go back to mdberta and increase sample size

app.py

@@ -43,11 +43,12 @@ dimension = len(id2label)
 st.write("dimension", dimension)
 
 student_model_config = teacher_model.config
-student_model_config.num_attention_heads = 6 
-student_model_config.num_hidden_layers = 4
+student_model_config.num_attention_heads = 8
+student_model_config.num_hidden_layers = 6
 student_model = DebertaV2ForTokenClassification.from_pretrained(
-    "microsoft/deberta-v3-small",
-    config=student_model_config, ignore_mismatched_sizes=True)
+    "microsoft/mdeberta-v3-base",
+    config=student_model_config)
+#    ignore_mismatched_sizes=True)
 print(student_model)
 print_trainable_parameters(student_model)
 
@@ -174,7 +175,7 @@ def distillation_loss(student_logits, teacher_logits, true_labels, temperature,
 # hyperparameters
 batch_size = 32
 lr = 1e-4
-num_epochs = 10
+num_epochs = 30
 temperature = 2.0
 alpha = 0.5