| license: mit | |
| library_name: transformers | |
| # sweep-lr5e5-wd001 | |
| This model was selected as the best checkpoint from a hyperparameter sweep. | |
| ## Model Architecture | |
| <div align="center"> | |
| <img src="figures/arch_diagram.png" width="70%" alt="Architecture Diagram" /> | |
| </div> | |
| ## Training Configuration | |
| | Parameter | Value | | |
| |---|---| | |
| | learning_rate | 5e-05 | | |
| | weight_decay | 0.001 | | |
| | epochs | 30 | | |
| | batch_size | 16 | | |
| | model_arch | deberta-v3-base | | |
| ## Evaluation Metrics | |
| | Metric | Value | | |
| |---|---| | |
| | val_loss | 0.198 | | |
| | val_accuracy | 0.934 | | |
| | f1_score | 0.921 | | |
| | inference_latency_ms | 14.2 | | |
| ## Training Loss Curve | |
| <div align="center"> | |
| <img src="figures/loss_curve.png" width="80%" alt="Training Loss Curve" /> | |
| </div> | |
| ## Usage | |
| ```python | |
| from transformers import AutoModelForSequenceClassification, AutoTokenizer | |
| model = AutoModelForSequenceClassification.from_pretrained("BestSweepModel-TestRepo") | |
| tokenizer = AutoTokenizer.from_pretrained("BestSweepModel-TestRepo") | |
| ``` | |
| ## License | |
| This model is released under the [MIT License](LICENSE). | |