DistilBERT Base Uncased Fine-tuned SST-2
| Property | Value |
|---|---|
| Parameter Count | 67M |
| License | Apache-2.0 |
| Paper | View Paper |
| Accuracy | 91.05% |
| F1 Score | 0.914 |
What is distilbert-base-uncased-finetuned-sst-2-english?
This is a lightweight sentiment analysis model that leverages DistilBERT's architecture, fine-tuned specifically on the Stanford Sentiment Treebank (SST-2) dataset. It represents a powerful balance between performance and efficiency, achieving 91.3% accuracy on sentiment classification while maintaining a smaller footprint compared to BERT.
Implementation Details
The model was trained using carefully optimized hyperparameters including a learning rate of 1e-5, batch size of 32, and 3 training epochs. It employs the DistilBERT architecture, which maintains 97% of BERT's performance while being 40% smaller.
- Maximum sequence length: 128 tokens
- Warmup steps: 600
- Optimized for binary sentiment classification
- Supports both PyTorch and TensorFlow frameworks
Core Capabilities
- Binary sentiment classification (positive/negative)
- High-performance text classification with 91.05% accuracy
- Efficient inference with reduced model size
- Production-ready with multiple framework support
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its optimal balance between performance and efficiency. While it sacrifices only 1.4 percentage points in accuracy compared to BERT-base, it offers significantly faster inference times and lower resource requirements.
Q: What are the recommended use cases?
The model is ideal for production sentiment analysis tasks, particularly in scenarios requiring real-time analysis of user feedback, review classification, or social media monitoring. However, users should be aware of potential biases in predictions, particularly regarding geographical references.
