diff --git a/tensorflow/lite/experimental/examples/lstm/rnn_cell.py b/tensorflow/lite/experimental/examples/lstm/rnn_cell.py
index 44446fb509d..6fdd16d079c 100644
--- a/tensorflow/lite/experimental/examples/lstm/rnn_cell.py
+++ b/tensorflow/lite/experimental/examples/lstm/rnn_cell.py
@@ -319,6 +319,8 @@ class TFLiteLSTMCell(rnn_cell_impl.LayerRNNCell):
     else:
       bias_initializer = init_ops.zeros_initializer(dtype=self.dtype)
 
+    forget_bias_initializer = init_ops.constant_initializer(self._forget_bias)
+
     self.input_to_input_w = add_variable_wrapped(
         "input_to_input_w", input_weight_shape, weight_initializer, 1,
         maybe_partitioner)
@@ -346,8 +348,9 @@ class TFLiteLSTMCell(rnn_cell_impl.LayerRNNCell):
 
     self.input_bias = add_variable_wrapped(
         "input_bias", bias_shape, bias_initializer, 12, maybe_partitioner)
-    self.forget_bias = add_variable_wrapped(
-        "forget_bias", bias_shape, bias_initializer, 13, maybe_partitioner)
+    self.forget_bias = add_variable_wrapped("forget_bias", bias_shape,
+                                            forget_bias_initializer, 13,
+                                            maybe_partitioner)
     self.cell_bias = add_variable_wrapped(
         "cell_bias", bias_shape, bias_initializer, 14, maybe_partitioner)
     self.output_bias = add_variable_wrapped(
@@ -473,12 +476,10 @@ class TFLiteLSTMCell(rnn_cell_impl.LayerRNNCell):
     # Diagonal connections
     if self._use_peepholes:
       c = (
-          sigmoid(f + self._forget_bias + self._w_f_diag * c_prev) * c_prev +
+          sigmoid(f + self._w_f_diag * c_prev) * c_prev +
           sigmoid(i + self._w_i_diag * c_prev) * self._activation(j))
     else:
-      c = (
-          sigmoid(f + self._forget_bias) * c_prev +
-          sigmoid(i) * self._activation(j))
+      c = (sigmoid(f) * c_prev + sigmoid(i) * self._activation(j))
 
     if self._cell_clip is not None:
       # pylint: disable=invalid-unary-operand-type
diff --git a/tensorflow/lite/experimental/examples/lstm/unidirectional_sequence_lstm_test.py b/tensorflow/lite/experimental/examples/lstm/unidirectional_sequence_lstm_test.py
index e40a3e5d3c1..e29c7510034 100644
--- a/tensorflow/lite/experimental/examples/lstm/unidirectional_sequence_lstm_test.py
+++ b/tensorflow/lite/experimental/examples/lstm/unidirectional_sequence_lstm_test.py
@@ -56,9 +56,9 @@ class UnidirectionalSequenceLstmTest(test_util.TensorFlowTestCase):
   def buildLstmLayer(self):
     return tf.keras.layers.StackedRNNCells([
         tf.lite.experimental.nn.TFLiteLSTMCell(
-            self.num_units, use_peepholes=True, forget_bias=0, name="rnn1"),
+            self.num_units, use_peepholes=True, forget_bias=1.0, name="rnn1"),
         tf.lite.experimental.nn.TFLiteLSTMCell(
-            self.num_units, num_proj=8, forget_bias=0, name="rnn2"),
+            self.num_units, num_proj=8, forget_bias=1.0, name="rnn2"),
         tf.lite.experimental.nn.TFLiteLSTMCell(
             self.num_units // 2,
             use_peepholes=True,
@@ -66,7 +66,7 @@ class UnidirectionalSequenceLstmTest(test_util.TensorFlowTestCase):
             forget_bias=0,
             name="rnn3"),
         tf.lite.experimental.nn.TFLiteLSTMCell(
-            self.num_units, forget_bias=0, name="rnn4")
+            self.num_units, forget_bias=1.0, name="rnn4")
     ])
 
   def buildModel(self, lstm_layer, is_dynamic_rnn):