Update app.py

app.py

@@ -26,7 +26,10 @@ tgt_index_to_word[0] = ""  # padding
 # Note: this matches the training architecture above.
 encoder_inputs = model.input[0]
 encoder_emb_output = model.layers[2](encoder_inputs)
-encoder_lstm_output, state_h_enc, state_c_enc = model.layers[3](encoder_emb_output)
+encoder_lstm = model.layers[3]
+encoder_outputs = encoder_lstm(encoder_emb_output)
+state_h_enc = encoder_outputs[1]
+state_c_enc = encoder_outputs[2]
 encoder_model = tf.keras.Model(encoder_inputs, [state_h_enc, state_c_enc])
 
 
@@ -39,9 +42,12 @@ decoder_dense = model.layers[6]
 dec_state_input_h = tf.keras.Input(shape=(decoder_lstm.units,))
 dec_state_input_c = tf.keras.Input(shape=(decoder_lstm.units,))
 dec_emb2 = dec_emb_layer(decoder_inputs)
-dec_outputs, state_h_dec, state_c_dec = decoder_lstm(
+decoder_lstm_outputs = decoder_lstm(
     dec_emb2, initial_state=[dec_state_input_h, dec_state_input_c]
 )
+dec_outputs = decoder_lstm_outputs[0]
+state_h_dec = decoder_lstm_outputs[1]
+state_c_dec = decoder_lstm_outputs[2]
 dec_outputs = decoder_dense(dec_outputs)
 decoder_model = tf.keras.Model(
     [decoder_inputs, dec_state_input_h, dec_state_input_c],