza2020bertimbau,
|
81 |
+
author = {F{\'a}bio Souza and
|
82 |
+
Rodrigo Nogueira and
|
83 |
+
Roberto Lotufo},
|
84 |
+
title = {{BERT}imbau: pretrained {BERT} models for {B}razilian {P}ortuguese},
|
85 |
+
booktitle = {9th Brazilian Conference on Intelligent Systems, {BRACIS}, Rio Grande do Sul, Brazil, October 20-23 (to appear)},
|
86 |
+
year = {2020}
|
87 |
+
}
|
88 |
+
|
89 |
+
@inproceedings{fonseca2016assin,
|
90 |
+
title={ASSIN: Avaliacao de similaridade semantica e inferencia textual},
|
91 |
+
author={Fonseca, E and Santos, L and Criscuolo, Marcelo and Aluisio, S},
|
92 |
+
booktitle={Computational Processing of the Portuguese Language-12th International Conference, Tomar, Portugal},
|
93 |
+
pages={13--15},
|
94 |
+
year={2016}
|
95 |
+
}
|
96 |
+
|
97 |
+
@inproceedings{real2020assin,
|
98 |
+
title={The assin 2 shared task: a quick overview},
|
99 |
+
author={Real, Livy and Fonseca, Erick and Oliveira, Hugo Goncalo},
|
100 |
+
booktitle={International Conference on Computational Processing of the Portuguese Language},
|
101 |
+
pages={406--412},
|
102 |
+
year={2020},
|
103 |
+
organization={Springer}
|
104 |
+
}
|
105 |
+
@InProceedings{huggingface:dataset:stsb_multi_mt,
|
106 |
+
title = {Machine translated multilingual STS benchmark dataset.},
|
107 |
+
author={Philip May},
|
108 |
+
year={2021},
|
109 |
+
url={https://github.com/PhilipMay/stsb-multi-mt}
|
110 |
+
}
|
111 |
+
|
112 |
+
```
|