Filter 1/11 of the data:

app.py

@@ -61,7 +61,8 @@ if torch.cuda.is_available():
 # Load data. 
 raw_dataset = load_dataset("ai4privacy/pii-masking-400k", split='train')
 raw_dataset = raw_dataset.filter(lambda example: example["language"].startswith("en"))
-raw_dataset = raw_dataset.select(range(2000)) 
+#raw_dataset = raw_dataset.select(range(2000))
+raw_dataset = raw_dataset.filter(lambda example, idx: idx % 11 == 0, with_indices=True)
 raw_dataset = raw_dataset.train_test_split(test_size=0.2)
 print(raw_dataset)
 print(raw_dataset.column_names)
@@ -177,7 +178,7 @@ def distillation_loss(student_logits, teacher_logits, true_labels, temperature,
 # hyperparameters
 batch_size = 32
 lr = 1e-4
-num_epochs = 300
+num_epochs = 50
 temperature = 2.0
 alpha = 0.5