diff --git a/site/en/tutorials/keras/overfit_and_underfit.ipynb b/site/en/tutorials/keras/overfit_and_underfit.ipynb
index 87c80de937..29fbfc49c8 100644
--- a/site/en/tutorials/keras/overfit_and_underfit.ipynb
+++ b/site/en/tutorials/keras/overfit_and_underfit.ipynb
@@ -543,10 +543,10 @@
         "  model.summary()\n",
         "\n",
         "  history = model.fit(\n",
-        "    train_ds,\n",
+        "    train_ds.map(lambda x, y: (x, tf.expand_dims(y, axis=-1))),\n",
         "    steps_per_epoch = STEPS_PER_EPOCH,\n",
         "    epochs=max_epochs,\n",
-        "    validation_data=validate_ds,\n",
+        "    validation_data=validate_ds.map(lambda x, y: (x, tf.expand_dims(y, axis=-1))),\n",
         "    callbacks=get_callbacks(name),\n",
         "    verbose=0)\n",
         "  return history"
@@ -977,7 +977,7 @@
       "source": [
         "`l2(0.001)` means that every coefficient in the weight matrix of the layer will add `0.001 * weight_coefficient_value**2` to the total **loss** of the network.\n",
         "\n",
-        "That is why we're monitoring the `binary_crossentropy` directly. Because it doesn't have this regularization component mixed in.\n",
+        "That is why you need to monitor the `binary_crossentropy` directly. Because it doesn't have this regularization component mixed in.\n",
         "\n",
         "So, that same `\"Large\"` model with an `L2` regularization penalty performs much better:\n"
       ]
@@ -1228,10 +1228,9 @@
     }
   ],
   "metadata": {
-    "accelerator": "GPU",
     "colab": {
       "name": "overfit_and_underfit.ipynb",
-            "toc_visible": true
+      "toc_visible": true
     },
     "kernelspec": {
       "display_name": "Python 3",