Paraphrase Bert Pt

🚀 Prompsit/paraphrase-bert-pt

This model enables the evaluation of paraphrases for a given phrase, offering valuable insights in text - classification tasks.

🚀 Quick Start

This model is designed to answer the question: Is "phrase B" a paraphrase of "phrase A"? It's important to note that the model operates on phrases rather than sentences. Thus, it doesn't expect to encounter punctuation marks or long text passages.

The resulting probabilities correspond to the following classes:

• 0: Not a paraphrase
• 1: It's a paraphrase

💻 Usage Examples

Basic Usage

# Import necessary libraries
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

# Initialize tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("Prompsit/paraphrase-bert-pt")
model = AutoModelForSequenceClassification.from_pretrained("Prompsit/paraphrase-bert-pt")

# Prepare input
input = tokenizer('logo após o homicídio','pouco depois do assassinato',return_tensors='pt')

# Get logits
logits = model(**input).logits

# Apply softmax
soft = torch.nn.Softmax(dim=1)

# Print results
print(soft(logits))

Advanced Usage

# The model was fine - tuned from pretrained "neuralmind/bert-base-portuguese-cased".
# Here is an example of using it in a more complex scenario where you can loop through multiple pairs of phrases.
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("Prompsit/paraphrase-bert-pt")
model = AutoModelForSequenceClassification.from_pretrained("Prompsit/paraphrase-bert-pt")

phrase_pairs = [
    ('logo após o homicídio', 'pouco depois do assassinato'),
    ('outro par de frases', 'mais um par de candidatos')
]

for pair in phrase_pairs:
    input = tokenizer(pair[0], pair[1], return_tensors='pt')
    logits = model(**input).logits
    soft = torch.nn.Softmax(dim=1)
    print(soft(logits))

📚 Documentation

This model was fine - tuned from the pretrained "neuralmind/bert - base - portuguese - cased". It was built under a TSI - 100905 - 2019 - 4 project, co - financed by the Ministry of Economic Affairs and Digital Transformation from the Government of Spain.

🔧 Technical Details

Evaluation Results

We used a test dataset consisting of 16500 pairs of human - tagged phrases. The following metrics were obtained:

metrics={
 'test_loss': 0.6074697375297546, 
 'test_accuracy': 0.7809, 
 'test_precision': 0.7157638466220329, 
 'test_recall': 0.40551724137931033, 
 'test_f1': 0.5177195685670262, 
 'test_matthews_correlation': 0.41603913834665324, 
 'test_runtime': 16.4585, 
 'test_samples_per_second': 607.587, 
 'test_steps_per_second': 19.017
}