diff --git a/paddlespeech/s2t/models/wav2vec2/wav2vec2_ASR.py b/paddlespeech/s2t/models/wav2vec2/wav2vec2_ASR.py
index baa7392eb4f0d64e8aeb9fe37db05b4595f1a837..7468fdce01d7df702b55a3a969d497fbebc14fd2 100755
--- a/paddlespeech/s2t/models/wav2vec2/wav2vec2_ASR.py
+++ b/paddlespeech/s2t/models/wav2vec2/wav2vec2_ASR.py
@@ -12,7 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from collections import defaultdict
-from turtle import Turtle
 from typing import Dict
 from typing import List
 from typing import Tuple
diff --git a/paddlespeech/s2t/training/gradclip.py b/paddlespeech/s2t/training/gradclip.py
index be6fcf5899f4f17cb6ee676b3cf0affd042bcd5c..06587c749b5a965e6881b34236f0d5b77f118ebd 100644
--- a/paddlespeech/s2t/training/gradclip.py
+++ b/paddlespeech/s2t/training/gradclip.py
@@ -56,15 +56,15 @@ class ClipGradByGlobalNormWithLog(paddle.nn.ClipGradByGlobalNorm):
         if len(sum_square_list) == 0:
             return params_grads
 
-        global_norm_var = layers.concat(sum_square_list)
+        global_norm_var = paddle.concat(sum_square_list)
         global_norm_var = paddle.sum(global_norm_var)
         global_norm_var = paddle.sqrt(global_norm_var)
 
         # debug log
         logger.debug(f"Grad Global Norm: {float(global_norm_var)}!!!!")
 
-        max_global_norm = layers.fill_constant(
-            shape=[1], dtype=global_norm_var.dtype, value=self.clip_norm)
+        max_global_norm = paddle.full(
+            shape=[1], dtype=global_norm_var.dtype, fill_value=self.clip_norm)
         clip_var = paddle.divide(
             x=max_global_norm,
             y=paddle.maximum(x=global_norm_var, y=max_global_norm))