From 814a66021f3f4135208ce0b7ebd6f7c6797fe25e Mon Sep 17 00:00:00 2001 From: barrierye Date: Thu, 9 Apr 2020 11:09:19 +0800 Subject: [PATCH] download bert model from bos in CI script --- python/examples/bert/README.md | 23 ++++++++------ python/examples/bert/README_CN.md | 23 ++++++++------ python/examples/bert/bert_client.py | 2 +- python/examples/bert/bert_web_service.py | 2 +- tools/serving_build.sh | 38 ++++++++++++++++++------ 5 files changed, 59 insertions(+), 29 deletions(-) diff --git a/python/examples/bert/README.md b/python/examples/bert/README.md index bd2af745..96b659ae 100644 --- a/python/examples/bert/README.md +++ b/python/examples/bert/README.md @@ -15,12 +15,17 @@ pip install paddlehub run ``` -python prepare_model.py 20 +python prepare_model.py 128 ``` -the 20 in the command above means max_seq_len in BERT model, which is the length of sample after preprocessing. -the config file and model file for server side are saved in the folder bert_seq20_model. -the config file generated for client side is saved in the folder bert_seq20_client. +the 128 in the command above means max_seq_len in BERT model, which is the length of sample after preprocessing. +the config file and model file for server side are saved in the folder bert_seq128_model. +the config file generated for client side is saved in the folder bert_seq128_client. +You can also download the above model from BOS(max_seq_len=128). After decompression, the config file and model file for server side are stored in the bert_chinese_L-12_H-768_A-12_model folder, and the config file generated for client side is stored in the bert_chinese_L-12_H-768_A-12_client folder: +```shell +wget https://paddle-serving.bj.bcebos.com/paddle_hub_models/text/SemanticModel/bert_chinese_L-12_H-768_A-12.tar.gz +tar -xzf bert_chinese_L-12_H-768_A-12.tar.gz +``` ### Getting Dict and Sample Dataset @@ -32,11 +37,11 @@ this script will download Chinese Dictionary File vocab.txt and Chinese Sample D ### RPC Inference Service Run ``` -python -m paddle_serving_server.serve --model bert_seq20_model/ --port 9292 #cpu inference service +python -m paddle_serving_server.serve --model bert_seq128_model/ --port 9292 #cpu inference service ``` Or ``` -python -m paddle_serving_server_gpu.serve --model bert_seq20_model/ --port 9292 --gpu_ids 0 #launch gpu inference service at GPU 0 +python -m paddle_serving_server_gpu.serve --model bert_seq128_model/ --port 9292 --gpu_ids 0 #launch gpu inference service at GPU 0 ``` ### RPC Inference @@ -47,7 +52,7 @@ pip install paddle_serving_app ``` Run ``` -head data-c.txt | python bert_client.py --model bert_seq20_client/serving_client_conf.prototxt +head data-c.txt | python bert_client.py --model bert_seq128_client/serving_client_conf.prototxt ``` the client reads data from data-c.txt and send prediction request, the prediction is given by word vector. (Due to massive data in the word vector, we do not print it). @@ -58,7 +63,7 @@ the client reads data from data-c.txt and send prediction request, the predictio ``` set environmental variable to specify which gpus are used, the command above means gpu 0 and gpu 1 is used. ``` - python bert_web_service.py bert_seq20_model/ 9292 #launch gpu inference service + python bert_web_service.py bert_seq128_model/ 9292 #launch gpu inference service ``` ### HTTP Inference @@ -75,7 +80,7 @@ GPU：GPU V100 * 1 CUDA/cudnn Version：CUDA 9.2，cudnn 7.1.4 -In the test, 10 thousand samples in the sample data are copied into 100 thousand samples. Each client thread sends a sample of the number of threads. The batch size is 1, the max_seq_len is 20, and the time unit is seconds. +In the test, 10 thousand samples in the sample data are copied into 100 thousand samples. Each client thread sends a sample of the number of threads. The batch size is 1, the max_seq_len is 20(not 128 as described above), and the time unit is seconds. When the number of client threads is 4, the prediction speed can reach 432 samples per second. Because a single GPU can only perform serial calculations internally, increasing the number of client threads can only reduce the idle time of the GPU. Therefore, after the number of threads reaches 4, the increase in the number of threads does not improve the prediction speed. diff --git a/python/examples/bert/README_CN.md b/python/examples/bert/README_CN.md index 305010ba..bba8d8d2 100644 --- a/python/examples/bert/README_CN.md +++ b/python/examples/bert/README_CN.md @@ -13,11 +13,16 @@ pip install paddlehub ``` 执行 ``` -python prepare_model.py 20 +python prepare_model.py 128 +``` +参数128表示BERT模型中的max_seq_len，即预处理后的样本长度。 +生成server端配置文件与模型文件，存放在bert_seq128_model文件夹。 +生成client端配置文件，存放在bert_seq128_client文件夹。 +您也可以从bos上直接下载上述模型（max_seq_len=128），解压后server端配置文件与模型文件存放在bert_chinese_L-12_H-768_A-12_model文件夹，client端配置文件存放在bert_chinese_L-12_H-768_A-12_client文件夹： +```shell +wget https://paddle-serving.bj.bcebos.com/paddle_hub_models/text/SemanticModel/bert_chinese_L-12_H-768_A-12.tar.gz +tar -xzf bert_chinese_L-12_H-768_A-12.tar.gz ``` -参数20表示BERT模型中的max_seq_len，即预处理后的样本长度。 -生成server端配置文件与模型文件，存放在bert_seq20_model文件夹 -生成client端配置文件，存放在bert_seq20_client文件夹 ### 获取词典和样例数据 @@ -29,11 +34,11 @@ sh get_data.sh ### 启动RPC预测服务 执行 ``` -python -m paddle_serving_server.serve --model bert_seq20_model/ --port 9292 #启动cpu预测服务 +python -m paddle_serving_server.serve --model bert_seq128_model/ --port 9292 #启动cpu预测服务 ``` 或者 ``` -python -m paddle_serving_server_gpu.serve --model bert_seq20_model/ --port 9292 --gpu_ids 0 #在gpu 0上启动gpu预测服务 +python -m paddle_serving_server_gpu.serve --model bert_seq128_model/ --port 9292 --gpu_ids 0 #在gpu 0上启动gpu预测服务 ``` ### 执行预测 @@ -44,7 +49,7 @@ pip install paddle_serving_app ``` 执行 ``` -head data-c.txt | python bert_client.py --model bert_seq20_client/serving_client_conf.prototxt +head data-c.txt | python bert_client.py --model bert_seq128_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。 @@ -54,7 +59,7 @@ head data-c.txt | python bert_client.py --model bert_seq20_client/serving_client ``` 通过环境变量指定gpu预测服务使用的gpu，示例中指定索引为0和1的两块gpu ``` - python bert_web_service.py bert_seq20_model/ 9292 #启动gpu预测服务 + python bert_web_service.py bert_seq128_model/ 9292 #启动gpu预测服务 ``` ### 执行预测 @@ -70,7 +75,7 @@ curl -H "Content-Type:application/json" -X POST -d '{"words": "hello", "fetch":[ 环境：CUDA 9.2，cudnn 7.1.4 -测试中将样例数据中的1W个样本复制为10W个样本，每个client线程发送线程数分之一个样本，batch size为1，max_seq_len为20，时间单位为秒. +测试中将样例数据中的1W个样本复制为10W个样本，每个client线程发送线程数分之一个样本，batch size为1，max_seq_len为20(而非上面的128)，时间单位为秒. 在client线程数为4时，预测速度可以达到432样本每秒。 由于单张GPU内部只能串行计算，client线程增多只能减少GPU的空闲时间，因此在线程数达到4之后，线程数增多对预测速度没有提升。 diff --git a/python/examples/bert/bert_client.py b/python/examples/bert/bert_client.py index 51364c67..b33a80d8 100644 --- a/python/examples/bert/bert_client.py +++ b/python/examples/bert/bert_client.py @@ -29,7 +29,7 @@ from paddle_serving_app import ChineseBertReader args = benchmark_args() -reader = ChineseBertReader({"max_seq_len": 20}) +reader = ChineseBertReader({"max_seq_len": 128}) fetch = ["pooled_output"] endpoint_list = ["127.0.0.1:9292"] client = Client() diff --git a/python/examples/bert/bert_web_service.py b/python/examples/bert/bert_web_service.py index 04462ca3..e22e379d 100644 --- a/python/examples/bert/bert_web_service.py +++ b/python/examples/bert/bert_web_service.py @@ -21,7 +21,7 @@ import os class BertService(WebService): def load(self): - self.reader = BertReader(vocab_file="vocab.txt", max_seq_len=20) + self.reader = BertReader(vocab_file="vocab.txt", max_seq_len=128) def preprocess(self, feed={}, fetch=[]): feed_res = self.reader.process(feed["words"].encode("utf-8")) diff --git a/tools/serving_build.sh b/tools/serving_build.sh index 6549838a..70c7238d 100644 --- a/tools/serving_build.sh +++ b/tools/serving_build.sh @@ -276,27 +276,47 @@ function python_test_bert() { case $TYPE in CPU) pip install paddlehub - python prepare_model.py 20 + # Because download from paddlehub may timeout, + # download the model from bos(max_seq_len=128). + wget https://paddle-serving.bj.bcebos.com/paddle_hub_models/text/SemanticModel/bert_chinese_L-12_H-768_A-12.tar.gz + tar -xzf bert_chinese_L-12_H-768_A-12.tar.gz sh get_data.sh - check_cmd "python -m paddle_serving_server.serve --model bert_seq20_model/ --port 9292 &" + check_cmd "python -m paddle_serving_server.serve --model bert_chinese_L-12_H-768_A-12_model --port 9292 &" sleep 5 pip install paddle_serving_app - check_cmd "head -n 10 data-c.txt | python bert_client.py --model bert_seq20_client/serving_client_conf.prototxt" + check_cmd "head -n 10 data-c.txt | python bert_client.py --model bert_chinese_L-12_H-768_A-12_client/serving_client_conf.prototxt" kill_server_process - ps -ef | grep "paddle_serving_server" | grep -v grep | awk '{print $2}' | xargs kill - ps -ef | grep "serving" | grep -v grep | awk '{print $2}' | xargs kill + # python prepare_model.py 20 + # sh get_data.sh + # check_cmd "python -m paddle_serving_server.serve --model bert_seq20_model/ --port 9292 &" + # sleep 5 + # pip install paddle_serving_app + # check_cmd "head -n 10 data-c.txt | python bert_client.py --model bert_seq20_client/serving_client_conf.prototxt" + # kill_server_process + # ps -ef | grep "paddle_serving_server" | grep -v grep | awk '{print $2}' | xargs kill + # ps -ef | grep "serving" | grep -v grep | awk '{print $2}' | xargs kill echo "bert RPC inference pass" ;; GPU) pip install paddlehub - python prepare_model.py 20 + # Because download from paddlehub may timeout, + # download the model from bos(max_seq_len=128). + wget https://paddle-serving.bj.bcebos.com/paddle_hub_models/text/SemanticModel/bert_chinese_L-12_H-768_A-12.tar.gz + tar -xzf bert_chinese_L-12_H-768_A-12.tar.gz sh get_data.sh - check_cmd "python -m paddle_serving_server_gpu.serve --model bert_seq20_model/ --port 9292 --gpu_ids 0 &" + check_cmd "python -m paddle_serving_server_gpu.serve --model bert_chinese_L-12_H-768_A-12_model --port 9292 --gpu_ids 0 &" sleep 5 pip install paddle_serving_app - check_cmd "head -n 10 data-c.txt | python bert_client.py --model bert_seq20_client/serving_client_conf.prototxt" + check_cmd "head -n 10 data-c.txt | python bert_client.py --model bert_chinese_L-12_H-768_A-12_client/serving_client_conf.prototxt" kill_server_process - ps -ef | grep "paddle_serving_server" | grep -v grep | awk '{print $2}' | xargs kill + # python prepare_model.py 20 + # sh get_data.sh + # check_cmd "python -m paddle_serving_server_gpu.serve --model bert_seq20_model/ --port 9292 --gpu_ids 0 &" + # sleep 5 + # pip install paddle_serving_app + # check_cmd "head -n 10 data-c.txt | python bert_client.py --model bert_seq20_client/serving_client_conf.prototxt" + # kill_server_process + # ps -ef | grep "paddle_serving_server" | grep -v grep | awk '{print $2}' | xargs kill echo "bert RPC inference pass" ;; *) -- GitLab