diff --git a/examples/text_classification/rnn/deploy/python/predict.py b/examples/text_classification/rnn/deploy/python/predict.py
index eeb4119b0eda4..61fc735c5a867 100644
--- a/examples/text_classification/rnn/deploy/python/predict.py
+++ b/examples/text_classification/rnn/deploy/python/predict.py
@@ -16,7 +16,6 @@
 
 import numpy as np
 import paddle
-from paddle import inference
 from paddlenlp.data import JiebaTokenizer, Stack, Tuple, Pad, Vocab
 from scipy.special import softmax
 
diff --git a/examples/text_classification/rnn/export_model.py b/examples/text_classification/rnn/export_model.py
index 477fd1a72143f..d61f82ba868ae 100644
--- a/examples/text_classification/rnn/export_model.py
+++ b/examples/text_classification/rnn/export_model.py
@@ -15,9 +15,10 @@
 import argparse
 
 import paddle
-import paddlenlp as ppnlp
 from paddlenlp.data import Vocab
 
+from model import BoWModel, BiLSTMAttentionModel, CNNModel, LSTMModel, GRUModel, RNNModel, SelfInteractiveAttention
+
 # yapf: disable
 parser = argparse.ArgumentParser(__doc__)
 parser.add_argument("--vocab_path", type=str, default="./senta_word_dict.txt", help="The path to vocabulary.")
@@ -56,7 +57,7 @@ def main():
             padding_idx=pad_token_id)
     elif network == 'bilstm_attn':
         lstm_hidden_size = 196
-        attention = SelfInteractiveAttention(hidden_size=2 * stm_hidden_size)
+        attention = SelfInteractiveAttention(hidden_size=2 * lstm_hidden_size)
         model = BiLSTMAttentionModel(
             attention_layer=attention,
             vocab_size=vocab_size,
diff --git a/examples/text_classification/rnn/predict.py b/examples/text_classification/rnn/predict.py
index ed8d8f361c307..f2a0e85e538d5 100644
--- a/examples/text_classification/rnn/predict.py
+++ b/examples/text_classification/rnn/predict.py
@@ -15,7 +15,6 @@
 
 import paddle
 import paddle.nn.functional as F
-import paddlenlp as ppnlp
 from paddlenlp.data import JiebaTokenizer, Stack, Tuple, Pad, Vocab
 
 from model import BoWModel, BiLSTMAttentionModel, CNNModel, LSTMModel, GRUModel, RNNModel, SelfInteractiveAttention
@@ -102,7 +101,7 @@ def predict(model, data, label_map, batch_size=1, pad_token_id=0):
             padding_idx=pad_token_id)
     elif network == 'bilstm_attn':
         lstm_hidden_size = 196
-        attention = SelfInteractiveAttention(hidden_size=2 * stm_hidden_size)
+        attention = SelfInteractiveAttention(hidden_size=2 * lstm_hidden_size)
         model = BiLSTMAttentionModel(
             attention_layer=attention,
             vocab_size=vocab_size,