O modelo foi treinado utilizando o dataset IMDB, amplamente utilizado para tarefas de classificação de texto, especialmente para análise de sentimentos. O dataset contém 50.000 revisões de filmes rotuladas, divididas igualmente entre revisões positivas e negativas, com 25.000 exemplos para treinamento e 25.000 para teste.
Para carregar o dataset, é preciso utilizar a biblioteca datasets da Hugging Face:
-
Carregar o dataset:
from datasets import load_dataset
dataset = load_dataset("imdb")
-
Pré-processamento:
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
tokenized_datasets = dataset.map(lambda x: tokenizer(x['text'], padding='max_length', truncation=True), batched=True)
-
Definir o Modelo e Argumentos de Treinamento:
from transformers import AutoModelForSequenceClassification, TrainingArguments, Trainer
import numpy as np
model = AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased", num_labels=2)
training_args = TrainingArguments(
output_dir="./results",
learning_rate=2e-5,
per_device_train_batch_size=32,
per_device_eval_batch_size=32,
num_train_epochs=1,
weight_decay=0.01,
evaluation_strategy="epoch",
push_to_hub=True
)
def compute_metrics(eval_pred):
logits, labels = eval_pred
predictions = np.argmax(logits, axis=-1)
return {"accuracy": (predictions == labels).mean()}
-
Treinamento:
small_train_dataset = tokenized_datasets["train"].shuffle(seed=42).select(range(1000))
small_eval_dataset = tokenized_datasets["test"].shuffle(seed=42).select(range(100))
trainer = Trainer(
model=model,
args=training_args,
train_dataset=small_train_dataset,
eval_dataset=small_eval_dataset,
compute_metrics=compute_metrics
)
trainer.train()
result = pipe("I loved this movie! It was fantastic and thrilling.")
print(result)
Carregando o Modelo Diretamente:
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import numpy as np
tokenizer = AutoTokenizer.from_pretrained("pedro123483/results")
model = AutoModelForSequenceClassification.from_pretrained("pedro123483/results")
inputs = tokenizer("I loved this movie! It was fantastic and thrilling.", return_tensors="pt")
outputs = model(**inputs)
predictions = np.argmax(outputs.logits.detach().numpy(), axis=-1)
print(predictions)
This model is a fine-tuned version of
distilbert-base-uncased on an unknown dataset.
