diff --git a/python/examples/bert/README.md b/python/examples/bert/README.md index 612ba8beb979f5a1ee9d1a8052f73169191f4fd7..acbd3d05cd4197e0e74f9ab04b141488145094ec 100644 --- a/python/examples/bert/README.md +++ b/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。 diff --git a/python/examples/bert/benchmark_batch.py b/python/examples/bert/benchmark_batch.py index e0f677146a47c0366a1bbafe9eff049e2671a617..265521d484259b0e6ea2b182dbf61e2a5cf43b8d 100644 --- a/python/examples/bert/benchmark_batch.py +++ b/python/examples/bert/benchmark_batch.py @@ -35,21 +35,29 @@ 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: - result = client.batch_predict( - feed_batch=feed_batch, fetch=fetch) + 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.predict(feed_batch=feed_batch, fetch=fetch) else: print("unsupport batch size {}".format(args.batch_size)) @@ -62,7 +70,7 @@ 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}) diff --git a/python/examples/bert/bert_client.py b/python/examples/bert/bert_client.py index cceb07c7334aa0cc9b9c347ae53351b7a6fa3513..51364c6745731017b31923d246990497115dc780 100644 --- a/python/examples/bert/bert_client.py +++ b/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) diff --git a/python/examples/bert/bert_web_service.py b/python/examples/bert/bert_web_service.py index e9a222db958bdd0df18e7f15908b722c2b4b71a9..04462ca3b16fecf818aadad63b4f67a8d97014fd 100644 --- a/python/examples/bert/bert_web_service.py +++ b/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() diff --git a/python/examples/criteo_ctr/benchmark_batch.py b/python/examples/criteo_ctr/benchmark_batch.py index 47b63a6ade0c21bdc82a5c67d65b39ffc614e06c..ab67507355d0eba187d47ec9577eb5a3eda5dc46 100644 --- a/python/examples/criteo_ctr/benchmark_batch.py +++ b/python/examples/criteo_ctr/benchmark_batch.py @@ -55,8 +55,7 @@ def single_func(idx, resource): for i in range(1, 27): feed_dict["sparse_{}".format(i - 1)] = data[0][i] feed_batch.append(feed_dict) - result = client.batch_predict( - feed_batch=feed_batch, fetch=fetch) + result = client.predict(feed_batch=feed_batch, fetch=fetch) else: print("unsupport batch size {}".format(args.batch_size)) diff --git a/python/examples/imagenet/benchmark_batch.py b/python/examples/imagenet/benchmark_batch.py index 7477100971d1a098bb075c94f75490c16b53b862..915544d4f6f9d4636b14f4be92ad75ba08013389 100644 --- a/python/examples/imagenet/benchmark_batch.py +++ b/python/examples/imagenet/benchmark_batch.py @@ -50,8 +50,7 @@ def single_func(idx, resource): img = reader.process_image(img_list[i]) img = img.reshape(-1) feed_batch.append({"image": img}) - result = client.batch_predict( - feed_batch=feed_batch, fetch=fetch) + result = client.predict(feed_batch=feed_batch, fetch=fetch) else: print("unsupport batch size {}".format(args.batch_size)) diff --git a/python/examples/imdb/benchmark_batch.py b/python/examples/imdb/benchmark_batch.py index 302d63352ca20bf7e455ad1a66ead22f63dbe846..3ac52ec5472ff97f5d273dc230494223f3a71907 100644 --- a/python/examples/imdb/benchmark_batch.py +++ b/python/examples/imdb/benchmark_batch.py @@ -42,7 +42,7 @@ def single_func(idx, resource): for bi in range(args.batch_size): word_ids, label = imdb_dataset.get_words_and_label(line) feed_batch.append({"words": word_ids}) - result = client.batch_predict( + result = client.predict( feed_batch=feed_batch, fetch=["prediction"]) else: print("unsupport batch size {}".format(args.batch_size))