Fix a bug in public doc.

PiperOrigin-RevId: 277124116

docs/_index.yaml

@@ -74,15 +74,15 @@ landing_page:
 
         # Wrap a Keras model for use with TFF.
         def model_fn():
-         model = tf.keras.models.Sequential([
-            tf.keras.layers.Dense(10, tf.nn.softmax, input_shape=(784,),
-                                  kernel_initializer='zeros')
-         ])
-         model.compile(
-             loss=tf.keras.losses.SparseCategoricalCrossentropy(),
-             optimizer=tf.keras.optimizers.SGD(0.1),
-             metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])
-         return tff.learning.from_compiled_keras_model(model, sample_batch)
+          model = tf.keras.models.Sequential([
+              tf.keras.layers.Dense(10, tf.nn.softmax, input_shape=(784,),
+                                    kernel_initializer='zeros')
+          ])
+          model.compile(
+              loss=tf.keras.losses.SparseCategoricalCrossentropy(),
+              optimizer=tf.keras.optimizers.SGD(0.1),
+              metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])
+          return tff.learning.from_compiled_keras_model(model, sample_batch)
 
         # Simulate a few rounds of training with the selected client devices.
         trainer = tff.learning.build_federated_averaging_process(model_fn)