diff --git a/generator/ctrl/training_utils/training.py b/generator/ctrl/training_utils/training.py
index e723040..734d4e3 100644
--- a/generator/ctrl/training_utils/training.py
+++ b/generator/ctrl/training_utils/training.py
@@ -133,7 +133,9 @@ model = tf.keras.Model(inputs=tokens, outputs=logits)
 
 # the loss function is a simple categorical crossentropy between the logits and the labels
 def loss(labels, logits):
-    return tf.keras.losses.sparse_categorical_crossentropy(labels, logits, from_logits=True)
+    loss = tf.keras.losses.sparse_categorical_crossentropy(labels, logits, from_logits=True)
+    loss = tf.Print(loss, [loss])
+    return loss
 
 
 # the optimizer is not used since this code only supports inference