MiniLM-evidence-types
| Property | Value |
|---|---|
| License | MIT |
| Base Model | microsoft/MiniLM-L12-H384-uncased |
| Best Accuracy | 71.61% |
| Training Framework | PyTorch 1.11.0 |
What is MiniLM-evidence-types?
MiniLM-evidence-types is a specialized text classification model fine-tuned from Microsoft's MiniLM architecture, designed specifically for identifying and categorizing different types of evidence in text. The model demonstrates robust performance with a 71.61% accuracy and a weighted F1 score of 0.7030, making it particularly suitable for evidence-based text analysis tasks.
Implementation Details
The model was trained using a carefully optimized configuration including Adam optimizer with a learning rate of 2e-05, batch size of 16, and ran for 20 epochs using mixed precision training. The training process showed consistent improvement in performance, with the final model achieving balanced metrics across different evidence categories.
- Native AMP mixed precision training
- Linear learning rate scheduler
- 20 epochs of training with careful monitoring
- Optimized batch size for performance
Core Capabilities
- Evidence type classification with high accuracy
- Balanced performance across categories (0.3616 balanced accuracy)
- Efficient inference with MiniLM architecture
- Robust macro F1 score of 0.3726
Frequently Asked Questions
Q: What makes this model unique?
The model's specialization in evidence type classification, combined with its efficient MiniLM architecture, makes it particularly valuable for tasks requiring nuanced understanding of evidence in text while maintaining computational efficiency.
Q: What are the recommended use cases?
This model is ideal for applications involving evidence classification in academic or research contexts, fact-checking systems, and automated content analysis where identifying different types of evidence is crucial.
