fix bert demo

core/general-client/src/general_model.cpp

@@ -132,13 +132,12 @@ int PredictorClient::create_predictor() {
   _api.thrd_initialize();
 }
 
-int PredictorClient::predict(
-    const std::vector<std::vector<float>>& float_feed,
-    const std::vector<std::string>& float_feed_name,
-    const std::vector<std::vector<int64_t>>& int_feed,
-    const std::vector<std::string>& int_feed_name,
-    const std::vector<std::string>& fetch_name,
-    PredictorRes & predict_res) { // NOLINT
+int PredictorClient::predict(const std::vector<std::vector<float>> &float_feed,
+                             const std::vector<std::string> &float_feed_name,
+                             const std::vector<std::vector<int64_t>> &int_feed,
+                             const std::vector<std::string> &int_feed_name,
+                             const std::vector<std::string> &fetch_name,
+                             PredictorRes &predict_res) {  // NOLINT
   predict_res._int64_map.clear();
   predict_res._float_map.clear();
   Timer timeline;
@@ -218,6 +217,7 @@ int PredictorClient::predict(
       VLOG(2) << "fetch name: " << name;
       if (_fetch_name_to_type[name] == 0) {
         int len = res.insts(0).tensor_array(idx).int64_data_size();
+        VLOG(2) << "fetch tensor : " << name << " type: int64 len : " << len;
         predict_res._int64_map[name].resize(1);
         predict_res._int64_map[name][0].resize(len);
         for (int i = 0; i < len; ++i) {
@@ -226,6 +226,7 @@ int PredictorClient::predict(
         }
       } else if (_fetch_name_to_type[name] == 1) {
         int len = res.insts(0).tensor_array(idx).float_data_size();
+        VLOG(2) << "fetch tensor : " << name << " type: float32 len : " << len;
         predict_res._float_map[name].resize(1);
         predict_res._float_map[name][0].resize(len);
         for (int i = 0; i < len; ++i) {
@@ -342,7 +343,7 @@ std::vector<std::vector<std::vector<float>>> PredictorClient::batch_predict(
     }
 
     VLOG(2) << "batch [" << bi << "] "
-            << "itn feed value prepared";
+            << "int feed value prepared";
   }
 
   int64_t preprocess_end = timeline.TimeStampUS();

python/examples/bert/bert_client.py

@@ -120,7 +120,6 @@ class BertService():
 
 
 def test():
-
     bc = BertService(
         model_name='bert_chinese_L-12_H-768_A-12',
         max_seq_len=20,
@@ -130,9 +129,13 @@ def test():
     config_file = './serving_client_conf/serving_client_conf.prototxt'
     fetch = ["pooled_output"]
     bc.load_client(config_file, server_addr)
-    batch_size = 4
+    batch_size = 1
     batch = []
     for line in sys.stdin:
+        if batch_size == 1:
+            result = bc.run_general([[line.strip()]], fetch)
+            print(result)
+        else:
             if len(batch) < batch_size:
                 batch.append([line.strip()])
             else:
@@ -140,6 +143,11 @@ def test():
                 batch = []
                 for r in result:
                     print(r)
+    if len(batch) > 0:
+        result = bc.run_batch_general(batch, fetch)
+        batch = []
+        for r in result:
+            print(r)
 
 
 if __name__ == '__main__':

python/examples/bert/bert_server.py

@@ -31,8 +31,6 @@ op_seq_maker.add_op(general_response_op)
 server = Server()
 server.set_op_sequence(op_seq_maker.get_op_sequence())
 server.set_num_threads(4)
-server.set_local_bin(
-    "~/github/Serving/build_server/core/general-server/serving")
 
 server.load_model_config(sys.argv[1])
 port = int(sys.argv[2])

python/examples/bert/get_data.sh

+wget https://paddle-serving.bj.bcebos.com/bert_example/data-c.txt --no-check-certificate