diff --git a/15_processing_sequences_using_rnns_and_cnns.ipynb b/15_processing_sequences_using_rnns_and_cnns.ipynb
index f6e2e5b..0d1face 100644
--- a/15_processing_sequences_using_rnns_and_cnns.ipynb
+++ b/15_processing_sequences_using_rnns_and_cnns.ipynb
@@ -3988,7 +3988,7 @@
     }
    ],
    "source": [
-    "model.save(\"my_sketchrnn\")"
+    "model.save(\"my_sketchrnn\", save_format=\"tf\")"
    ]
   },
   {
@@ -4638,7 +4638,7 @@
     }
    ],
    "source": [
-    "model.save(\"my_bach_model.h5\")\n",
+    "model.save(\"my_bach_model\", save_format=\"tf\")\n",
     "model.evaluate(test_set)"
    ]
   },
diff --git a/19_training_and_deploying_at_scale.ipynb b/19_training_and_deploying_at_scale.ipynb
index ead10a4..352b3db 100644
--- a/19_training_and_deploying_at_scale.ipynb
+++ b/19_training_and_deploying_at_scale.ipynb
@@ -229,7 +229,7 @@
     "model_name = \"my_mnist_model\"\n",
     "model_version = \"0001\"\n",
     "model_path = Path(model_name) / model_version\n",
-    "model.save(model_path)"
+    "model.save(model_path, save_format=\"tf\")"
    ]
   },
   {
@@ -706,7 +706,7 @@
    "source": [
     "model_version = \"0002\"\n",
     "model_path = Path(model_name) / model_version\n",
-    "model.save(model_path)"
+    "model.save(model_path, save_format=\"tf\")"
    ]
   },
   {
@@ -1830,7 +1830,7 @@
    "source": [
     "# extra code – shows that saving a model does not preserve its distribution\n",
     "#              strategy\n",
-    "model.save(\"my_mirrored_model\")\n",
+    "model.save(\"my_mirrored_model\", save_format=\"tf\")\n",
     "model = tf.keras.models.load_model(\"my_mirrored_model\")\n",
     "type(model.weights[0])"
    ]
@@ -2151,10 +2151,10 @@
     "model.fit(X_train, y_train, validation_data=(X_valid, y_valid), epochs=10)\n",
     "\n",
     "if resolver.task_id == 0:  # the chief saves the model to the right location\n",
-    "    model.save(\"my_mnist_multiworker_model\")\n",
+    "    model.save(\"my_mnist_multiworker_model\", save_format=\"tf\")\n",
     "else:\n",
     "    tmpdir = tempfile.mkdtemp()  # other workers save to a temporary directory\n",
-    "    model.save(tmpdir)\n",
+    "    model.save(tmpdir, save_format=\"tf\")\n",
     "    tf.io.gfile.rmtree(tmpdir)  # and we can delete this directory at the end!"
    ]
   },
@@ -2319,7 +2319,7 @@
     "\n",
     "model.fit(X_train, y_train, validation_data=(X_valid, y_valid), epochs=10,\n",
     "          callbacks=callbacks)\n",
-    "model.save(model_dir)"
+    "model.save(model_dir, save_format=\"tf\")"
    ]
   },
   {
@@ -2471,7 +2471,7 @@
     "model = build_model(args)\n",
     "history = model.fit(X_train, y_train, validation_data=(X_valid, y_valid),\n",
     "                    epochs=10, callbacks=callbacks)\n",
-    "model.save(model_dir)  # extra code\n",
+    "model.save(model_dir, save_format=\"tf\")  # extra code\n",
     "\n",
     "import hypertune\n",
     "\n",
@@ -2771,7 +2771,7 @@
     "if tuner_id == \"chief\":\n",
     "    best_hp = hyperband_tuner.get_best_hyperparameters()[0]\n",
     "    best_model = hyperband_tuner.hypermodel.build(best_hp)\n",
-    "    best_model.save(os.getenv(\"AIP_MODEL_DIR\"))"
+    "    best_model.save(os.getenv(\"AIP_MODEL_DIR\"), save_format=\"tf\")"
    ]
   },
   {