Merge pull request #359 from MRXLT/0.2.0-cherry

cherry pick bug fix to 0.2.0

python/examples/bert/README.md

@@ -42,7 +42,7 @@ pip install paddle_serving_app
 ```
 执行
 ```
-cat data-c.txt | python bert_client.py 
+head data-c.txt | python bert_client.py --model bert_seq20_client/serving_client_conf.prototxt
 ```
 启动client读取data-c.txt中的数据进行预测，预测结果为文本的向量表示（由于数据较多，脚本中没有将输出进行打印），server端的地址在脚本中修改。
 

python/examples/bert/benchmark_batch.py

@@ -35,19 +35,28 @@ def single_func(idx, resource):
     dataset = []
     for line in fin:
         dataset.append(line.strip())
+    profile_flags = False
+    if os.environ["FLAGS_profile_client"]:
+        profile_flags = True
     if args.request == "rpc":
         reader = BertReader(vocab_file="vocab.txt", max_seq_len=20)
         fetch = ["pooled_output"]
         client = Client()
         client.load_client_config(args.model)
         client.connect([resource["endpoint"][idx % len(resource["endpoint"])]])
-        feed_batch = []
-        for bi in range(args.batch_size):
-            feed_batch.append(reader.process(dataset[bi]))
-
         start = time.time()
         for i in range(1000):
             if args.batch_size >= 1:
+                feed_batch = []
+                b_start = time.time()
+                for bi in range(args.batch_size):
+                    feed_batch.append(reader.process(dataset[bi]))
+                b_end = time.time()
+                if profile_flags:
+                    print("PROFILE\tpid:{}\tbert+pre_0:{} bert_pre_1:{}".format(
+                        os.getpid(),
+                        int(round(b_start * 1000000)),
+                        int(round(b_end * 1000000))))
                 result = client.batch_predict(
                     feed_batch=feed_batch, fetch=fetch)
             else:
@@ -62,7 +71,7 @@ def single_func(idx, resource):
 if __name__ == '__main__':
     multi_thread_runner = MultiThreadRunner()
     endpoint_list = [
-        "127.0.0.1:9295", "127.0.0.1:9296", "127.0.0.1:9297", "127.0.0.1:9298"
+        "127.0.0.1:9292", "127.0.0.1:9293", "127.0.0.1:9294", "127.0.0.1:9295"
     ]
     result = multi_thread_runner.run(single_func, args.thread,
                                      {"endpoint": endpoint_list})

python/examples/bert/bert_client.py

@@ -29,13 +29,13 @@ from paddle_serving_app import ChineseBertReader
 
 args = benchmark_args()
 
-reader = ChineseBertReader(max_seq_len=20)
+reader = ChineseBertReader({"max_seq_len": 20})
 fetch = ["pooled_output"]
 endpoint_list = ["127.0.0.1:9292"]
 client = Client()
 client.load_client_config(args.model)
 client.connect(endpoint_list)
 
-for line in fin:
+for line in sys.stdin:
     feed_dict = reader.process(line)
     result = client.predict(feed=feed_dict, fetch=fetch)

python/examples/bert/bert_web_service.py

@@ -32,8 +32,7 @@ bert_service = BertService(name="bert")
 bert_service.load()
 bert_service.load_model_config(sys.argv[1])
 gpu_ids = os.environ["CUDA_VISIBLE_DEVICES"]
-gpus = [int(x) for x in gpu_ids.split(",")]
-bert_service.set_gpus(gpus)
+bert_service.set_gpus(gpu_ids)
 bert_service.prepare_server(
     workdir="workdir", port=int(sys.argv[2]), device="gpu")
 bert_service.run_server()