| --- |
| license: mit |
| library_name: transformers |
| --- |
| # TextClassifier |
|
|
| <div align="center"> |
| <img src="figures/fig1.png" width="70%" alt="Training Curves" /> |
| </div> |
|
|
| ## Model Description |
|
|
| This is a BERT-based text classification model fine-tuned on a 5-class dataset. The best checkpoint was selected based on validation F1 score across multiple hyperparameter sweeps. |
|
|
| ## Training Details |
|
|
| - **Base Model**: bert-base-uncased |
| - **Best Run ID**: run-def456 |
| - **Best Run Name**: sweep-lr5e5-bs16 |
| - **Learning Rate**: 5e-05 |
| - **Batch Size**: 16 |
| - **Weight Decay**: 0.01 |
| - **Best Epoch**: 10 |
|
|
| ## Evaluation Results |
|
|
| - **Validation F1**: 0.851 |
| - **Validation Accuracy**: 0.865 |
| - **Final Validation Loss**: 0.487 |
|
|
| <div align="center"> |
| <img src="figures/fig2.png" width="60%" alt="Validation Metrics" /> |
| </div> |
|
|
| ### Run Comparison (sorted by val_f1 descending) |
| |
| | Run ID | Run Name | Learning Rate | Batch Size | Weight Decay | Val F1 | Val Accuracy | Val Loss | Best Epoch | |
| |--------|----------|---------------|------------|--------------|--------|--------------|----------|-------------| |
| | run-def456 | sweep-lr5e5-bs16 | 5e-05 | 16 | 0.01 | 0.851 | 0.865 | 0.487 | 10 | |
| | run-pqr678 | sweep-lr5e5-bs16-wd005 | 5e-05 | 16 | 0.005 | 0.841 | 0.855 | 0.512 | 10 | |
| | run-jkl012 | sweep-lr5e5-bs32-wd0 | 5e-05 | 32 | 0.0 | 0.829 | 0.841 | 0.583 | 8 | |
| | run-abc123 | sweep-lr3e5-bs32 | 3e-05 | 32 | 0.01 | 0.811 | 0.826 | 0.585 | 10 | |
| | run-ghi789 | sweep-lr2e5-bs64 | 2e-05 | 64 | 0.02 | 0.782 | 0.796 | 0.649 | 10 | |
| | run-mno345 | sweep-lr1e4-bs32 | 0.0001 | 32 | 0.01 | 0.735 | 0.751 | 0.821 | 7 | |
| |
| <div align="center"> |
| <img src="figures/fig3.png" width="60%" alt="Confusion Matrix" /> |
| </div> |
| |
| ## Intended Use |
| |
| This model is intended for text classification tasks with 5 output classes. It should not be used for generating text or for tasks outside its training distribution. |
| |
| ## Limitations |
| |
| The model's performance is benchmark-specific and may not generalize to out-of-distribution inputs or domains not seen during training. |
| |
| ## License |
| |
| This model is released under the MIT License. |
