Update MODEL_CARD.md with metrics for all runs (3 decimal places)

MODEL_CARD.md

@@ -0,0 +1,39 @@
+---
+license: apache-2.0
+library_name: transformers
+---
+# BestSweepModel
+
+This model was selected from a hyperparameter sweep over 5 configurations. The run with the lowest validation loss was chosen as the final model.
+
+## Training Configuration
+
+All runs used the same base architecture (BERT-base) and training setup, varying only the learning rate and weight decay.
+
+## Sweep Results
+
+| Run | Learning Rate | Weight Decay | Train Loss | Val Loss | Val Accuracy | Val F1 |
+|---|---|---|---|---|---|---|
+| run_lr1e-3_wd0.01 | 0.001000 | 0.0100 | 0.082 | 0.347 | 0.882 | 0.871 |
+| run_lr5e-4_wd0.001 | 0.000500 | 0.0010 | 0.119 | 0.285 | 0.904 | 0.898 |
+| run_lr1e-4_wd0.1 | 0.000100 | 0.1000 | 0.451 | 0.523 | 0.791 | 0.776 |
+| run_lr3e-4_wd0.01 | 0.000300 | 0.0100 | 0.098 | 0.261 | 0.917 | 0.911 |
+| run_lr2e-4_wd0.05 | 0.000200 | 0.0500 | 0.203 | 0.312 | 0.895 | 0.887 |
+
+**Best run**: run_lr3e-4_wd0.01 (lowest val_loss = 0.261)
+
+## Intended Use
+
+This model is intended for text classification tasks. It was trained on a standard benchmark dataset.
+
+## How to Use
+
+```python
+from transformers import AutoModel, AutoTokenizer
+model = AutoModel.from_pretrained("BestSweepModel-TestRepo")
+tokenizer = AutoTokenizer.from_pretrained("BestSweepModel-TestRepo")
+```
+
+## License
+
+Apache 2.0