Update README.md

README.md

@@ -8,6 +8,34 @@ This is the 650M parameter version of ESM-2, finetuned with QLoRA to predict bin
 No multiple sequence alignment or structure is required. The embeddings from this model can also be used in structural models. The model is trained on 
 approximately 12M protein sequences from UniProt, with an 80/20 train/test split. 
 
+## Metrics
+
+Based on preliminary tests on 20% samples of the train test split, the metrics are as follows:
+
+### Train Metrics
+
+```python
+'eval_loss': 0.05597764626145363, 
+'eval_accuracy': 0.9829392036087405, 
+'eval_precision': 0.5626191259397847, 
+'eval_recall': 0.9488112528941492, 
+'eval_f1': 0.7063763773187873, 
+'eval_auc': 0.9662524626230765, 
+'eval_mcc': 0.7235838533979579
+```
+
+### Test Metrics
+
+```python
+'eval_loss': 0.16281947493553162, 
+'eval_accuracy': 0.9569658774883986, 
+'eval_precision': 0.3209956738348438, 
+'eval_recall': 0.7883697002335764, 
+'eval_f1': 0.4562306866120791, 
+'eval_auc': 0.8746433990040084, 
+'eval_mcc': 0.48648765699020435
+```
+
 ## Using the Model
 
 ```python