bert-imdb-finetuned

Fine-Tuned BERT for IMDb Sentiment Analysis

This repository contains a fine-tuned BERT-base-uncased model for binary sentiment classification (positive/negative) on the IMDb movie review dataset.

Model Overview

• Base model: bert-base-uncased
• Task: Sentiment classification
• Dataset: IMDb (binary labels)
• Metrics: Accuracy, Precision, Recall, F1-score

Performance

Training Details

• Framework: Transformers, PyTorch
• Hardware: Google Colab Pro
• Techniques used:
  • Label smoothing
  • Early stopping
  • Weight decay
  • Dropout regularization
  • WANDB for tracking metrics
• Hyperparameters:
  • Learning rate: 2e-5
  • Batch size: 16
  • Epochs: 10
  • Dropout: 0.3
  • Label smoothing factor: 0.1

Usage

Inference

from transformers import pipeline

classifier = pipeline("text-classification", model="your-username/your-model-name")
result = classifier("This movie was fantastic!")
print(result)
# [{'label': 'POSITIVE', 'score': 0.99}]