Skip to content

S
Serving

PaddlePaddle / Serving
大约 1 年 前同步成功

通知 185
Star 833
Fork 253
S
Serving

体验新版 GitCode，发现更多精彩内容 >>

未验证 提交 fcc97c73 编写于 作者: T Thomas Young 提交者： GitHub
浏览文件
操作

Merge pull request #3 from PaddlePaddle/develop 

update
上级 6bf4c3a8 7ea59bda
隐藏空白更改
内联 并排
Showing with 346 addition and 2028 deletion
README.md
浏览文件 @ fcc97c73
......@@ -45,10 +45,11 @@ nvidia-docker exec -it test bash
```
```shell
pip install paddle-serving-client==0.3.2
pip install paddle-serving-server==0.3.2 # CPU
pip install paddle-serving-server-gpu==0.3.2.post9 # GPU with CUDA9.0
pip install paddle-serving-server-gpu==0.3.2.post10 # GPU with CUDA10.0
pip install paddle-serving-client==0.4.0
pip install paddle-serving-server==0.4.0 # CPU
pip install paddle-serving-server-gpu==0.4.0.post9 # GPU with CUDA9.0
pip install paddle-serving-server-gpu==0.4.0.post10 # GPU with CUDA10.0
pip install paddle-serving-server-gpu==0.4.0.trt # GPU with CUDA10.1+TensorRT
```
You may need to use a domestic mirror source (in China, you can use the Tsinghua mirror source, add `-i https://pypi.tuna.tsinghua.edu.cn/simple` to pip command) to speed up the download.
......@@ -57,7 +58,7 @@ If you need install modules compiled with develop branch, please download packag
Packages of paddle-serving-server and paddle-serving-server-gpu support Centos 6/7, Ubuntu 16/18, Windows 10.
Packages of paddle-serving-client and paddle-serving-app support Linux and Windows, but paddle-serving-client only support python2.7/3.6/3.7.
Packages of paddle-serving-client and paddle-serving-app support Linux and Windows, but paddle-serving-client only support python2.7/3.5/3.6/3.7.
Recommended to install paddle >= 1.8.4.
......@@ -65,15 +66,6 @@ For **Windows Users**, please read the document [Paddle Serving for Windows User
<h2 align="center"> Pre-built services with Paddle Serving</h2>
<h3 align="center">Latest release</h4>
<p align="center">
<a href="https://github.com/PaddlePaddle/Serving/tree/develop/python/examples/ocr">Optical Character Recognition</a>
<br>
<a href="https://github.com/PaddlePaddle/Serving/tree/develop/python/examples/faster_rcnn_model">Object Detection</a>
<br>
<a href="https://github.com/PaddlePaddle/Serving/tree/develop/python/examples/deeplabv3">Image Segmentation</a>
<p>
<h3 align="center">Chinese Word Segmentation</h4>
``` shell
......@@ -113,11 +105,11 @@ tar -xzf uci_housing.tar.gz
Paddle Serving provides HTTP and RPC based service for users to access
### HTTP service
### RPC service
Paddle Serving provides a built-in python module called `paddle_serving_server.serve` that can start a RPC service or a http service with one-line command. If we specify the argument `--name uci`, it means that we will have a HTTP service with a url of `$IP:$PORT/uci/prediction`
A user can also start a RPC service with `paddle_serving_server.serve`. RPC service is usually faster than HTTP service, although a user needs to do some coding based on Paddle Serving's python client API. Note that we do not specify `--name` here.
``` shell
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9292 --name uci
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9292
```
<center>
......@@ -125,42 +117,63 @@ python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --po
|--------------|------|-----------|--------------------------------|
| `thread` | int | `4` | Concurrency of current service |
| `port` | int | `9292` | Exposed port of current service to users|
| `name` | str | `""` | Service name, can be used to generate HTTP request url |
| `model` | str | `""` | Path of paddle model directory to be served |
| `mem_optim_off` | - | - | Disable memory / graphic memory optimization |
| `ir_optim` | - | - | Enable analysis and optimization of calculation graph |
| `use_mkl` (Only for cpu version) | - | - | Run inference with MKL |
| `use_trt` (Only for trt version) | - | - | Run inference with TensorRT |
Here, we use `curl` to send a HTTP POST request to the service we just started. Users can use any python library to send HTTP POST as well, e.g, [requests](https://requests.readthedocs.io/en/master/).
</center>
``` shell
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]}], "fetch":["price"]}' http://127.0.0.1:9292/uci/prediction
```
### RPC service
A user can also start a RPC service with `paddle_serving_server.serve`. RPC service is usually faster than HTTP service, although a user needs to do some coding based on Paddle Serving's python client API. Note that we do not specify `--name` here.
``` shell
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9292
```
``` python
```python
# A user can visit rpc service through paddle_serving_client API
from paddle_serving_client import Client
import numpy as np
client = Client()
client.load_client_config("uci_housing_client/serving_client_conf.prototxt")
client.connect(["127.0.0.1:9292"])
data = [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727,
-0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]
fetch_map = client.predict(feed={"x": data}, fetch=["price"])
fetch_map = client.predict(feed={"x": np.array(data).reshape(1,13,1)}, fetch=["price"])
print(fetch_map)
```
Here, `client.predict` function has two arguments. `feed` is a `python dict` with model input variable alias name and values. `fetch` assigns the prediction variables to be returned from servers. In the example, the name of `"x"` and `"price"` are assigned when the servable model is saved during training.
### WEB service
Users can also put the data format processing logic on the server side, so that they can directly use curl to access the service, refer to the following case whose path is `python/examples/fit_a_line`
```python
from paddle_serving_server.web_service import WebService
import numpy as np
class UciService(WebService):
def preprocess(self, feed=[], fetch=[]):
feed_batch = []
is_batch = True
new_data = np.zeros((len(feed), 1, 13)).astype("float32")
for i, ins in enumerate(feed):
nums = np.array(ins["x"]).reshape(1, 1, 13)
new_data[i] = nums
feed = {"x": new_data}
return feed, fetch, is_batch
uci_service = UciService(name="uci")
uci_service.load_model_config("uci_housing_model")
uci_service.prepare_server(workdir="workdir", port=9292)
uci_service.run_rpc_service()
uci_service.run_web_service()
```
for client side,
```
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]}], "fetch":["price"]}' http://127.0.0.1:9292/uci/prediction
```
the response is
```
{"result":{"price":[[18.901151657104492]]}}
```
<h2 align="center">Some Key Features of Paddle Serving</h2>
- Integrate with Paddle training pipeline seamlessly, most paddle models can be deployed **with one line command**.
......@@ -215,6 +228,10 @@ To connect with other users and contributors, welcome to join our [Slack channel
If you want to contribute code to Paddle Serving, please reference [Contribution Guidelines](doc/CONTRIBUTE.md)
- Special Thanks to [@BeyondYourself](https://github.com/BeyondYourself) in complementing the gRPC tutorial, updating the FAQ doc and modifying the mdkir command
- Special Thanks to [@mcl-stone](https://github.com/mcl-stone) in updating faster_rcnn benchmark
- Special Thanks to [@cg82616424](https://github.com/cg82616424) in updating the unet benchmark and modifying resize comment error
### Feedback
For any feedback or to report a bug, please propose a [GitHub Issue](https://github.com/PaddlePaddle/Serving/issues).
......
README_CN.md
浏览文件 @ fcc97c73
......@@ -47,10 +47,11 @@ nvidia-docker run -p 9292:9292 --name test -dit hub.baidubce.com/paddlepaddle/se
nvidia-docker exec -it test bash
```
```shell
pip install paddle-serving-client==0.3.2
pip install paddle-serving-server==0.3.2 # CPU
pip install paddle-serving-server-gpu==0.3.2.post9 # GPU with CUDA9.0
pip install paddle-serving-server-gpu==0.3.2.post10 # GPU with CUDA10.0
pip install paddle-serving-client==0.4.0
pip install paddle-serving-server==0.4.0 # CPU
pip install paddle-serving-server-gpu==0.4.0.post9 # GPU with CUDA9.0
pip install paddle-serving-server-gpu==0.4.0.post10 # GPU with CUDA10.0
pip install paddle-serving-server-gpu==0.4.0.trt # GPU with CUDA10.1+TensorRT
```
您可能需要使用国内镜像源（例如清华源, 在pip命令中添加`-i https://pypi.tuna.tsinghua.edu.cn/simple`）来加速下载。
......@@ -107,13 +108,12 @@ tar -xzf uci_housing.tar.gz
Paddle Serving 为用户提供了基于 HTTP 和 RPC 的服务
<h3 align="center">RPC服务</h3>
<h3 align="center">HTTP服务</h3>
Paddle Serving提供了一个名为`paddle_serving_server.serve`的内置python模块，可以使用单行命令启动RPC服务或HTTP服务。如果我们指定参数`--name uci`，则意味着我们将拥有一个HTTP服务，其URL为$IP:$PORT/uci/prediction`。
用户还可以使用`paddle_serving_server.serve`启动RPC服务。 尽管用户需要基于Paddle Serving的python客户端API进行一些开发，但是RPC服务通常比HTTP服务更快。需要指出的是这里我们没有指定`--name`。
``` shell
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9292 --name uci
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9292
```
<center>
......@@ -128,21 +128,10 @@ python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --po
| `use_mkl` (Only for cpu version) | - | - | Run inference with MKL |
| `use_trt` (Only for trt version) | - | - | Run inference with TensorRT |
我们使用 `curl` 命令来发送HTTP POST请求给刚刚启动的服务。用户也可以调用python库来发送HTTP POST请求，请参考英文文档 [requests](https://requests.readthedocs.io/en/master/)。
我们使用 `curl` 命令来发送HTTP POST请求给刚刚启动的服务。用户也可以调用python库来发送HTTP POST请求，请参考英文文
档 [requests](https://requests.readthedocs.io/en/master/)。
</center>
``` shell
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]}], "fetch":["price"]}' http://127.0.0.1:9292/uci/prediction
```
<h3 align="center">RPC服务</h3>
用户还可以使用`paddle_serving_server.serve`启动RPC服务。 尽管用户需要基于Paddle Serving的python客户端API进行一些开发，但是RPC服务通常比HTTP服务更快。需要指出的是这里我们没有指定`--name`。
``` shell
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9292
```
``` python
# A user can visit rpc service through paddle_serving_client API
from paddle_serving_client import Client
......@@ -152,12 +141,45 @@ client.load_client_config("uci_housing_client/serving_client_conf.prototxt")
client.connect(["127.0.0.1:9292"])
data = [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727,
-0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]
fetch_map = client.predict(feed={"x": data}, fetch=["price"])
fetch_map = client.predict(feed={"x": np.array(data).reshape(1,13,1)}, fetch=["price"])
print(fetch_map)
```
在这里，`client.predict`函数具有两个参数。 `feed`是带有模型输入变量别名和值的`python dict`。 `fetch`被要从服务器返回的预测变量赋值。 在该示例中，在训练过程中保存可服务模型时，被赋值的tensor名为`"x"`和`"price"`。
<h3 align="center">HTTP服务</h3>
用户也可以将数据格式处理逻辑放在服务器端进行，这样就可以直接用curl去访问服务，参考如下案例，在目录`python/examples/fit_a_line`
```python
from paddle_serving_server.web_service import WebService
import numpy as np
class UciService(WebService):
def preprocess(self, feed=[], fetch=[]):
feed_batch = []
is_batch = True
new_data = np.zeros((len(feed), 1, 13)).astype("float32")
for i, ins in enumerate(feed):
nums = np.array(ins["x"]).reshape(1, 1, 13)
new_data[i] = nums
feed = {"x": new_data}
return feed, fetch, is_batch
uci_service = UciService(name="uci")
uci_service.load_model_config("uci_housing_model")
uci_service.prepare_server(workdir="workdir", port=9292)
uci_service.run_rpc_service()
uci_service.run_web_service()
```
客户端输入
```
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]}], "fetch":["price"]}' http://127.0.0.1:9292/uci/prediction
```
返回结果
```
{"result":{"price":[[18.901151657104492]]}}
```
<h2 align="center">Paddle Serving的核心功能</h2>
- 与Paddle训练紧密连接，绝大部分Paddle模型可以 **一键部署**.
......@@ -210,6 +232,10 @@ print(fetch_map)
如果您想为Paddle Serving贡献代码，请参考 [Contribution Guidelines](doc/CONTRIBUTE.md)
- 特别感谢 [@BeyondYourself](https://github.com/BeyondYourself) 提供grpc教程，更新FAQ教程，整理文件目录。
- 特别感谢 [@mcl-stone](https://github.com/mcl-stone) 提供faster rcnn benchmark脚本
- 特别感谢 [@cg82616424](https://github.com/cg82616424) 提供unet benchmark脚本和修改部分注释错误
### 反馈
如有任何反馈或是bug，请在 [GitHub Issue](https://github.com/PaddlePaddle/Serving/issues)提交
......
cmake/paddlepaddle.cmake
浏览文件 @ fcc97c73
......@@ -114,7 +114,7 @@ ADD_LIBRARY(openblas STATIC IMPORTED GLOBAL)
SET_PROPERTY(TARGET openblas PROPERTY IMPORTED_LOCATION ${PADDLE_INSTALL_DIR}/third_party/install/openblas/lib/libopenblas.a)
ADD_LIBRARY(paddle_fluid SHARED IMPORTED GLOBAL)
SET_PROPERTY(TARGET paddle_fluid PROPERTY IMPORTED_LOCATION ${PADDLE_INSTALL_DIR}/lib/libpaddle_fluid.a)
SET_PROPERTY(TARGET paddle_fluid PROPERTY IMPORTED_LOCATION ${PADDLE_INSTALL_DIR}/lib/libpaddle_fluid.so)
if (WITH_TRT)
ADD_LIBRARY(nvinfer SHARED IMPORTED GLOBAL)
......@@ -127,17 +127,12 @@ endif()
ADD_LIBRARY(xxhash STATIC IMPORTED GLOBAL)
SET_PROPERTY(TARGET xxhash PROPERTY IMPORTED_LOCATION ${PADDLE_INSTALL_DIR}/third_party/install/xxhash/lib/libxxhash.a)
ADD_LIBRARY(cryptopp STATIC IMPORTED GLOBAL)
SET_PROPERTY(TARGET cryptopp PROPERTY IMPORTED_LOCATION ${PADDLE_INSTALL_DIR}/third_party/install/cryptopp/lib/libcryptopp.a)
LIST(APPEND external_project_dependencies paddle)
LIST(APPEND paddle_depend_libs
xxhash cryptopp)
xxhash)
if(WITH_TRT)
LIST(APPEND paddle_depend_libs
nvinfer nvinfer_plugin)
endif()
core/predictor/tools/seq_generator.cpp
浏览文件 @ fcc97c73
......@@ -17,11 +17,11 @@
#include <fstream>
#include <iostream>
#include <memory>
#include <thread> //NOLINT
#include <thread>
#include "core/predictor/framework.pb.h"
#include "quant.h" // NOLINT
#include "seq_file.h" // NOLINT
#include "quant.h"
#include "seq_file.h"
inline uint64_t time_diff(const struct timeval &start_time,
const struct timeval &end_time) {
......@@ -113,15 +113,13 @@ int dump_parameter(const char *input_file, const char *output_file) {
// std::cout << "key_len " << key_len << " value_len " << value_buf_len
// << std::endl;
memcpy(value_buf, tensor_buf + offset, value_buf_len);
seq_file_writer.write(
std::to_string(i).c_str(), sizeof(i), value_buf, value_buf_len);
seq_file_writer.write((char *)&i, sizeof(i), value_buf, value_buf_len);
offset += value_buf_len;
}
return 0;
}
float *read_embedding_table(const char *file1,
std::vector<int64_t> &dims) { // NOLINT
float *read_embedding_table(const char *file1, std::vector<int64_t> &dims) {
std::ifstream is(file1);
// Step 1: is read version, os write version
uint32_t version;
......@@ -244,7 +242,7 @@ int compress_parameter_parallel(const char *file1,
float x = *(emb_table + k * emb_size + e);
int val = round((x - xmin) / scale);
val = std::max(0, val);
val = std::min(static_cast<int>(pow2bits) - 1, val);
val = std::min((int)pow2bits - 1, val);
*(tensor_temp + 2 * sizeof(float) + e) = val;
}
result[k] = tensor_temp;
......@@ -264,8 +262,7 @@ int compress_parameter_parallel(const char *file1,
}
SeqFileWriter seq_file_writer(file2);
for (int64_t i = 0; i < dict_size; i++) {
seq_file_writer.write(
std::to_string(i).c_str(), sizeof(i), result[i], per_line_size);
seq_file_writer.write((char *)&i, sizeof(i), result[i], per_line_size);
}
return 0;
}
......
doc/BERT_10_MINS.md
浏览文件 @ fcc97c73
......@@ -56,21 +56,25 @@ the script of client side bert_client.py is as follow:
[//file]:#bert_client.py
``` python
import os
import sys
from paddle_serving_client import Client
from paddle_serving_client.utils import benchmark_args
from paddle_serving_app.reader import ChineseBertReader
import numpy as np
args = benchmark_args()
reader = ChineseBertReader()
reader = ChineseBertReader({"max_seq_len": 128})
fetch = ["pooled_output"]
endpoint_list = ["127.0.0.1:9292"]
endpoint_list = ['127.0.0.1:9292']
client = Client()
client.load_client_config("bert_seq20_client/serving_client_conf.prototxt")
client.load_client_config(args.model)
client.connect(endpoint_list)
for line in sys.stdin:
feed_dict = reader.process(line)
result = client.predict(feed=feed_dict, fetch=fetch)
for key in feed_dict.keys():
feed_dict[key] = np.array(feed_dict[key]).reshape((128, 1))
result = client.predict(feed=feed_dict, fetch=fetch, batch=False)
```
run
......
doc/BERT_10_MINS_CN.md
浏览文件 @ fcc97c73
......@@ -52,18 +52,23 @@ pip install paddle_serving_app
``` python
import sys
from paddle_serving_client import Client
from paddle_serving_client.utils import benchmark_args
from paddle_serving_app.reader import ChineseBertReader
import numpy as np
args = benchmark_args()
reader = ChineseBertReader()
reader = ChineseBertReader({"max_seq_len": 128})
fetch = ["pooled_output"]
endpoint_list = ["127.0.0.1:9292"]
endpoint_list = ['127.0.0.1:9292']
client = Client()
client.load_client_config("bert_seq20_client/serving_client_conf.prototxt")
client.load_client_config(args.model)
client.connect(endpoint_list)
for line in sys.stdin:
feed_dict = reader.process(line)
result = client.predict(feed=feed_dict, fetch=fetch)
for key in feed_dict.keys():
feed_dict[key] = np.array(feed_dict[key]).reshape((128, 1))
result = client.predict(feed=feed_dict, fetch=fetch, batch=False)
```
执行
......
doc/RUN_IN_DOCKER.md
浏览文件 @ fcc97c73
......@@ -32,63 +32,9 @@ The `-p` option is to map the `9292` port of the container to the `9292` port of
### Install PaddleServing
In order to make the image smaller, the PaddleServing package is not installed in the image. You can run the following command to install it:
```bash
pip install paddle-serving-server
```
You may need to use a domestic mirror source (in China, you can use the Tsinghua mirror source of the following example) to speed up the download:
```shell
pip install paddle-serving-server -i https://pypi.tuna.tsinghua.edu.cn/simple
```
### Test example
Get the trained Boston house price prediction model by the following command:
```bash
wget --no-check-certificate https://paddle-serving.bj.bcebos.com/uci_housing.tar.gz
tar -xzf uci_housing.tar.gz
```
- Test HTTP service
Running on the Server side (inside the container):
```bash
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9292 --name uci >std.log 2>err.log &
```
Running on the Client side (inside or outside the container):
```bash
curl -H "Content-Type:application/json" -X POST -d '{"feed":{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]}, "fetch":["price"]}' http://127.0.0.1:9292/uci/prediction
```
- Test RPC service
Running on the Server side (inside the container):
```bash
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9292 >std.log 2>err.log &
```
Running following Python code on the Client side (inside or outside the container, The `paddle-serving-client` package needs to be installed):
```bash
from paddle_serving_client import Client
client = Client()
client.load_client_config("uci_housing_client/serving_client_conf.prototxt")
client.connect(["127.0.0.1:9292"])
data = [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727,
-0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]
fetch_map = client.predict(feed={"x": data}, fetch=["price"])
print(fetch_map)
```
The mirror comes with `paddle_serving_server`, `paddle_serving_client`, and `paddle_serving_app` corresponding to the mirror tag version. If users don’t need to change the version, they can use it directly, which is suitable for environments without extranet services.
If you need to change the version, please refer to the instructions on the homepage to download the pip package of the corresponding version.
## GPU
......@@ -100,7 +46,7 @@ The GPU version is basically the same as the CPU version, with only some differe
Refer to [this document](DOCKER_IMAGES.md) for a docker image, the following is an example of an `cuda9.0-cudnn7` image:
```shell
nvidia-docker pull hub.baidubce.com/paddlepaddle/serving:latest-cuda9.0-cudnn7
docker pull hub.baidubce.com/paddlepaddle/serving:latest-cuda9.0-cudnn7
```
### Create container
......@@ -110,77 +56,21 @@ nvidia-docker run -p 9292:9292 --name test -dit hub.baidubce.com/paddlepaddle/se
nvidia-docker exec -it test bash
```
The `-p` option is to map the `9292` port of the container to the `9292` port of the host.
### Install PaddleServing
In order to make the image smaller, the PaddleServing package is not installed in the image. You can run the following command to install it:
or
```bash
pip install paddle-serving-server-gpu
```
You may need to use a domestic mirror source (in China, you can use the Tsinghua mirror source of the following example) to speed up the download:
```shell
pip install paddle-serving-server-gpu -i https://pypi.tuna.tsinghua.edu.cn/simple
```
### Test example
When running the GPU Server, you need to set the GPUs used by the prediction service through the `--gpu_ids` option, and the CPU is used by default. An error will be reported when the value of `--gpu_ids` exceeds the environment variable `CUDA_VISIBLE_DEVICES`. The following example specifies to use a GPU with index 0:
```shell
export CUDA_VISIBLE_DEVICES=0,1
python -m paddle_serving_server_gpu.serve --model uci_housing_model --port 9292 --gpu_ids 0
```
Get the trained Boston house price prediction model by the following command:
```bash
wget --no-check-certificate https://paddle-serving.bj.bcebos.com/uci_housing.tar.gz
tar -xzf uci_housing.tar.gz
docker run --gpus all -p 9292:9292 --name test -dit hub.baidubce.com/paddlepaddle/serving:latest-cuda9.0-cudnn7
docker exec -it test bash
```
- Test HTTP service
Running on the Server side (inside the container):
```bash
python -m paddle_serving_server_gpu.serve --model uci_housing_model --thread 10 --port 9292 --name uci --gpu_ids 0
```
Running on the Client side (inside or outside the container):
```bash
curl -H "Content-Type:application/json" -X POST -d '{"feed":{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]}, "fetch":["price"]}' http://127.0.0.1:9292/uci/prediction
```
- Test RPC service
Running on the Server side (inside the container):
```bash
python -m paddle_serving_server_gpu.serve --model uci_housing_model --thread 10 --port 9292 --gpu_ids 0
```
Running following Python code on the Client side (inside or outside the container, The `paddle-serving-client` package needs to be installed):
```bash
from paddle_serving_client import Client
client = Client()
client.load_client_config("uci_housing_client/serving_client_conf.prototxt")
client.connect(["127.0.0.1:9292"])
data = [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727,
-0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]
fetch_map = client.predict(feed={"x": data}, fetch=["price"])
print(fetch_map)
```
The `-p` option is to map the `9292` port of the container to the `9292` port of the host.
### Install PaddleServing
The mirror comes with `paddle_serving_server_gpu`, `paddle_serving_client`, and `paddle_serving_app` corresponding to the mirror tag version. If users don’t need to change the version, they can use it directly, which is suitable for environments without extranet services.
If you need to change the version, please refer to the instructions on the homepage to download the pip package of the corresponding version.
## Attention
## Precautious
Runtime images cannot be used for compilation. If you want to compile from source, refer to [COMPILE](COMPILE.md).
doc/RUN_IN_DOCKER_CN.md
浏览文件 @ fcc97c73
......@@ -20,7 +20,6 @@ Docker（GPU版本需要在GPU机器上安装nvidia-docker）
docker pull hub.baidubce.com/paddlepaddle/serving:latest
```
### 创建容器并进入
```bash
......@@ -32,74 +31,11 @@ docker exec -it test bash
### 安装PaddleServing
为了减小镜像的体积，镜像中没有安装Serving包，要执行下面命令进行安装。
```bash
pip install paddle-serving-server
```
您可能需要使用国内镜像源（例如清华源）来加速下载。
```shell
pip install paddle-serving-server -i https://pypi.tuna.tsinghua.edu.cn/simple
```
### 测试example
通过下面命令获取训练好的Boston房价预估模型：
```bash
wget --no-check-certificate https://paddle-serving.bj.bcebos.com/uci_housing.tar.gz
tar -xzf uci_housing.tar.gz
```
- 测试HTTP服务
在Server端（容器内）运行：
镜像里自带对应镜像tag版本的`paddle_serving_server`，`paddle_serving_client`，`paddle_serving_app`，如果用户不需要更改版本，可以直接使用，适用于没有外网服务的环境。
```bash
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9292 --name uci >std.log 2>err.log &
```
如果需要更换版本，请参照首页的指导，下载对应版本的pip包。
在Client端（容器内或容器外）运行：
```bash
curl -H "Content-Type:application/json" -X POST -d '{"feed":{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]}, "fetch":["price"]}' http://127.0.0.1:9292/uci/prediction
```
- 测试RPC服务
在Server端（容器内）运行：
```bash
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9292 >std.log 2>err.log &
```
在Client端（容器内或容器外，需要安装`paddle-serving-client`包）运行下面Python代码：
```python
from paddle_serving_client import Client
client = Client()
client.load_client_config("uci_housing_client/serving_client_conf.prototxt")
client.connect(["127.0.0.1:9292"])
data = [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727,
-0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]
fetch_map = client.predict(feed={"x": data}, fetch=["price"])
print(fetch_map)
```
## GPU版本
GPU版本与CPU版本基本一致，只有部分接口命名的差别（GPU版本需要在GPU机器上安装nvidia-docker）。
### 获取镜像
参考[该文档](DOCKER_IMAGES_CN.md)获取镜像，这里以 `cuda9.0-cudnn7` 的镜像为例：
```shell
nvidia-docker pull hub.baidubce.com/paddlepaddle/serving:latest-cuda9.0-cudnn7
```
## GPU 版本
### 创建容器并进入
......@@ -107,74 +43,19 @@ nvidia-docker pull hub.baidubce.com/paddlepaddle/serving:latest-cuda9.0-cudnn7
nvidia-docker run -p 9292:9292 --name test -dit hub.baidubce.com/paddlepaddle/serving:latest-cuda9.0-cudnn7
nvidia-docker exec -it test bash
```
`-p`选项是为了将容器的`9292`端口映射到宿主机的`9292`端口。
### 安装PaddleServing
为了减小镜像的体积，镜像中没有安装Serving包，要执行下面命令进行安装。
```bash
pip install paddle-serving-server-gpu
```
您可能需要使用国内镜像源（例如清华源）来加速下载。
```shell
pip install paddle-serving-server-gpu -i https://pypi.tuna.tsinghua.edu.cn/simple
```
### 测试example
在运行GPU版Server时需要通过`--gpu_ids`选项设置预测服务使用的GPU，缺省状态默认使用CPU。当设置的`--gpu_ids`超出环境变量`CUDA_VISIBLE_DEVICES`时会报错。下面的示例为指定使用索引为0的GPU：
```shell
export CUDA_VISIBLE_DEVICES=0,1
python -m paddle_serving_server_gpu.serve --model uci_housing_model --port 9292 --gpu_ids 0
```
通过下面命令获取训练好的Boston房价预估模型：
或者
```bash
wget --no-check-certificate https://paddle-serving.bj.bcebos.com/uci_housing.tar.gz
tar -xzf uci_housing.tar.gz
docker run --gpus all -p 9292:9292 --name test -dit hub.baidubce.com/paddlepaddle/serving:latest-cuda9.0-cudnn7
docker exec -it test bash
```
- 测试HTTP服务
在Server端（容器内）运行：
```bash
python -m paddle_serving_server_gpu.serve --model uci_housing_model --thread 10 --port 9292 --name uci --gpu_ids 0
```
在Client端（容器内或容器外）运行：
```bash
curl -H "Content-Type:application/json" -X POST -d '{"feed":{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]}, "fetch":["price"]}' http://127.0.0.1:9292/uci/prediction
```
- 测试RPC服务
在Server端（容器内）运行：
`-p`选项是为了将容器的`9292`端口映射到宿主机的`9292`端口。
```bash
python -m paddle_serving_server_gpu.serve --model uci_housing_model --thread 10 --port 9292 --gpu_ids 0
```
### 安装PaddleServing
在Client端（容器内或容器外，需要安装`paddle-serving-client`包）运行下面Python代码：
镜像里自带对应镜像tag版本的`paddle_serving_server_gpu`，`paddle_serving_client`，`paddle_serving_app`，如果用户不需要更改版本，可以直接使用，适用于没有外网服务的环境。
```bash
from paddle_serving_client import Client
client = Client()
client.load_client_config("uci_housing_client/serving_client_conf.prototxt")
client.connect(["127.0.0.1:9292"])
data = [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727,
-0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]
fetch_map = client.predict(feed={"x": data}, fetch=["price"])
print(fetch_map)
```
如果需要更换版本，请参照首页的指导，下载对应版本的pip包。
## 注意事项
......
doc/WINDOWS_TUTORIAL.md
浏览文件 @ fcc97c73
......@@ -8,7 +8,7 @@ This document guides users how to build Paddle Serving service on the Windows pl
### Running Paddle Serving on Native Windows System
**Configure Python environment variables to PATH**: First, you need to add the directory where the Python executable program is located to the PATH. Usually in **System Properties/My Computer Properties**-**Advanced**-**Environment Variables**, click Path and add the path at the beginning. For example, `C:\Users\$USER\AppData\Local\Programs\Python\Python36`, and finally click **OK** continuously. If you enter python on Powershell, you can enter the python interactive interface, indicating that the environment variable configuration is successful.
**Configure Python environment variables to PATH**: **We only support Python 3.5+ on Native Windows System.**. First, you need to add the directory where the Python executable program is located to the PATH. Usually in **System Properties/My Computer Properties**-**Advanced**-**Environment Variables**, click Path and add the path at the beginning. For example, `C:\Users\$USER\AppData\Local\Programs\Python\Python36`, and finally click **OK** continuously. If you enter python on Powershell, you can enter the python interactive interface, indicating that the environment variable configuration is successful.
**Install wget**: Because all the downloads in the tutorial and the built-in model download function in `paddle_serving_app` all use the wget tool, download the binary package at the [link](http://gnuwin32.sourceforge.net/packages/wget.htm), unzip and copy it to `C:\Windows\System32`, if there is a security prompt, you need to pass it.
......@@ -32,6 +32,7 @@ python -m pip install -U paddle_serving_server_gpu paddle_serving_client paddle_
```
git clone https://github.com/paddlepaddle/Serving
pip install -r python/requirements_win.txt
```
**Run OCR example**:
......@@ -42,7 +43,7 @@ python -m paddle_serving_app.package --get_model ocr_rec
tar -xzvf ocr_rec.tar.gz
python -m paddle_serving_app.package --get_model ocr_det
tar -xzvf ocr_det.tar.gz
python ocr_debugger_server.py &
python ocr_debugger_server.py cpu &
python ocr_web_client.py
```
......
doc/WINDOWS_TUTORIAL_CN.md
浏览文件 @ fcc97c73
......@@ -8,7 +8,7 @@
### 原生Windows系统运行Paddle Serving
**配置Python环境变量到PATH**：首先需要将Python的可执行程序所在目录加入到PATH当中。通常在**系统属性/我的电脑属性**-**高级**-**环境变量** ，点选Path，并在开头加上路径。例如`C:\Users\$USER\AppData\Local\Programs\Python\Python36`，最后连续点击**确定** 。在Powershell上如果输入python可以进入python交互界面，说明环境变量配置成功。
**配置Python环境变量到PATH**：**目前原生Windows仅支持Python 3.5或更高版本**。首先需要将Python的可执行程序所在目录加入到PATH当中。通常在**系统属性/我的电脑属性**-**高级**-**环境变量** ，点选Path，并在开头加上路径。例如`C:\Users\$USER\AppData\Local\Programs\Python\Python36`，最后连续点击**确定** 。在Powershell上如果输入python可以进入python交互界面，说明环境变量配置成功。
**安装wget工具**：由于教程当中所有的下载，以及`paddle_serving_app`当中内嵌的模型下载功能，都是用到wget工具，在链接[下载wget](http://gnuwin32.sourceforge.net/packages/wget.htm)，解压后复制到`C:\Windows\System32`下，如有安全提示需要通过。
......@@ -32,6 +32,7 @@ python -m pip install -U paddle_serving_server_gpu paddle_serving_client paddle_
```
git clone https://github.com/paddlepaddle/Serving
pip install -r python/requirements_win.txt
```
**运行OCR示例**：
......@@ -42,7 +43,7 @@ python -m paddle_serving_app.package --get_model ocr_rec
tar -xzvf ocr_rec.tar.gz
python -m paddle_serving_app.package --get_model ocr_det
tar -xzvf ocr_det.tar.gz
python ocr_debugger_server.py &
python ocr_debugger_server.py cpu &
python ocr_web_client.py
```
......
java/README.md
浏览文件 @ fcc97c73
## Java Demo
## Tutorial of Java Client for Paddle Serving
(English|[简体中文](./README_CN.md))
### Development Environment
In order to facilitate users to use java for development, we provide the compiled Serving project to be placed in the java mirror. The way to get the mirror and enter the development environment is
```
docker pull hub.baidubce.com/paddlepaddle/serving:0.4.0-java
docker run --rm -dit --name java_serving hub.baidubce.com/paddlepaddle/serving:0.4.0-java
docker exec -it java_serving bash
cd Serving/java
```
The Serving folder is at the develop branch when the docker image is generated. You need to git pull to the latest version or git checkout to the desired branch.
### Install client dependencies
Due to the large number of dependent libraries, the image has been compiled once at the time of generation, and the user can perform the following operations
### Install package
```
mvn compile
mvn install
......@@ -9,18 +27,49 @@ mvn compile
mvn install
```
### Start Server
### Start the server
take the fit_a_line demo as example
Take the fit_a_line model as an example, the server starts
```
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9393 --use_multilang #CPU
python -m paddle_serving_server_gpu.serve --model uci_housing_model --thread 10 --port 9393 --use_multilang #GPU
cd ../../python/examples/fit_a_line
sh get_data.sh
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9393 --use_multilang &
```
### Client Predict
Client prediction
```
cd ../../../java/examples/target
java -cp paddle-serving-sdk-java-examples-0.0.1-jar-with-dependencies.jar PaddleServingClientExample fit_a_line
```
The Java example also contains the prediction client of Bert, Model_enaemble, asyn_predict, batch_predict, Cube_local, Cube_quant, and Yolov4 models.
Take yolov4 as an example, the server starts
```
python -m paddle_serving_app.package --get_model yolov4
tar -xzvf yolov4.tar.gz
python -m paddle_serving_server_gpu.serve --model yolov4_model --port 9393 --gpu_ids 0 --use_multilang & #It needs to be executed in GPU Docker, otherwise the execution method of CPU must be used.
```
Client prediction
```
# in /Serving/java/examples/target
java -cp paddle-serving-sdk-java-examples-0.0.1-jar-with-dependencies.jar PaddleServingClientExample yolov4 ../../../python/examples/yolov4/000000570688.jpg
# The case of yolov4 needs to specify a picture as input
```
### Customization guidance
The above example is running in CPU mode. If GPU mode is required, there are two options.
The first is that GPU Serving and Java Client are in the same image. After starting the corresponding image, the user needs to move /Serving/java in the java image to the corresponding image.
The second is to deploy GPU Serving and Java Client separately. If they are on the same host, you can learn the IP address of the corresponding container through ifconfig, and then when you connect to client.connect in `examples/src/main/java/PaddleServingClientExample.java` Make changes to the endpoint, and then compile it again. Or select `--net=host` to bind the network device of docker and host when docker starts, so that it can run directly without customizing java code.
**It should be noted that in the example, all models need to use `--use_multilang` to start GRPC multi-programming language support, and the port number is 9393. If you need another port, you need to modify it in the java file**
**Currently Serving has launched the Pipeline mode (see [Pipeline Serving](../doc/PIPELINE_SERVING.md) for details). The next version (0.4.1) of the Pipeline Serving Client for Java will be released. **
java/README_CN.md
浏览文件 @ fcc97c73
## Java 示例
## 用于Paddle Serving的Java客户端
([English](./README.md)|简体中文)
### 开发环境
为了方便用户使用java进行开发，我们提供了编译好的Serving工程放置在java镜像当中，获取镜像并进入开发环境的方式是
```
docker pull hub.baidubce.com/paddlepaddle/serving:0.4.0-java
docker run --rm -dit --name java_serving hub.baidubce.com/paddlepaddle/serving:0.4.0-java
docker exec -it java_serving bash
cd Serving/java
```
Serving文件夹是镜像生成时的develop分支工程目录，需要git pull 到最新版本，或者git checkout 到想要的分支。
### 安装客户端依赖
由于依赖库数量庞大，因此镜像已经在生成时编译过一次，用户执行以下操作即可
```
mvn compile
mvn install
......@@ -11,16 +29,47 @@ mvn install
### 启动服务端
以fit_a_line模型为例
以fit_a_line模型为例，服务端启动
```
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9393 --use_multilang #CPU
python -m paddle_serving_server_gpu.serve --model uci_housing_model --thread 10 --port 9393 --use_multilang #GPU
cd ../../python/examples/fit_a_line
sh get_data.sh
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9393 --use_multilang &
```
### 客户端预测
客户端预测
```
cd ../../../java/examples/target
java -cp paddle-serving-sdk-java-examples-0.0.1-jar-with-dependencies.jar PaddleServingClientExample fit_a_line
```
java示例中还包含了bert、model_enaemble、asyn_predict、batch_predict、cube_local、cube_quant、yolov4模型的预测客户端。
以yolov4为例子，服务端启动
```
python -m paddle_serving_app.package --get_model yolov4
tar -xzvf yolov4.tar.gz
python -m paddle_serving_server_gpu.serve --model yolov4_model --port 9393 --gpu_ids 0 --use_multilang & #需要在GPU Docker当中执行，否则要使用CPU的执行方式。
```
客户端预测
```
# in /Serving/java/examples/target
java -cp paddle-serving-sdk-java-examples-0.0.1-jar-with-dependencies.jar PaddleServingClientExample yolov4 ../../../python/examples/yolov4/000000570688.jpg
# yolov4的案例需要指定一个图片作为输入
```
### 二次开发指导
上述示例是在CPU模式下运行，如果需要GPU模式，可以有两种选择。
第一种是GPU Serving和Java Client在同一个镜像，需要用户在启动对应的镜像后，把java镜像当中的/Serving/java移动到对应的镜像中。
第二种是GPU Serving和Java Client分开部署，如果在同一台宿主机，可以通过ifconfig了解对应容器的IP地址，然后在`examples/src/main/java/PaddleServingClientExample.java`当中对client.connect时的endpoint做修改，然后再编译一次。 或者在docker启动时选择 `--net=host`来绑定docker和宿主机的网络设备，这样不需要定制java代码可以直接运行。
**需要注意的是，在示例中，所有模型都需要使用`--use_multilang`来启动GRPC多编程语言支持，以及端口号都是9393，如果需要别的端口，需要在java文件里修改**
**目前Serving已推出Pipeline模式（详见[Pipeline Serving](../doc/PIPELINE_SERVING_CN.md)），下个版本（0.4.1）面向Java的Pipeline Serving Client将会发布，敬请期待。**
paddle_inference/inferencer-fluid-cpu/include/fluid_cpu_engine.h
浏览文件 @ fcc97c73
......@@ -13,6 +13,7 @@
// limitations under the License.
#pragma once
#include <pthread.h>
#include <fstream>
#include <map>
......@@ -28,6 +29,7 @@ namespace paddle_serving {
namespace fluid_cpu {
using configure::SigmoidConf;
class AutoLock {
public:
explicit AutoLock(pthread_mutex_t& mutex) : _mut(mutex) {
......@@ -528,60 +530,7 @@ class FluidCpuAnalysisDirWithSigmoidCore : public FluidCpuWithSigmoidCore {
return 0;
}
};
class FluidCpuAnalysisEncryptCore : public FluidFamilyCore {
public:
void ReadBinaryFile(const std::string& filename, std::string* contents) {
std::ifstream fin(filename, std::ios::in | std::ios::binary);
fin.seekg(0, std::ios::end);
contents->clear();
contents->resize(fin.tellg());
fin.seekg(0, std::ios::beg);
fin.read(&(contents->at(0)), contents->size());
fin.close();
}
int create(const predictor::InferEngineCreationParams& params) {
std::string data_path = params.get_path();
if (access(data_path.c_str(), F_OK) == -1) {
LOG(ERROR) << "create paddle predictor failed, path note exits: "
<< data_path;
return -1;
}
std::string model_buffer, params_buffer, key_buffer;
ReadBinaryFile(data_path + "encrypt_model", &model_buffer);
ReadBinaryFile(data_path + "encrypt_params", &params_buffer);
ReadBinaryFile(data_path + "key", &key_buffer);
VLOG(2) << "prepare for encryption model";
auto cipher = paddle::MakeCipher("");
std::string real_model_buffer = cipher->Decrypt(model_buffer, key_buffer);
std::string real_params_buffer = cipher->Decrypt(params_buffer, key_buffer);
paddle::AnalysisConfig analysis_config;
analysis_config.SetModelBuffer(&real_model_buffer[0],
real_model_buffer.size(),
&real_params_buffer[0],
real_params_buffer.size());
analysis_config.DisableGpu();
analysis_config.SetCpuMathLibraryNumThreads(1);
if (params.enable_memory_optimization()) {
analysis_config.EnableMemoryOptim();
}
analysis_config.SwitchSpecifyInputNames(true);
AutoLock lock(GlobalPaddleCreateMutex::instance());
VLOG(2) << "decrypt model file sucess";
_core =
paddle::CreatePaddlePredictor<paddle::AnalysisConfig>(analysis_config);
if (NULL == _core.get()) {
LOG(ERROR) << "create paddle predictor failed, path: " << data_path;
return -1;
}
VLOG(2) << "create paddle predictor sucess, path: " << data_path;
return 0;
}
};
} // namespace fluid_cpu
} // namespace paddle_serving
} // namespace baidu
paddle_inference/inferencer-fluid-cpu/src/fluid_cpu_engine.cpp
浏览文件 @ fcc97c73
......@@ -52,13 +52,6 @@ REGIST_FACTORY_OBJECT_IMPL_WITH_NAME(
::baidu::paddle_serving::predictor::InferEngine,
"FLUID_CPU_NATIVE_DIR_SIGMOID");
#if 1
REGIST_FACTORY_OBJECT_IMPL_WITH_NAME(
::baidu::paddle_serving::predictor::FluidInferEngine<
FluidCpuAnalysisEncryptCore>,
::baidu::paddle_serving::predictor::InferEngine,
"FLUID_CPU_ANALYSIS_ENCRYPT");
#endif
} // namespace fluid_cpu
} // namespace paddle_serving
} // namespace baidu
paddle_inference/inferencer-fluid-gpu/include/fluid_gpu_engine.h
浏览文件 @ fcc97c73
......@@ -25,6 +25,7 @@
#include "core/configure/inferencer_configure.pb.h"
#include "core/predictor/framework/infer.h"
#include "paddle_inference_api.h" // NOLINT
DECLARE_int32(gpuid);
namespace baidu {
......@@ -590,60 +591,6 @@ class FluidGpuAnalysisDirWithSigmoidCore : public FluidGpuWithSigmoidCore {
}
};
class FluidGpuAnalysisEncryptCore : public FluidFamilyCore {
public:
void ReadBinaryFile(const std::string& filename, std::string* contents) {
std::ifstream fin(filename, std::ios::in | std::ios::binary);
fin.seekg(0, std::ios::end);
contents->clear();
contents->resize(fin.tellg());
fin.seekg(0, std::ios::beg);
fin.read(&(contents->at(0)), contents->size());
fin.close();
}
int create(const predictor::InferEngineCreationParams& params) {
std::string data_path = params.get_path();
if (access(data_path.c_str(), F_OK) == -1) {
LOG(ERROR) << "create paddle predictor failed, path note exits: "
<< data_path;
return -1;
}
std::string model_buffer, params_buffer, key_buffer;
ReadBinaryFile(data_path + "encrypt_model", &model_buffer);
ReadBinaryFile(data_path + "encrypt_params", &params_buffer);
ReadBinaryFile(data_path + "key", &key_buffer);
VLOG(2) << "prepare for encryption model";
auto cipher = paddle::MakeCipher("");
std::string real_model_buffer = cipher->Decrypt(model_buffer, key_buffer);
std::string real_params_buffer = cipher->Decrypt(params_buffer, key_buffer);
paddle::AnalysisConfig analysis_config;
analysis_config.SetModelBuffer(&real_model_buffer[0],
real_model_buffer.size(),
&real_params_buffer[0],
real_params_buffer.size());
analysis_config.EnableUseGpu(100, FLAGS_gpuid);
analysis_config.SetCpuMathLibraryNumThreads(1);
if (params.enable_memory_optimization()) {
analysis_config.EnableMemoryOptim();
}
analysis_config.SwitchSpecifyInputNames(true);
AutoLock lock(GlobalPaddleCreateMutex::instance());
VLOG(2) << "decrypt model file sucess";
_core =
paddle::CreatePaddlePredictor<paddle::AnalysisConfig>(analysis_config);
if (NULL == _core.get()) {
LOG(ERROR) << "create paddle predictor failed, path: " << data_path;
return -1;
}
VLOG(2) << "create paddle predictor sucess, path: " << data_path;
return 0;
}
};
} // namespace fluid_gpu
} // namespace paddle_serving
} // namespace baidu
paddle_inference/inferencer-fluid-gpu/src/fluid_gpu_engine.cpp
浏览文件 @ fcc97c73
......@@ -54,12 +54,6 @@ REGIST_FACTORY_OBJECT_IMPL_WITH_NAME(
::baidu::paddle_serving::predictor::InferEngine,
"FLUID_GPU_NATIVE_DIR_SIGMOID");
REGIST_FACTORY_OBJECT_IMPL_WITH_NAME(
::baidu::paddle_serving::predictor::FluidInferEngine<
FluidGpuAnalysisEncryptCore>,
::baidu::paddle_serving::predictor::InferEngine,
"FLUID_GPU_ANALYSIS_ENCRPT")
} // namespace fluid_gpu
} // namespace paddle_serving
} // namespace baidu
python/examples/criteo_ctr_with_cube/README.md 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
## Criteo CTR with Sparse Parameter Indexing Service
([简体中文](./README_CN.md)|English)
### Get Sample Dataset
go to directory `python/examples/criteo_ctr_with_cube`
```
sh get_data.sh
```
### Download Model and Sparse Parameter Sequence Files
```
wget https://paddle-serving.bj.bcebos.com/unittest/ctr_cube_unittest.tar.gz
tar xf ctr_cube_unittest.tar.gz
mv models/ctr_client_conf ./
mv models/ctr_serving_model_kv ./
mv models/data ./cube/
```
the model will be in ./ctr_server_model_kv and ./ctr_client_config.
### Start Sparse Parameter Indexing Service
```
wget https://paddle-serving.bj.bcebos.com/others/cube_app.tar.gz
tar xf cube_app.tar.gz
mv cube_app/cube* ./cube/
sh cube_prepare.sh &
```
Here, the sparse parameter is loaded by cube sparse parameter indexing service Cube.
### Start RPC Predictor, the number of serving thread is 4（configurable in test_server.py）
```
python test_server.py ctr_serving_model_kv
```
### Run Prediction
```
python test_client.py ctr_client_conf/serving_client_conf.prototxt ./raw_data
```
### Benchmark
CPU ：Intel(R) Xeon(R) CPU 6148 @ 2.40GHz
Model ：[Criteo CTR](https://github.com/PaddlePaddle/Serving/blob/develop/python/examples/criteo_ctr_with_cube/network_conf.py)
server core/thread num ： 4/8
Run
```
bash benchmark.sh
```
1000 batches will be sent by every client
| client thread num | prepro | client infer | op0 | op1 | op2 | postpro | avg_latency | qps |
| ------------------ | ------ | ------------ | ------ | ----- | ------ | ------- | ----- | ----- |
| 1 | 0.035 | 1.596 | 0.021 | 0.518 | 0.0024 | 0.0025 | 6.774 | 147.7 |
| 2 | 0.034 | 1.780 | 0.027 | 0.463 | 0.0020 | 0.0023 | 6.931 | 288.3 |
| 4 | 0.038 | 2.954 | 0.025 | 0.455 | 0.0019 | 0.0027 | 8.378 | 477.5 |
| 8 | 0.044 | 8.230 | 0.028 | 0.464 | 0.0023 | 0.0034 | 14.191 | 563.8 |
| 16 | 0.048 | 21.037 | 0.028 | 0.455 | 0.0025 | 0.0041 | 27.236 | 587.5 |
the average latency of threads
![avg cost](../../../doc/criteo-cube-benchmark-avgcost.png)
The QPS is
![qps](../../../doc/criteo-cube-benchmark-qps.png)
python/examples/criteo_ctr_with_cube/README_CN.md 已删除 100644 → 0
浏览文件 @ 6bf4c3a8
## 带稀疏参数索引服务的CTR预测服务
(简体中文|[English](./README.md))
### 获取样例数据
进入目录 `python/examples/criteo_ctr_with_cube`
```
sh get_data.sh
```
### 下载模型和稀疏参数序列文件
```
wget https://paddle-serving.bj.bcebos.com/unittest/ctr_cube_unittest.tar.gz
tar xf ctr_cube_unittest.tar.gz
mv models/ctr_client_conf ./
mv models/ctr_serving_model_kv ./
mv models/data ./cube/
```
执行脚本后会在当前目录有ctr_server_model_kv和ctr_client_config文件夹。
### 启动稀疏参数索引服务
```
wget https://paddle-serving.bj.bcebos.com/others/cube_app.tar.gz
tar xf cube_app.tar.gz
mv cube_app/cube* ./cube/
sh cube_prepare.sh &
```
此处，模型当中的稀疏参数会被存放在稀疏参数索引服务Cube当中。
### 启动RPC预测服务，服务端线程数为4（可在test_server.py配置）
```
python test_server.py ctr_serving_model_kv
```
### 执行预测
```
python test_client.py ctr_client_conf/serving_client_conf.prototxt ./raw_data
```
### Benchmark
设备 ：Intel(R) Xeon(R) CPU 6148 @ 2.40GHz
模型 ：[Criteo CTR](https://github.com/PaddlePaddle/Serving/blob/develop/python/examples/criteo_ctr_with_cube/network_conf.py)
server core/thread num ： 4/8
执行
```
bash benchmark.sh
```
客户端每个线程会发送1000个batch
| client thread num | prepro | client infer | op0 | op1 | op2 | postpro | avg_latency | qps |
| ------------------ | ------ | ------------ | ------ | ----- | ------ | ------- | ----- | ----- |
| 1 | 0.035 | 1.596 | 0.021 | 0.518 | 0.0024 | 0.0025 | 6.774 | 147.7 |
| 2 | 0.034 | 1.780 | 0.027 | 0.463 | 0.0020 | 0.0023 | 6.931 | 288.3 |
| 4 | 0.038 | 2.954 | 0.025 | 0.455 | 0.0019 | 0.0027 | 8.378 | 477.5 |
| 8 | 0.044 | 8.230 | 0.028 | 0.464 | 0.0023 | 0.0034 | 14.191 | 563.8 |
| 16 | 0.048 | 21.037 | 0.028 | 0.455 | 0.0025 | 0.0041 | 27.236 | 587.5 |
平均每个线程耗时图如下
![avg cost](../../../doc/criteo-cube-benchmark-avgcost.png)
每个线程QPS耗时如下
![qps](../../../doc/criteo-cube-benchmark-qps.png)
python/examples/criteo_ctr_with_cube/args.py 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
import argparse
def parse_args():
parser = argparse.ArgumentParser(description="PaddlePaddle CTR example")
parser.add_argument(
'--train_data_path',
type=str,
default='./data/raw/train.txt',
help="The path of training dataset")
parser.add_argument(
'--sparse_only',
type=bool,
default=False,
help="Whether we use sparse features only")
parser.add_argument(
'--test_data_path',
type=str,
default='./data/raw/valid.txt',
help="The path of testing dataset")
parser.add_argument(
'--batch_size',
type=int,
default=1000,
help="The size of mini-batch (default:1000)")
parser.add_argument(
'--embedding_size',
type=int,
default=10,
help="The size for embedding layer (default:10)")
parser.add_argument(
'--num_passes',
type=int,
default=10,
help="The number of passes to train (default: 10)")
parser.add_argument(
'--model_output_dir',
type=str,
default='models',
help='The path for model to store (default: models)')
parser.add_argument(
'--sparse_feature_dim',
type=int,
default=1000001,
help='sparse feature hashing space for index processing')
parser.add_argument(
'--is_local',
type=int,
default=1,
help='Local train or distributed train (default: 1)')
parser.add_argument(
'--cloud_train',
type=int,
default=0,
help='Local train or distributed train on paddlecloud (default: 0)')
parser.add_argument(
'--async_mode',
action='store_true',
default=False,
help='Whether start pserver in async mode to support ASGD')
parser.add_argument(
'--no_split_var',
action='store_true',
default=False,
help='Whether split variables into blocks when update_method is pserver')
parser.add_argument(
'--role',
type=str,
default='pserver', # trainer or pserver
help='The path for model to store (default: models)')
parser.add_argument(
'--endpoints',
type=str,
default='127.0.0.1:6000',
help='The pserver endpoints, like: 127.0.0.1:6000,127.0.0.1:6001')
parser.add_argument(
'--current_endpoint',
type=str,
default='127.0.0.1:6000',
help='The path for model to store (default: 127.0.0.1:6000)')
parser.add_argument(
'--trainer_id',
type=int,
default=0,
help='The path for model to store (default: models)')
parser.add_argument(
'--trainers',
type=int,
default=1,
help='The num of trianers, (default: 1)')
return parser.parse_args()
python/examples/criteo_ctr_with_cube/benchmark.py 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# -*- coding: utf-8 -*-
#
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
from paddle_serving_client import Client
import sys
import os
import criteo as criteo
import time
from paddle_serving_client.utils import MultiThreadRunner
from paddle_serving_client.utils import benchmark_args
from paddle_serving_client.metric import auc
py_version = sys.version_info[0]
args = benchmark_args()
def single_func(idx, resource):
client = Client()
print([resource["endpoint"][idx % len(resource["endpoint"])]])
client.load_client_config('ctr_client_conf/serving_client_conf.prototxt')
client.connect(['127.0.0.1:9292'])
batch = 1
buf_size = 100
dataset = criteo.CriteoDataset()
dataset.setup(1000001)
test_filelists = [
"./raw_data/part-%d" % x for x in range(len(os.listdir("./raw_data")))
]
reader = dataset.infer_reader(test_filelists[len(test_filelists) - 40:],
batch, buf_size)
if args.request == "rpc":
fetch = ["prob"]
start = time.time()
itr = 1000
for ei in range(itr):
if args.batch_size > 0:
feed_batch = []
for bi in range(args.batch_size):
if py_version == 2:
data = reader().next()
else:
data = reader().__next__()
feed_dict = {}
feed_dict['dense_input'] = data[0][0]
for i in range(1, 27):
feed_dict["embedding_{}.tmp_0".format(i - 1)] = data[0][
i]
feed_batch.append(feed_dict)
result = client.predict(feed=feed_batch, fetch=fetch)
else:
print("unsupport batch size {}".format(args.batch_size))
elif args.request == "http":
raise ("Not support http service.")
end = time.time()
qps = itr * args.batch_size / (end - start)
return [[end - start, qps]]
if __name__ == '__main__':
multi_thread_runner = MultiThreadRunner()
endpoint_list = ["127.0.0.1:9292"]
#result = single_func(0, {"endpoint": endpoint_list})
start = time.time()
result = multi_thread_runner.run(single_func, args.thread,
{"endpoint": endpoint_list})
end = time.time()
total_cost = end - start
avg_cost = 0
qps = 0
for i in range(args.thread):
avg_cost += result[0][i * 2 + 0]
qps += result[0][i * 2 + 1]
avg_cost = avg_cost / args.thread
print("total cost: {}".format(total_cost))
print("average total cost {} s.".format(avg_cost))
print("qps {} ins/s".format(qps))
python/examples/criteo_ctr_with_cube/benchmark.sh 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
rm profile_log
export FLAGS_profile_client=1
export FLAGS_profile_server=1
wget https://paddle-serving.bj.bcebos.com/unittest/ctr_cube_unittest.tar.gz --no-check-certificate
tar xf ctr_cube_unittest.tar.gz
mv models/ctr_client_conf ./
mv models/ctr_serving_model_kv ./
mv models/data ./cube/
wget https://paddle-serving.bj.bcebos.com/others/cube_app.tar.gz --no-check-certificate
tar xf cube_app.tar.gz
mv cube_app/cube* ./cube/
sh cube_prepare.sh &
python test_server.py ctr_serving_model_kv > serving_log 2>&1 &
for thread_num in 1 4 16
do
for batch_size in 1 4 16 64
do
$PYTHONROOT/bin/python benchmark.py --thread $thread_num --batch_size $batch_size --model serving_client_conf/serving_client_conf.prototxt --request rpc > profile 2>&1
echo "batch size : $batch_size"
echo "thread num : $thread_num"
echo "========================================"
echo "batch size : $batch_size" >> profile_log
$PYTHONROOT/bin/python ../util/show_profile.py profile $thread_num >> profile_log
tail -n 3 profile >> profile_log
done
done
ps -ef|grep 'serving'|grep -v grep|cut -c 9-15 | xargs kill -9
python/examples/criteo_ctr_with_cube/benchmark_cube.sh 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
rm profile_log
#wget https://paddle-serving.bj.bcebos.com/unittest/ctr_cube_unittest.tar.gz --no-check-certificate
#tar xf ctr_cube_unittest.tar.gz
mv models/ctr_client_conf ./
mv models/ctr_serving_model_kv ./
mv models/data ./cube/
#wget https://paddle-serving.bj.bcebos.com/others/cube_app.tar.gz --no-check-certificate
#tar xf cube_app.tar.gz
mv cube_app/cube* ./cube/
sh cube_prepare.sh &
cp ../../../build_server/core/cube/cube-api/cube-cli .
python gen_key.py
for thread_num in 1 4 16 32
do
for batch_size in 1000
do
./cube-cli -config_file ./cube/conf/cube.conf -keys key -dict test_dict -thread_num $thread_num --batch $batch_size > profile 2>&1
echo "batch size : $batch_size"
echo "thread num : $thread_num"
echo "========================================"
echo "batch size : $batch_size" >> profile_log
echo "thread num : $thread_num" >> profile_log
tail -n 8 profile >> profile_log
done
done
ps -ef|grep 'cube'|grep -v grep|cut -c 9-15 | xargs kill -9
python/examples/criteo_ctr_with_cube/clean.sh 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
ps -ef | grep cube | awk {'print $2'} | xargs kill -9
rm -rf cube/cube_data cube/data cube/log* cube/nohup* cube/output/ cube/donefile cube/input cube/monitor cube/cube-builder.INFO
ps -ef | grep test | awk {'print $2'} | xargs kill -9
ps -ef | grep serving | awk {'print $2'} | xargs kill -9
python/examples/criteo_ctr_with_cube/criteo.py 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import sys
class CriteoDataset(object):
def setup(self, sparse_feature_dim):
self.cont_min_ = [0, -3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
self.cont_max_ = [
20, 600, 100, 50, 64000, 500, 100, 50, 500, 10, 10, 10, 50
]
self.cont_diff_ = [
20, 603, 100, 50, 64000, 500, 100, 50, 500, 10, 10, 10, 50
]
self.hash_dim_ = sparse_feature_dim
# here, training data are lines with line_index < train_idx_
self.train_idx_ = 41256555
self.continuous_range_ = range(1, 14)
self.categorical_range_ = range(14, 40)
def _process_line(self, line):
features = line.rstrip('\n').split('\t')
dense_feature = []
sparse_feature = []
for idx in self.continuous_range_:
if features[idx] == '':
dense_feature.append(0.0)
else:
dense_feature.append((float(features[idx]) - self.cont_min_[idx - 1]) / \
self.cont_diff_[idx - 1])
for idx in self.categorical_range_:
sparse_feature.append(
[hash(str(idx) + features[idx]) % self.hash_dim_])
return dense_feature, sparse_feature, [int(features[0])]
def infer_reader(self, filelist, batch, buf_size):
def local_iter():
for fname in filelist:
with open(fname.strip(), "r") as fin:
for line in fin:
dense_feature, sparse_feature, label = self._process_line(
line)
#yield dense_feature, sparse_feature, label
yield [dense_feature] + sparse_feature + [label]
import paddle
batch_iter = paddle.batch(
paddle.reader.shuffle(
local_iter, buf_size=buf_size),
batch_size=batch)
return batch_iter
def generate_sample(self, line):
def data_iter():
dense_feature, sparse_feature, label = self._process_line(line)
feature_name = ["dense_input"]
for idx in self.categorical_range_:
feature_name.append("C" + str(idx - 13))
feature_name.append("label")
yield zip(feature_name, [dense_feature] + sparse_feature + [label])
return data_iter
if __name__ == "__main__":
criteo_dataset = CriteoDataset()
criteo_dataset.setup(int(sys.argv[1]))
criteo_dataset.run_from_stdin()
python/examples/criteo_ctr_with_cube/criteo_reader.py 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
import sys
import paddle.fluid.incubate.data_generator as dg
class CriteoDataset(dg.MultiSlotDataGenerator):
def setup(self, sparse_feature_dim):
self.cont_min_ = [0, -3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
self.cont_max_ = [
20, 600, 100, 50, 64000, 500, 100, 50, 500, 10, 10, 10, 50
]
self.cont_diff_ = [
20, 603, 100, 50, 64000, 500, 100, 50, 500, 10, 10, 10, 50
]
self.hash_dim_ = sparse_feature_dim
# here, training data are lines with line_index < train_idx_
self.train_idx_ = 41256555
self.continuous_range_ = range(1, 14)
self.categorical_range_ = range(14, 40)
def _process_line(self, line):
features = line.rstrip('\n').split('\t')
dense_feature = []
sparse_feature = []
for idx in self.continuous_range_:
if features[idx] == '':
dense_feature.append(0.0)
else:
dense_feature.append((float(features[idx]) - self.cont_min_[idx - 1]) / \
self.cont_diff_[idx - 1])
for idx in self.categorical_range_:
sparse_feature.append(
[hash(str(idx) + features[idx]) % self.hash_dim_])
return dense_feature, sparse_feature, [int(features[0])]
def infer_reader(self, filelist, batch, buf_size):
def local_iter():
for fname in filelist:
with open(fname.strip(), "r") as fin:
for line in fin:
dense_feature, sparse_feature, label = self._process_line(
line)
#yield dense_feature, sparse_feature, label
yield [dense_feature] + sparse_feature + [label]
import paddle
batch_iter = paddle.batch(
paddle.reader.shuffle(
local_iter, buf_size=buf_size),
batch_size=batch)
return batch_iter
def generate_sample(self, line):
def data_iter():
dense_feature, sparse_feature, label = self._process_line(line)
feature_name = ["dense_input"]
for idx in self.categorical_range_:
feature_name.append("C" + str(idx - 13))
feature_name.append("label")
yield zip(feature_name, [dense_feature] + sparse_feature + [label])
return data_iter
if __name__ == "__main__":
criteo_dataset = CriteoDataset()
criteo_dataset.setup(int(sys.argv[1]))
criteo_dataset.run_from_stdin()
python/examples/criteo_ctr_with_cube/cube/conf/cube.conf 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
[{
"dict_name": "test_dict",
"shard": 1,
"dup": 1,
"timeout": 200,
"retry": 3,
"backup_request": 100,
"type": "ipport_list",
"load_balancer": "rr",
"nodes": [{
"ipport_list": "list://127.0.0.1:8027"
}]
}]
python/examples/criteo_ctr_with_cube/cube/conf/gflags.conf 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
--port=8027
--dict_split=1
--in_mem=true
--log_dir=./log/
python/examples/criteo_ctr_with_cube/cube_prepare.sh 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
#! /bin/bash
mkdir -p cube_model
mkdir -p cube/data
./cube/cube-builder -dict_name=test_dict -job_mode=base -last_version=0 -cur_version=0 -depend_version=0 -input_path=./cube_model -output_path=${PWD}/cube/data -shard_num=1 -only_build=false
cd cube && ./cube
python/examples/criteo_ctr_with_cube/cube_quant_prepare.sh 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
#! /bin/bash
mkdir -p cube_model
mkdir -p cube/data
./seq_generator ctr_serving_model/SparseFeatFactors ./cube_model/feature 8
./cube/cube-builder -dict_name=test_dict -job_mode=base -last_version=0 -cur_version=0 -depend_version=0 -input_path=./cube_model -output_path=${PWD}/cube/data -shard_num=1 -only_build=false
mv ./cube/data/0_0/test_dict_part0/* ./cube/data/
cd cube && ./cube
python/examples/criteo_ctr_with_cube/gen_key.py 已删除 100644 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import sys
import random
with open("key", "w") as f:
for i in range(1000000):
f.write("{}\n".format(random.randint(0, 999999)))
python/examples/criteo_ctr_with_cube/get_data.sh 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
wget --no-check-certificate https://paddle-serving.bj.bcebos.com/data/ctr_prediction/ctr_data.tar.gz
tar -zxvf ctr_data.tar.gz
python/examples/criteo_ctr_with_cube/local_train.py 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
from __future__ import print_function
from args import parse_args
import os
import paddle.fluid as fluid
import sys
from network_conf import dnn_model
dense_feature_dim = 13
def train():
args = parse_args()
sparse_only = args.sparse_only
if not os.path.isdir(args.model_output_dir):
os.mkdir(args.model_output_dir)
dense_input = fluid.layers.data(
name="dense_input", shape=[dense_feature_dim], dtype='float32')
sparse_input_ids = [
fluid.layers.data(
name="C" + str(i), shape=[1], lod_level=1, dtype="int64")
for i in range(1, 27)
]
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
#nn_input = None if sparse_only else dense_input
nn_input = dense_input
predict_y, loss, auc_var, batch_auc_var, infer_vars = dnn_model(
nn_input, sparse_input_ids, label, args.embedding_size,
args.sparse_feature_dim)
optimizer = fluid.optimizer.SGD(learning_rate=1e-4)
optimizer.minimize(loss)
exe = fluid.Executor(fluid.CPUPlace())
exe.run(fluid.default_startup_program())
dataset = fluid.DatasetFactory().create_dataset("InMemoryDataset")
dataset.set_use_var([dense_input] + sparse_input_ids + [label])
python_executable = "python"
pipe_command = "{} criteo_reader.py {}".format(python_executable,
args.sparse_feature_dim)
dataset.set_pipe_command(pipe_command)
dataset.set_batch_size(128)
thread_num = 10
dataset.set_thread(thread_num)
whole_filelist = [
"raw_data/part-%d" % x for x in range(len(os.listdir("raw_data")))
]
print(whole_filelist)
dataset.set_filelist(whole_filelist[:100])
dataset.load_into_memory()
fluid.layers.Print(auc_var)
epochs = 1
for i in range(epochs):
exe.train_from_dataset(
program=fluid.default_main_program(), dataset=dataset, debug=True)
print("epoch {} finished".format(i))
import paddle_serving_client.io as server_io
feed_var_dict = {}
feed_var_dict['dense_input'] = dense_input
for i, sparse in enumerate(sparse_input_ids):
feed_var_dict["embedding_{}.tmp_0".format(i)] = sparse
fetch_var_dict = {"prob": predict_y}
feed_kv_dict = {}
feed_kv_dict['dense_input'] = dense_input
for i, emb in enumerate(infer_vars):
feed_kv_dict["embedding_{}.tmp_0".format(i)] = emb
fetch_var_dict = {"prob": predict_y}
server_io.save_model("ctr_serving_model", "ctr_client_conf", feed_var_dict,
fetch_var_dict, fluid.default_main_program())
server_io.save_model("ctr_serving_model_kv", "ctr_client_conf_kv",
feed_kv_dict, fetch_var_dict,
fluid.default_main_program())
if __name__ == '__main__':
train()
python/examples/criteo_ctr_with_cube/network_conf.py 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
import paddle.fluid as fluid
import math
def dnn_model(dense_input, sparse_inputs, label, embedding_size,
sparse_feature_dim):
def embedding_layer(input):
emb = fluid.layers.embedding(
input=input,
is_sparse=True,
is_distributed=False,
size=[sparse_feature_dim, embedding_size],
param_attr=fluid.ParamAttr(
name="SparseFeatFactors",
initializer=fluid.initializer.Uniform()))
x = fluid.layers.sequence_pool(input=emb, pool_type='sum')
return emb, x
def mlp_input_tensor(emb_sums, dense_tensor):
#if isinstance(dense_tensor, fluid.Variable):
# return fluid.layers.concat(emb_sums, axis=1)
#else:
return fluid.layers.concat(emb_sums + [dense_tensor], axis=1)
def mlp(mlp_input):
fc1 = fluid.layers.fc(input=mlp_input,
size=400,
act='relu',
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.Normal(
scale=1 / math.sqrt(mlp_input.shape[1]))))
fc2 = fluid.layers.fc(input=fc1,
size=400,
act='relu',
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.Normal(
scale=1 / math.sqrt(fc1.shape[1]))))
fc3 = fluid.layers.fc(input=fc2,
size=400,
act='relu',
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.Normal(
scale=1 / math.sqrt(fc2.shape[1]))))
pre = fluid.layers.fc(input=fc3,
size=2,
act='softmax',
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.Normal(
scale=1 / math.sqrt(fc3.shape[1]))))
return pre
emb_pair_sums = list(map(embedding_layer, sparse_inputs))
emb_sums = [x[1] for x in emb_pair_sums]
infer_vars = [x[0] for x in emb_pair_sums]
mlp_in = mlp_input_tensor(emb_sums, dense_input)
predict = mlp(mlp_in)
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.reduce_sum(cost)
accuracy = fluid.layers.accuracy(input=predict, label=label)
auc_var, batch_auc_var, auc_states = \
fluid.layers.auc(input=predict, label=label, num_thresholds=2 ** 12, slide_steps=20)
return predict, avg_cost, auc_var, batch_auc_var, infer_vars
python/examples/criteo_ctr_with_cube/test_client.py 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
from paddle_serving_client import Client
import sys
import os
import criteo as criteo
import time
from paddle_serving_client.metric import auc
import numpy as np
py_version = sys.version_info[0]
client = Client()
client.load_client_config(sys.argv[1])
client.connect(["127.0.0.1:9292"])
batch = 1
buf_size = 100
dataset = criteo.CriteoDataset()
dataset.setup(1000001)
test_filelists = ["{}/part-0".format(sys.argv[2])]
reader = dataset.infer_reader(test_filelists, batch, buf_size)
label_list = []
prob_list = []
start = time.time()
for ei in range(10000):
if py_version == 2:
data = reader().next()
else:
data = reader().__next__()
feed_dict = {}
feed_dict['dense_input'] = np.array(data[0][0]).astype("float32").reshape(
1, 13)
feed_dict['dense_input.lod'] = [0, 1]
for i in range(1, 27):
tmp_data = np.array(data[0][i]).astype(np.int64)
feed_dict["embedding_{}.tmp_0".format(i - 1)] = tmp_data.reshape(
(1, len(data[0][i])))
feed_dict["embedding_{}.tmp_0.lod".format(i - 1)] = [0, 1]
fetch_map = client.predict(feed=feed_dict, fetch=["prob"], batch=True)
prob_list.append(fetch_map['prob'][0][1])
label_list.append(data[0][-1][0])
print(auc(label_list, prob_list))
end = time.time()
print(end - start)
python/examples/criteo_ctr_with_cube/test_server.py 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
import os
import sys
from paddle_serving_server import OpMaker
from paddle_serving_server import OpSeqMaker
from paddle_serving_server import Server
op_maker = OpMaker()
read_op = op_maker.create('general_reader')
general_dist_kv_infer_op = op_maker.create('general_dist_kv_infer')
response_op = op_maker.create('general_response')
op_seq_maker = OpSeqMaker()
op_seq_maker.add_op(read_op)
op_seq_maker.add_op(general_dist_kv_infer_op)
op_seq_maker.add_op(response_op)
server = Server()
server.set_op_sequence(op_seq_maker.get_op_sequence())
server.set_num_threads(4)
server.load_model_config(sys.argv[1])
server.prepare_server(
workdir="work_dir1",
port=9292,
device="cpu",
cube_conf="./cube/conf/cube.conf")
server.run_server()
python/examples/criteo_ctr_with_cube/test_server_gpu.py 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
import os
import sys
from paddle_serving_server_gpu import OpMaker
from paddle_serving_server_gpu import OpSeqMaker
from paddle_serving_server_gpu import Server
op_maker = OpMaker()
read_op = op_maker.create('general_reader')
general_dist_kv_infer_op = op_maker.create('general_dist_kv_infer')
response_op = op_maker.create('general_response')
op_seq_maker = OpSeqMaker()
op_seq_maker.add_op(read_op)
op_seq_maker.add_op(general_dist_kv_infer_op)
op_seq_maker.add_op(response_op)
server = Server()
server.set_op_sequence(op_seq_maker.get_op_sequence())
server.set_num_threads(4)
server.load_model_config(sys.argv[1])
server.prepare_server(
workdir="work_dir1",
port=9292,
device="cpu",
cube_conf="./cube/conf/cube.conf")
server.run_server()
python/examples/criteo_ctr_with_cube/test_server_quant.py 已删除 100755 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
import os
import sys
from paddle_serving_server import OpMaker
from paddle_serving_server import OpSeqMaker
from paddle_serving_server import Server
op_maker = OpMaker()
read_op = op_maker.create('general_reader')
general_dist_kv_infer_op = op_maker.create('general_dist_kv_quant_infer')
response_op = op_maker.create('general_response')
op_seq_maker = OpSeqMaker()
op_seq_maker.add_op(read_op)
op_seq_maker.add_op(general_dist_kv_infer_op)
op_seq_maker.add_op(response_op)
server = Server()
server.set_op_sequence(op_seq_maker.get_op_sequence())
server.set_num_threads(4)
server.load_model_config(sys.argv[1])
server.prepare_server(
workdir="work_dir1",
port=9292,
device="cpu",
cube_conf="./cube/conf/cube.conf")
server.run_server()
python/examples/encryption/README.md 已删除 100644 → 0
浏览文件 @ 6bf4c3a8
# Encryption Model Prediction
([简体中文](README_CN.md)|English)
## Get Origin Model
The example uses the model file of the fit_a_line example as a origin model
```
sh get_data.sh
```
## Encrypt Model
```
python encrypt.py
```
The key is stored in the `key` file, and the encrypted model file and server-side configuration file are stored in the `encrypt_server` directory.
client-side configuration file are stored in the `encrypt_client` directory.
## Start Encryption Service
CPU Service
```
python -m paddle_serving_server.serve --model encrypt_server/ --port 9300 --use_encryption_model
```
GPU Service
```
python -m paddle_serving_server_gpu.serve --model encrypt_server/ --port 9300 --use_encryption_model --gpu_ids 0
```
## Prediction
```
python test_client.py uci_housing_client/serving_client_conf.prototxt
```
python/examples/encryption/README_CN.md 已删除 100644 → 0
浏览文件 @ 6bf4c3a8
# 加密模型预测
(简体中文|[English](README.md))
## 获取明文模型
示例中使用fit_a_line示例的模型文件作为明文模型
```
sh get_data.sh
```
## 模型加密
```
python encrypt.py
```
密钥保存在`key`文件中，加密模型文件以及server端配置文件保存在`encrypt_server`目录下，client端配置文件保存在`encrypt_client`目录下。
## 启动加密预测服务
CPU预测服务
```
python -m paddle_serving_server.serve --model encrypt_server/ --port 9300 --use_encryption_model
```
GPU预测服务
```
python -m paddle_serving_server_gpu.serve --model encrypt_server/ --port 9300 --use_encryption_model --gpu_ids 0
```
## 预测
```
python test_client.py uci_housing_client/serving_client_conf.prototxt
```
python/examples/encryption/encrypt.py 已删除 100644 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from paddle_serving_client.io import inference_model_to_serving
def serving_encryption():
inference_model_to_serving(
dirname="./uci_housing_model",
serving_server="encrypt_server",
serving_client="encrypt_client",
encryption=True)
if __name__ == "__main__":
serving_encryption()
python/examples/encryption/get_data.sh 已删除 100644 → 0
浏览文件 @ 6bf4c3a8
wget --no-check-certificate https://paddle-serving.bj.bcebos.com/uci_housing_example/encrypt.tar.gz
tar -xzf encrypt.tar.gz
cp -rvf ../fit_a_line/uci_housing_model .
cp -rvf ../fit_a_line/uci_housing_client .
python/examples/encryption/test_client.py 已删除 100644 → 0
浏览文件 @ 6bf4c3a8
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=doc-string-missing
from paddle_serving_client import Client
import sys
client = Client()
client.load_client_config(sys.argv[1])
client.use_key("./key")
client.connect(["127.0.0.1:9300"], encryption=True)
import paddle
test_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.uci_housing.test(), buf_size=500),
batch_size=1)
for data in test_reader():
fetch_map = client.predict(feed={"x": data[0][0]}, fetch=["price"])
print("{} {}".format(fetch_map["price"][0], data[0][1][0]))
python/examples/fit_a_line/README.md
浏览文件 @ fcc97c73
......@@ -14,12 +14,6 @@ sh get_data.sh
### Start server
``` shell
python test_server.py uci_housing_model/
```
You can also start the default RPC service with the following line of code:
```shell
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9393
```
......@@ -40,7 +34,7 @@ python test_client.py uci_housing_client/serving_client_conf.prototxt
Start a web service with default web service hosting modules:
``` shell
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9393 --name uci
python test_server.py
```
### Client prediction
......
python/examples/fit_a_line/README_CN.md
浏览文件 @ fcc97c73
......@@ -41,7 +41,7 @@ python test_client.py uci_housing_client/serving_client_conf.prototxt
通过下面的一行代码开启默认web服务：
``` shell
python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9393 --name uci
python test_server.py
```
### 客户端预测
......
python/examples/fit_a_line/test_client.py
浏览文件 @ fcc97c73
......@@ -15,6 +15,7 @@
from paddle_serving_client import Client
import sys
import numpy as np
client = Client()
client.load_client_config(sys.argv[1])
......@@ -27,7 +28,6 @@ test_reader = paddle.batch(
batch_size=1)
for data in test_reader():
import numpy as np
new_data = np.zeros((1, 1, 13)).astype("float32")
new_data[0] = data[0][0]
fetch_map = client.predict(
......
python/examples/fit_a_line/test_server.py
浏览文件 @ fcc97c73
......@@ -13,24 +13,24 @@
# limitations under the License.
# pylint: disable=doc-string-missing
import os
import sys
from paddle_serving_server import OpMaker
from paddle_serving_server import OpSeqMaker
from paddle_serving_server import Server
from paddle_serving_server.web_service import WebService
import numpy as np
op_maker = OpMaker()
read_op = op_maker.create('general_reader')
general_infer_op = op_maker.create('general_infer')
response_op = op_maker.create('general_response')
op_seq_maker = OpSeqMaker()
op_seq_maker.add_op(read_op)
op_seq_maker.add_op(general_infer_op)
op_seq_maker.add_op(response_op)
class UciService(WebService):
def preprocess(self, feed=[], fetch=[]):
feed_batch = []
is_batch = True
new_data = np.zeros((len(feed), 1, 13)).astype("float32")
for i, ins in enumerate(feed):
nums = np.array(ins["x"]).reshape(1, 1, 13)
new_data[i] = nums
feed = {"x": new_data}
return feed, fetch, is_batch
server = Server()
server.set_op_sequence(op_seq_maker.get_op_sequence())
server.load_model_config(sys.argv[1])
server.prepare_server(workdir="work_dir1", port=9393, device="cpu")
server.run_server()
uci_service = UciService(name="uci")
uci_service.load_model_config("uci_housing_model")
uci_service.prepare_server(workdir="workdir", port=9292)
uci_service.run_rpc_service()
uci_service.run_web_service()
python/paddle_serving_client/__init__.py
浏览文件 @ fcc97c73
......@@ -13,19 +13,16 @@
# limitations under the License.
# pylint: disable=doc-string-missing
import paddle_serving_client
import os
from .proto import sdk_configure_pb2 as sdk
from .proto import general_model_config_pb2 as m_config
import google.protobuf.text_format
import numpy as np
import time
import sys
import requests
import json
import base64
import numpy as np
import paddle_serving_client
import google.protobuf.text_format
import grpc
from .proto import sdk_configure_pb2 as sdk
from .proto import general_model_config_pb2 as m_config
from .proto import multi_lang_general_model_service_pb2
sys.path.append(
os.path.join(os.path.abspath(os.path.dirname(__file__)), 'proto'))
......@@ -164,7 +161,6 @@ class Client(object):
self.fetch_names_to_idx_ = {}
self.lod_tensor_set = set()
self.feed_tensor_len = {}
self.key = None
for i, var in enumerate(model_conf.feed_var):
self.feed_names_to_idx_[var.alias_name] = i
......@@ -197,28 +193,7 @@ class Client(object):
else:
self.rpc_timeout_ms = rpc_timeout
def use_key(self, key_filename):
with open(key_filename, "r") as f:
self.key = f.read()
def get_serving_port(self, endpoints):
if self.key is not None:
req = json.dumps({"key": base64.b64encode(self.key)})
else:
req = json.dumps({})
r = requests.post("http://" + endpoints[0], req)
result = r.json()
print(result)
if "endpoint_list" not in result:
raise ValueError("server not ready")
else:
endpoints = [
endpoints[0].split(":")[0] + ":" +
str(result["endpoint_list"][0])
]
return endpoints
def connect(self, endpoints=None, encryption=False):
def connect(self, endpoints=None):
# check whether current endpoint is available
# init from client config
# create predictor here
......@@ -228,8 +203,6 @@ class Client(object):
"You must set the endpoints parameter or use add_variant function to create a variant."
)
else:
if encryption:
endpoints = self.get_serving_port(endpoints)
if self.predictor_sdk_ is None:
self.add_variant('default_tag_{}'.format(id(self)), endpoints,
100)
......
python/paddle_serving_client/io/__init__.py
浏览文件 @ fcc97c73
......@@ -21,9 +21,6 @@ from paddle.fluid.framework import Program
from paddle.fluid import CPUPlace
from paddle.fluid.io import save_inference_model
import paddle.fluid as fluid
from paddle.fluid.core import CipherUtils
from paddle.fluid.core import CipherFactory
from paddle.fluid.core import Cipher
from ..proto import general_model_config_pb2 as model_conf
import os
......@@ -32,10 +29,7 @@ def save_model(server_model_folder,
client_config_folder,
feed_var_dict,
fetch_var_dict,
main_program=None,
encryption=False,
key_len=128,
encrypt_conf=None):
main_program=None):
executor = Executor(place=CPUPlace())
feed_var_names = [feed_var_dict[x].name for x in feed_var_dict]
......@@ -44,29 +38,14 @@ def save_model(server_model_folder,
for key in sorted(fetch_var_dict.keys()):
target_vars.append(fetch_var_dict[key])
target_var_names.append(key)
if not encryption:
save_inference_model(
server_model_folder,
feed_var_names,
target_vars,
executor,
main_program=main_program)
else:
if encrypt_conf == None:
aes_cipher = CipherFactory.create_cipher()
else:
#todo: more encryption algorithms
pass
key = CipherUtils.gen_key_to_file(128, "key")
params = fluid.io.save_persistables(
executor=executor, dirname=None, main_program=main_program)
model = main_program.desc.serialize_to_string()
if not os.path.exists(server_model_folder):
os.makedirs(server_model_folder)
os.chdir(server_model_folder)
aes_cipher.encrypt_to_file(params, key, "encrypt_params")
aes_cipher.encrypt_to_file(model, key, "encrypt_model")
os.chdir("..")
save_inference_model(
server_model_folder,
feed_var_names,
target_vars,
executor,
main_program=main_program)
config = model_conf.GeneralModelConfig()
#int64 = 0; float32 = 1; int32 = 2;
......@@ -137,10 +116,7 @@ def inference_model_to_serving(dirname,
serving_server="serving_server",
serving_client="serving_client",
model_filename=None,
params_filename=None,
encryption=False,
key_len=128,
encrypt_conf=None):
params_filename=None):
place = fluid.CPUPlace()
exe = fluid.Executor(place)
inference_program, feed_target_names, fetch_targets = \
......@@ -151,7 +127,7 @@ def inference_model_to_serving(dirname,
}
fetch_dict = {x.name: x for x in fetch_targets}
save_model(serving_server, serving_client, feed_dict, fetch_dict,
inference_program, encryption, key_len, encrypt_conf)
inference_program)
feed_names = feed_dict.keys()
fetch_names = fetch_dict.keys()
return feed_names, fetch_names
python/paddle_serving_server/__init__.py
浏览文件 @ fcc97c73
......@@ -157,7 +157,6 @@ class Server(object):
self.cur_path = os.getcwd()
self.use_local_bin = False
self.mkl_flag = False
self.encryption_model = False
self.product_name = None
self.container_id = None
self.model_config_paths = None # for multi-model in a workflow
......@@ -198,9 +197,6 @@ class Server(object):
def set_ir_optimize(self, flag=False):
self.ir_optimization = flag
def use_encryption_model(self, flag=False):
self.encryption_model = flag
def set_product_name(self, product_name=None):
if product_name == None:
raise ValueError("product_name can't be None.")
......@@ -236,15 +232,9 @@ class Server(object):
engine.force_update_static_cache = False
if device == "cpu":
if self.encryption_model:
engine.type = "FLUID_CPU_ANALYSIS_ENCRYPT"
else:
engine.type = "FLUID_CPU_ANALYSIS_DIR"
engine.type = "FLUID_CPU_ANALYSIS_DIR"
elif device == "gpu":
if self.encryption_model:
engine.type = "FLUID_GPU_ANALYSIS_ENCRYPT"
else:
engine.type = "FLUID_GPU_ANALYSIS_DIR"
engine.type = "FLUID_GPU_ANALYSIS_DIR"
self.model_toolkit_conf.engines.extend([engine])
......
python/paddle_serving_server/serve.py
浏览文件 @ fcc97c73
......@@ -18,14 +18,8 @@ Usage:
python -m paddle_serving_server.serve --model ./serving_server_model --port 9292
"""
import argparse
import sys
import json
import base64
import time
from multiprocessing import Process
from web_service import WebService, port_is_available
from .web_service import WebService
from flask import Flask, request
from BaseHTTPServer import BaseHTTPRequestHandler, HTTPServer
def parse_args(): # pylint: disable=doc-string-missing
......@@ -59,11 +53,6 @@ def parse_args(): # pylint: disable=doc-string-missing
type=int,
default=512 * 1024 * 1024,
help="Limit sizes of messages")
parser.add_argument(
"--use_encryption_model",
default=False,
action="store_true",
help="Use encryption model")
parser.add_argument(
"--use_multilang",
default=False,
......@@ -82,18 +71,17 @@ def parse_args(): # pylint: disable=doc-string-missing
return parser.parse_args()
def start_standard_model(serving_port): # pylint: disable=doc-string-missing
def start_standard_model(): # pylint: disable=doc-string-missing
args = parse_args()
thread_num = args.thread
model = args.model
port = serving_port
port = args.port
workdir = args.workdir
device = args.device
mem_optim = args.mem_optim_off is False
ir_optim = args.ir_optim
max_body_size = args.max_body_size
use_mkl = args.use_mkl
use_encryption_model = args.use_encryption_model
use_multilang = args.use_multilang
if model == "":
......@@ -123,7 +111,6 @@ def start_standard_model(serving_port): # pylint: disable=doc-string-missing
server.use_mkl(use_mkl)
server.set_max_body_size(max_body_size)
server.set_port(port)
server.use_encryption_model(use_encryption_model)
if args.product_name != None:
server.set_product_name(args.product_name)
if args.container_id != None:
......@@ -134,88 +121,11 @@ def start_standard_model(serving_port): # pylint: disable=doc-string-missing
server.run_server()
class MainService(BaseHTTPRequestHandler):
def get_available_port(self):
default_port = 12000
for i in range(1000):
if port_is_available(default_port + i):
return default_port + i
def start_serving(self):
start_standard_model(serving_port)
def get_key(self, post_data):
if "key" not in post_data:
return False
else:
key = base64.b64decode(post_data["key"])
with open(args.model + "/key", "w") as f:
f.write(key)
return True
def check_key(self, post_data):
if "key" not in post_data:
return False
else:
key = base64.b64decode(post_data["key"])
with open(args.model + "/key", "r") as f:
cur_key = f.read()
return (key == cur_key)
def start(self, post_data):
post_data = json.loads(post_data)
global p_flag
if not p_flag:
if args.use_encryption_model:
print("waiting key for model")
if not self.get_key(post_data):
print("not found key in request")
return False
global serving_port
global p
serving_port = self.get_available_port()
p = Process(target=self.start_serving)
p.start()
time.sleep(3)
if p.is_alive():
p_flag = True
else:
return False
else:
if p.is_alive():
if not self.check_key(post_data):
return False
else:
return False
return True
def do_POST(self):
content_length = int(self.headers['Content-Length'])
post_data = self.rfile.read(content_length)
if self.start(post_data):
response = {"endpoint_list": [serving_port]}
else:
response = {"message": "start serving failed"}
self.send_response(200)
self.send_header('Content-type', 'application/json')
self.end_headers()
self.wfile.write(json.dumps(response))
if __name__ == "__main__":
args = parse_args()
if args.name == "None":
if args.use_encryption_model:
p_flag = False
p = None
serving_port = 0
server = HTTPServer(('localhost', int(args.port)), MainService)
print(
'Starting encryption server, waiting for key from client, use <Ctrl-C> to stop'
)
server.serve_forever()
else:
start_standard_model(args.port)
start_standard_model()
else:
service = WebService(name=args.name)
service.load_model_config(args.model)
......
python/paddle_serving_server/web_service.py
浏览文件 @ fcc97c73
......@@ -25,16 +25,6 @@ from paddle_serving_server import pipeline
from paddle_serving_server.pipeline import Op
def port_is_available(port):
with closing(socket.socket(socket.AF_INET, socket.SOCK_STREAM)) as sock:
sock.settimeout(2)
result = sock.connect_ex(('0.0.0.0', port))
if result != 0:
return True
else:
return False
class WebService(object):
def __init__(self, name="default_service"):
self.name = name
......@@ -68,7 +58,7 @@ class WebService(object):
if os.path.isdir(model_config):
client_config = "{}/serving_server_conf.prototxt".format(
model_config)
elif os.path.isfile(path):
elif os.path.isfile(model_config):
client_config = model_config
model_conf = m_config.GeneralModelConfig()
f = open(client_config, 'r')
......@@ -120,7 +110,7 @@ class WebService(object):
self.mem_optim = mem_optim
self.ir_optim = ir_optim
for i in range(1000):
if port_is_available(default_port + i):
if self.port_is_available(default_port + i):
self.port_list.append(default_port + i)
break
......
python/paddle_serving_server_gpu/__init__.py
浏览文件 @ fcc97c73
......@@ -25,7 +25,9 @@ from .version import serving_server_version
from contextlib import closing
import argparse
import collections
import fcntl
import sys
if sys.platform.startswith('win') is False:
import fcntl
import shutil
import numpy as np
import grpc
......@@ -68,11 +70,6 @@ def serve_args():
type=int,
default=512 * 1024 * 1024,
help="Limit sizes of messages")
parser.add_argument(
"--use_encryption_model",
default=False,
action="store_true",
help="Use encryption model")
parser.add_argument(
"--use_multilang",
default=False,
......@@ -282,8 +279,7 @@ class Server(object):
def set_trt(self):
self.use_trt = True
def _prepare_engine(self, model_config_paths, device, use_encryption_model):
def _prepare_engine(self, model_config_paths, device):
if self.model_toolkit_conf == None:
self.model_toolkit_conf = server_sdk.ModelToolkitConf()
......@@ -305,15 +301,9 @@ class Server(object):
engine.use_trt = self.use_trt
if device == "cpu":
if use_encryption_model:
engine.type = "FLUID_CPU_ANALYSIS_ENCRPT"
else:
engine.type = "FLUID_CPU_ANALYSIS_DIR"
engine.type = "FLUID_CPU_ANALYSIS_DIR"
elif device == "gpu":
if use_encryption_model:
engine.type = "FLUID_GPU_ANALYSIS_ENCRPT"
else:
engine.type = "FLUID_GPU_ANALYSIS_DIR"
engine.type = "FLUID_GPU_ANALYSIS_DIR"
self.model_toolkit_conf.engines.extend([engine])
......@@ -470,7 +460,6 @@ class Server(object):
workdir=None,
port=9292,
device="cpu",
use_encryption_model=False,
cube_conf=None):
if workdir == None:
workdir = "./tmp"
......@@ -484,8 +473,7 @@ class Server(object):
self.set_port(port)
self._prepare_resource(workdir, cube_conf)
self._prepare_engine(self.model_config_paths, device,
use_encryption_model)
self._prepare_engine(self.model_config_paths, device)
self._prepare_infer_service(port)
self.workdir = workdir
......
python/paddle_serving_server_gpu/serve.py
浏览文件 @ fcc97c73
......@@ -19,21 +19,19 @@ Usage:
"""
import argparse
import os
import json
import base64
from multiprocessing import Pool, Process
from paddle_serving_server_gpu import serve_args
from flask import Flask, request
from BaseHTTPServer import BaseHTTPRequestHandler, HTTPServer
def start_gpu_card_model(index, gpuid, port, args): # pylint: disable=doc-string-missing
def start_gpu_card_model(index, gpuid, args): # pylint: disable=doc-string-missing
gpuid = int(gpuid)
device = "gpu"
port = args.port
if gpuid == -1:
device = "cpu"
elif gpuid >= 0:
port = port + index
port = args.port + index
thread_num = args.thread
model = args.model
mem_optim = args.mem_optim_off is False
......@@ -75,20 +73,14 @@ def start_gpu_card_model(index, gpuid, port, args): # pylint: disable=doc-strin
server.set_container_id(args.container_id)
server.load_model_config(model)
server.prepare_server(
workdir=workdir,
port=port,
device=device,
use_encryption_model=args.use_encryption_model)
server.prepare_server(workdir=workdir, port=port, device=device)
if gpuid >= 0:
server.set_gpuid(gpuid)
server.run_server()
def start_multi_card(args, serving_port=None): # pylint: disable=doc-string-missing
def start_multi_card(args): # pylint: disable=doc-string-missing
gpus = ""
if serving_port == None:
serving_port = args.port
if args.gpu_ids == "":
gpus = []
else:
......@@ -105,16 +97,14 @@ def start_multi_card(args, serving_port=None): # pylint: disable=doc-string-mis
env_gpus = []
if len(gpus) <= 0:
print("gpu_ids not set, going to run cpu service.")
start_gpu_card_model(-1, -1, serving_port, args)
start_gpu_card_model(-1, -1, args)
else:
gpu_processes = []
for i, gpu_id in enumerate(gpus):
p = Process(
target=start_gpu_card_model,
args=(
target=start_gpu_card_model, args=(
i,
gpu_id,
serving_port,
args, ))
gpu_processes.append(p)
for p in gpu_processes:
......@@ -123,89 +113,10 @@ def start_multi_card(args, serving_port=None): # pylint: disable=doc-string-mis
p.join()
class MainService(BaseHTTPRequestHandler):
def get_available_port(self):
default_port = 12000
for i in range(1000):
if port_is_available(default_port + i):
return default_port + i
def start_serving(self):
start_multi_card(args, serving_port)
def get_key(self, post_data):
if "key" not in post_data:
return False
else:
key = base64.b64decode(post_data["key"])
with open(args.model + "/key", "w") as f:
f.write(key)
return True
def check_key(self, post_data):
if "key" not in post_data:
return False
else:
key = base64.b64decode(post_data["key"])
with open(args.model + "/key", "r") as f:
cur_key = f.read()
return (key == cur_key)
def start(self, post_data):
post_data = json.loads(post_data)
global p_flag
if not p_flag:
if args.use_encryption_model:
print("waiting key for model")
if not self.get_key(post_data):
print("not found key in request")
return False
global serving_port
global p
serving_port = self.get_available_port()
p = Process(target=self.start_serving)
p.start()
time.sleep(3)
if p.is_alive():
p_flag = True
else:
return False
else:
if p.is_alive():
if not self.check_key(post_data):
return False
else:
return False
return True
def do_POST(self):
content_length = int(self.headers['Content-Length'])
post_data = self.rfile.read(content_length)
if self.start(post_data):
response = {"endpoint_list": [serving_port]}
else:
response = {"message": "start serving failed"}
self.send_response(200)
self.send_header('Content-type', 'application/json')
self.end_headers()
self.wfile.write(json.dumps(response))
if __name__ == "__main__":
args = serve_args()
if args.name == "None":
from .web_service import port_is_available
if args.use_encryption_model:
p_flag = False
p = None
serving_port = 0
server = HTTPServer(('localhost', int(args.port)), MainService)
print(
'Starting encryption server, waiting for key from client, use <Ctrl-C> to stop'
)
server.serve_forever()
else:
start_multi_card(args)
start_multi_card(args)
else:
from .web_service import WebService
web_service = WebService(name=args.name)
......
python/paddle_serving_server_gpu/web_service.py
浏览文件 @ fcc97c73
......@@ -28,16 +28,6 @@ from paddle_serving_server_gpu import pipeline
from paddle_serving_server_gpu.pipeline import Op
def port_is_available(port):
with closing(socket.socket(socket.AF_INET, socket.SOCK_STREAM)) as sock:
sock.settimeout(2)
result = sock.connect_ex(('0.0.0.0', port))
if result != 0:
return True
else:
return False
class WebService(object):
def __init__(self, name="default_service"):
self.name = name
......@@ -74,7 +64,7 @@ class WebService(object):
if os.path.isdir(model_config):
client_config = "{}/serving_server_conf.prototxt".format(
model_config)
elif os.path.isfile(path):
elif os.path.isfile(model_config):
client_config = model_config
model_conf = m_config.GeneralModelConfig()
f = open(client_config, 'r')
......@@ -146,7 +136,7 @@ class WebService(object):
self.port_list = []
default_port = 12000
for i in range(1000):
if port_is_available(default_port + i):
if self.port_is_available(default_port + i):
self.port_list.append(default_port + i)
if len(self.port_list) > len(self.gpus):
break
......
python/pipeline/pipeline_server.py
浏览文件 @ fcc97c73
......@@ -333,7 +333,7 @@ class ServerYamlConfChecker(object):
raise SystemExit("Failed to prepare_server: only one of yml_file"
" or yml_dict can be selected as the parameter.")
if yml_file is not None:
with open(yml_file) as f:
with open(yml_file, encoding='utf-8') as f:
conf = yaml.load(f.read())
elif yml_dict is not None:
conf = yml_dict
......
python/setup.py.server.in
浏览文件 @ fcc97c73
......@@ -29,7 +29,7 @@ util.gen_pipeline_code("paddle_serving_server")
REQUIRED_PACKAGES = [
'six >= 1.10.0', 'protobuf >= 3.11.0', 'grpcio <= 1.33.2', 'grpcio-tools <= 1.33.2',
'paddle_serving_client', 'flask >= 1.1.1', 'paddle_serving_app', 'func_timeout', 'pyyaml'
'flask >= 1.1.1', 'func_timeout', 'pyyaml'
]
packages=['paddle_serving_server',
......
python/setup.py.server_gpu.in
浏览文件 @ fcc97c73
......@@ -31,7 +31,7 @@ util.gen_pipeline_code("paddle_serving_server_gpu")
REQUIRED_PACKAGES = [
'six >= 1.10.0', 'protobuf >= 3.11.0', 'grpcio <= 1.33.2', 'grpcio-tools <= 1.33.2',
'paddle_serving_client', 'flask >= 1.1.1', 'paddle_serving_app', 'func_timeout', 'pyyaml'
'flask >= 1.1.1', 'func_timeout', 'pyyaml'
]
packages=['paddle_serving_server_gpu',
......
tools/Dockerfile.centos6.cuda9.0-cudnn7.devel
浏览文件 @ fcc97c73
......@@ -39,8 +39,6 @@ RUN yum -y install wget && \
make clean && \
echo 'export PATH=/usr/local/python3.6/bin:$PATH' >> /root/.bashrc && \
echo 'export LD_LIBRARY_PATH=/usr/local/python3.6/lib:$LD_LIBRARY_PATH' >> /root/.bashrc && \
pip install requests && \
pip3 install requests && \
source /root/.bashrc && \
cd .. && rm -rf Python-3.6.8* && \
wget https://github.com/protocolbuffers/protobuf/releases/download/v3.11.2/protobuf-all-3.11.2.tar.gz && \
......
tools/Dockerfile.centos6.devel
浏览文件 @ fcc97c73
......@@ -49,8 +49,6 @@ RUN yum -y install wget && \
cd .. && rm -rf protobuf-* && \
yum -y install epel-release && yum -y install patchelf libXext libSM libXrender && \
yum clean all && \
pip install requests && \
pip3 install requests && \
localedef -c -i en_US -f UTF-8 en_US.UTF-8 && \
echo "export LANG=en_US.utf8" >> /root/.bashrc && \
echo "export LANGUAGE=en_US.utf8" >> /root/.bashrc
tools/Dockerfile.ci
浏览文件 @ fcc97c73
......@@ -23,8 +23,7 @@ RUN wget https://dl.google.com/go/go1.14.linux-amd64.tar.gz >/dev/null \
RUN yum -y install python-devel sqlite-devel >/dev/null \
&& curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py >/dev/null \
&& python get-pip.py >/dev/null \
&& rm get-pip.py \
&& pip install requests
&& rm get-pip.py
RUN wget http://nixos.org/releases/patchelf/patchelf-0.10/patchelf-0.10.tar.bz2 \
&& yum -y install bzip2 >/dev/null \
......@@ -35,9 +34,6 @@ RUN wget http://nixos.org/releases/patchelf/patchelf-0.10/patchelf-0.10.tar.bz2
&& cd .. \
&& rm -rf patchelf-0.10*
RUN yum install -y python3 python3-devel \
&& pip3 install requests
RUN wget https://github.com/protocolbuffers/protobuf/releases/download/v3.11.2/protobuf-all-3.11.2.tar.gz && \
tar zxf protobuf-all-3.11.2.tar.gz && \
cd protobuf-3.11.2 && \
......@@ -45,6 +41,8 @@ RUN wget https://github.com/protocolbuffers/protobuf/releases/download/v3.11.2/p
make clean && \
cd .. && rm -rf protobuf-*
RUN yum install -y python3 python3-devel
RUN yum -y update >/dev/null \
&& yum -y install dnf >/dev/null \
&& yum -y install dnf-plugins-core >/dev/null \
......
tools/Dockerfile.cuda10.0-cudnn7.devel
浏览文件 @ fcc97c73
......@@ -30,13 +30,11 @@ RUN wget https://dl.google.com/go/go1.14.linux-amd64.tar.gz >/dev/null \
RUN yum -y install python-devel sqlite-devel \
&& curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py >/dev/null \
&& python get-pip.py >/dev/null \
&& rm get-pip.py \
&& pip install requests
&& rm get-pip.py
RUN yum install -y python3 python3-devel \
&& yum -y install epel-release && yum -y install patchelf libXext libSM libXrender\
&& yum clean all \
&& pip3 install requests
&& yum clean all
RUN localedef -c -i en_US -f UTF-8 en_US.UTF-8 \
&& echo "export LANG=en_US.utf8" >> /root/.bashrc \
......
tools/Dockerfile.cuda9.0-cudnn7.devel
浏览文件 @ fcc97c73
......@@ -29,13 +29,11 @@ RUN wget https://dl.google.com/go/go1.14.linux-amd64.tar.gz >/dev/null \
RUN yum -y install python-devel sqlite-devel \
&& curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py >/dev/null \
&& python get-pip.py >/dev/null \
&& rm get-pip.py \
&& pip install requests
&& rm get-pip.py
RUN yum install -y python3 python3-devel \
&& yum -y install epel-release && yum -y install patchelf libXext libSM libXrender\
&& yum clean all \
&& pip3 install requests
&& yum clean all
RUN localedef -c -i en_US -f UTF-8 en_US.UTF-8 \
&& echo "export LANG=en_US.utf8" >> /root/.bashrc \
......
tools/Dockerfile.devel
浏览文件 @ fcc97c73
......@@ -19,13 +19,11 @@ RUN wget https://dl.google.com/go/go1.14.linux-amd64.tar.gz >/dev/null \
RUN yum -y install python-devel sqlite-devel \
&& curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py >/dev/null \
&& python get-pip.py >/dev/null \
&& rm get-pip.py \
&& pip install requests
&& rm get-pip.py
RUN yum install -y python3 python3-devel \
&& yum -y install epel-release && yum -y install patchelf libXext libSM libXrender\
&& yum clean all \
&& pip3 install requests
&& yum clean all
RUN localedef -c -i en_US -f UTF-8 en_US.UTF-8 \
&& echo "export LANG=en_US.utf8" >> /root/.bashrc \
......
tools/serving_build.sh
浏览文件 @ fcc97c73
......@@ -514,40 +514,6 @@ function python_test_lac() {
cd ..
}
function python_test_encryption(){
#pwd: /Serving/python/examples
cd encryption
sh get_data.sh
local TYPE=$1
export SERVING_BIN=${SERIVNG_WORKDIR}/build-server-${TYPE}/core/general-server/serving
case $TYPE in
CPU)
#check_cmd "python encrypt.py"
#sleep 5
check_cmd "python -m paddle_serving_server.serve --model encrypt_server/ --port 9300 --use_encryption_model > /dev/null &"
sleep 5
check_cmd "python test_client.py encrypt_client/serving_client_conf.prototxt"
kill_server_process
;;
GPU)
#check_cmd "python encrypt.py"
#sleep 5
check_cmd "python -m paddle_serving_server_gpu.serve --model encrypt_server/ --port 9300 --use_encryption_model --gpu_ids 0"
sleep 5
check_cmd "python test_client.py encrypt_client/serving_client_conf.prototxt"
kill_servere_process
;;
*)
echo "error type"
exit 1
;;
esac
echo "encryption $TYPE test finished as expected"
setproxy
unset SERVING_BIN
cd ..
}
function java_run_test() {
# pwd: /Serving
local TYPE=$1
......@@ -563,7 +529,7 @@ function java_run_test() {
cd examples # pwd: /Serving/java/examples
mvn compile > /dev/null
mvn install > /dev/null
# fit_a_line (general, asyn_predict, batch_predict)
cd ../../python/examples/grpc_impl_example/fit_a_line # pwd: /Serving/python/examples/grpc_impl_example/fit_a_line
sh get_data.sh
......@@ -820,7 +786,7 @@ function python_test_pipeline(){
python -m paddle_serving_server.serve --model imdb_cnn_model --port 9292 --workdir test9292 &> cnn.log &
python -m paddle_serving_server.serve --model imdb_bow_model --port 9393 --workdir test9393 &> bow.log &
sleep 5
# test: thread servicer & thread op
cat << EOF > config.yml
rpc_port: 18080
......@@ -994,7 +960,6 @@ function python_run_test() {
python_test_lac $TYPE # pwd: /Serving/python/examples
python_test_multi_process $TYPE # pwd: /Serving/python/examples
python_test_multi_fetch $TYPE # pwd: /Serving/python/examples
python_test_encryption $TYPE # pwd: /Serving/python/examples
python_test_yolov4 $TYPE # pwd: /Serving/python/examples
python_test_grpc_impl $TYPE # pwd: /Serving/python/examples
python_test_resnet50 $TYPE # pwd: /Serving/python/examples
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或