Este modelo de lenguaje pre-entrenado está ajustado para la tarea de detección de posturas específicamente para Joe Biden. Por favor, vea el repositorio oficial para más detalles.

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
import numpy as np

# Elegir GPU si disponible
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Seleccione la ruta del modelo aquí
pretrained_LM_path = "kornosk/bert-election2020-twitter-stance-biden-KE-MLM"

# Cargar modelo
tokenizer = AutoTokenizer.from_pretrained(pretrained_LM_path)
model = AutoModelForSequenceClassification.from_pretrained(pretrained_LM_path)

id2label = {
  0: "AGAINST",
  1: "FAVOR",
  2: "NONE"
}

##### Predicción Neutral #####
sentence = "Hello World."
inputs = tokenizer(sentence.lower(), return_tensors="pt")
outputs = model(**inputs)
predicted_probability = torch.softmax(outputs[0], dim=1)[0].tolist()

print("Sentence:", sentence)
print("Prediction:", id2label[np.argmax(predicted_probability)])
print("Against:", predicted_probability[0])
print("Favor:", predicted_probability[1])
print("Neutral:", predicted_probability[2])

##### Predicción Favor #####
sentence = "Go Go Biden!!!"
inputs = tokenizer(sentence.lower(), return_tensors="pt")
outputs = model(**inputs)
predicted_probability = torch.softmax(outputs[0], dim=1)[0].tolist()

print("Sentence:", sentence)
print("Prediction:", id2label[np.argmax(predicted_probability)])
print("Against:", predicted_probability[0])
print("Favor:", predicted_probability[1])
print("Neutral:", predicted_probability[2])

##### Predicción Against #####
sentence = "Biden is the worst."
inputs = tokenizer(sentence.lower(), return_tensors="pt")
outputs = model(**inputs)
predicted_probability = torch.softmax(outputs[0], dim=1)[0].tolist()

print("Sentence:", sentence)
print("Prediction:", id2label[np.argmax(predicted_probability)])
print("Against:", predicted_probability[0])
print("Favor:", predicted_probability[1])
print("Neutral:", predicted_probability[2])

Por favor considere citar nuestro trabajo si le resulta útil. :)