get error when predict about bert-for-tf2 HOT 4 CLOSED

diff --git a/BERT.py b/BERT.py
index 722d4bb..1a3b7f4 100644
--- a/BERT.py
+++ b/BERT.py
@@ -39,7 +39,7 @@ def BERT(max_seq_len = 128, bert_model_dir = 'models/chinese_L-12_H-768_A-12', d
     output = bert([input_token_ids, input_segment_ids]);
     # create model containing only bert layer
     model = tf.keras.Model(inputs = [input_token_ids, input_segment_ids], outputs = output);
-    model.build(input_shape = [(None, max_seq_len), (None, max_seq_len)]);
+    #model.build(input_shape = [(None, max_seq_len), (None, max_seq_len)]);
     # freeze_bert_layers
     freeze_bert_layers(bert);
     # load bert layer weights
diff --git a/Predictor.py b/Predictor.py
index 2c7e6da..dd1dcfe 100644
--- a/Predictor.py
+++ b/Predictor.py
@@ -151,15 +151,20 @@ class Predictor(object):
     def _classify(self, inputs, mask, training = None):
 
         # the first element of output sequence.
-        outputs = self.bert(inputs, mask, training);
+        outputs = self.bert.predict(inputs)
+
+        cls_input = tf.keras.Input((128,768,), dtype=tf.float32)
         # first_token.shape = (batch, hidden_size)
-        first_token = tf.keras.layers.Lambda(lambda seq: seq[:, 0, :])(outputs);
-        first_token = tf.keras.Dropout(rate = 0.5)(first_token);
+        first_token = tf.keras.layers.Lambda(lambda seq: seq[:, 0, :])(cls_input);
+        first_token = tf.keras.layers.Dropout(rate = 0.5)(first_token);
         pooled_output = tf.keras.layers.Dense(units = first_token.shape[-1], activation = tf.math.tanh)(first_token);
         dropout = tf.keras.layers.Dropout(rate = 0.5)(pooled_output);
         logits = tf.keras.layers.Dense(units = 2, activation = tf.nn.softmax)(dropout);
 
-        return logits;
+        model = tf.keras.models.Model(inputs=cls_input, outputs=logits)
+        model.build(input_shape = (None, 128, 768))
+
+        return model.predict(outputs)
 
     def predict(self, question, answer):
 
@@ -167,6 +172,10 @@ class Predictor(object):
         input_ids = tf.constant(input_ids, dtype = tf.int32);
         input_mask = tf.constant(input_mask, dtype = tf.int32);
         segment_ids = tf.constant(segment_ids, dtype = tf.int32);
+
+        input_ids = tf.expand_dims(input_ids, axis=0)
+        segment_ids = tf.expand_dims(segment_ids, axis=0)
+        
         logits = self._classify([input_ids, segment_ids], input_mask, False);
         probabilities = tf.nn.softmax(logits);
         out = tf.math.argmax(probabilities);
@@ -174,6 +183,7 @@ class Predictor(object):
 
 if __name__ == "__main__":
 
+    tf.enable_eager_execution()
     assert tf.executing_eagerly();
     predictor = Predictor();
     print(predictor.predict('今天天气如何？','感觉很不错！'));