Legal BERTimbau Sts Base Ma

🚀 rufimelo/Legal-BERTimbau-sts-base-ma

This is a sentence-transformers model that maps sentences and paragraphs to a 768-dimensional dense vector space. It can be used for tasks such as clustering or semantic search. rufimelo/Legal-BERTimbau-sts-base-ma is based on Legal-BERTimbau-base, which is derived from the BERTimbau large model. It is adapted to the Portuguese legal domain and trained for Semantic Textual Similarity (STS) on Portuguese datasets.

🚀 Quick Start

Prerequisites

To use this model, you need to have sentence-transformers installed. You can install it using the following command:

pip install -U sentence-transformers

Basic Usage

Here is a simple example of how to use the model:

from sentence_transformers import SentenceTransformer
sentences = ["Isto é um exemplo", "Isto é um outro exemplo"]

model = SentenceTransformer('rufimelo/Legal-BERTimbau-sts-base-ma')
embeddings = model.encode(sentences)
print(embeddings)

Advanced Usage with HuggingFace Transformers

from transformers import AutoTokenizer, AutoModel
import torch


#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)


# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('rufimelo/Legal-BERTimbau-sts-base-ma')
model = AutoModel.from_pretrained('rufimelo/Legal-BERTimbau-sts-base-ma')

# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')

# Compute token embeddings
with torch.no_grad():
    model_output = model(**encoded_input)

# Perform pooling. In this case, mean pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

print("Sentence embeddings:")
print(sentence_embeddings)

✨ Features

• Dense Vector Representation: Maps sentences and paragraphs to a 768-dimensional dense vector space.
• Adapted to Legal Domain: Specifically adapted to the Portuguese legal domain.
• Trained for STS: Trained for Semantic Textual Similarity on Portuguese datasets.

📦 Installation

To install the necessary dependencies, run the following command:

pip install -U sentence-transformers

💻 Usage Examples

Basic Usage

from sentence_transformers import SentenceTransformer
sentences = ["Isto é um exemplo", "Isto é um outro exemplo"]

model = SentenceTransformer('rufimelo/Legal-BERTimbau-sts-base-ma')
embeddings = model.encode(sentences)
print(embeddings)

Advanced Usage

from transformers import AutoTokenizer, AutoModel
import torch


#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)


# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('rufimelo/Legal-BERTimbau-sts-base-ma')
model = AutoModel.from_pretrained('rufimelo/Legal-BERTimbau-sts-base-ma')

# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')

# Compute token embeddings
with torch.no_grad():
    model_output = model(**encoded_input)

# Perform pooling. In this case, mean pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

print("Sentence embeddings:")
print(sentence_embeddings)

📚 Documentation

Model Information

• Model Type: Sentence-transformers
• Based on: Legal-BERTimbau-base, derived from BERTimbau large
• Adapted to: Portuguese legal domain
• Trained for: Semantic Textual Similarity (STS) on Portuguese datasets

Datasets Used for Training

• assin
• assin2
• stsb_multi_mt
• rufimelo/PortugueseLegalSentences-v0

Evaluation Results (STS)

Training Process

1. Multilingual Knowledge Distillation: Due to the lack of Portuguese datasets, the model was initially trained using multilingual knowledge distillation. The teacher model was 'sentence-transformers/paraphrase-xlm-r-multilingual-v1', with English as the supposed supported language and Portuguese as the language to learn.
2. Fine-tuning: The model was then fine-tuned for Semantic Textual Similarity on the assin, assin2, and stsb_multi_mt pt datasets.

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, '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})
)

🔧 Technical Details

The model is based on the Sentence-Transformers framework, which maps text to a dense vector space. It uses a BERT-based architecture (Legal-BERTimbau-base) and is fine-tuned for Semantic Textual Similarity on Portuguese legal datasets. The multilingual knowledge distillation process helps to leverage the knowledge from a multilingual teacher model, and the subsequent fine-tuning on Portuguese datasets adapts the model to the specific domain.

📄 License

The license information is not provided in the original README. Please refer to the model's repository on Hugging Face for the license details.

Citing & Authors

If you use this work, please cite the following papers:

@inproceedings{souza2020bertimbau,
  author    = {F{\'a}bio Souza and
               Rodrigo Nogueira and
               Roberto Lotufo},
  title     = {{BERT}imbau: pretrained {BERT} models for {B}razilian {P}ortuguese},
  booktitle = {9th Brazilian Conference on Intelligent Systems, {BRACIS}, Rio Grande do Sul, Brazil, October 20-23 (to appear)},
  year      = {2020}
}

@inproceedings{fonseca2016assin,
  title={ASSIN: Avaliacao de similaridade semantica e inferencia textual},
  author={Fonseca, E and Santos, L and Criscuolo, Marcelo and Aluisio, S},
  booktitle={Computational Processing of the Portuguese Language-12th International Conference, Tomar, Portugal},
  pages={13--15},
  year={2016}
}

@inproceedings{real2020assin,
  title={The assin 2 shared task: a quick overview},
  author={Real, Livy and Fonseca, Erick and Oliveira, Hugo Goncalo},
  booktitle={International Conference on Computational Processing of the Portuguese Language},
  pages={406--412},
  year={2020},
  organization={Springer}
}
@InProceedings{huggingface:dataset:stsb_multi_mt,
title = {Machine translated multilingual STS benchmark dataset.},
author={Philip May},
year={2021},
url={https://github.com/PhilipMay/stsb-multi-mt}
}