Skip to content

S
Serving

PaddlePaddle / Serving
大约 1 年 前同步成功

通知 186
Star 833
Fork 253
S
Serving
提交 47ae3ad2 编写于 作者: M MRXLT
浏览文件
操作

bug fix

上级 272d3afe 9c71c83e
展开全部 隐藏空白更改
内联 并排
Showing with 1944 addition and 725 deletion
.travis.yml 0 → 100644
浏览文件 @ 47ae3ad2
language: generic
sudo: required
dist: trusty
os:
- linux
env:
- COMPILE_TYPE=CPU DOCKERFILE_CPU=$PWD/tools/Dockerfile.ci
services:
- docker
before_install:
- docker build -f ${DOCKERFILE_CPU} -t serving-img:${COMPILE_TYPE} .
install:
- if [ $COMPILE_TYPE == "CPU" ]; then docker run -it -v $PWD:/Serving serving-img:${COMPILE_TYPE} /bin/bash Serving/tools/serving_check_style.sh ; fi;
- docker run -it -v $PWD:/Serving serving-img:${COMPILE_TYPE} /bin/bash Serving/tools/serving_build.sh $COMPILE_TYPE
core/configure/CMakeLists.txt
浏览文件 @ 47ae3ad2
......@@ -87,4 +87,3 @@ add_custom_command(TARGET general_model_config_py_proto POST_BUILD
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
endif()
endif()
core/configure/include/configure_parser.h
浏览文件 @ 47ae3ad2
......@@ -20,16 +20,16 @@ namespace baidu {
namespace paddle_serving {
namespace configure {
int read_proto_conf(const std::string &conf_full_path,
google::protobuf::Message *conf);
int read_proto_conf(const std::string &conf_path,
const std::string &conf_file,
google::protobuf::Message *conf);
int write_proto_conf(google::protobuf::Message *message,
const std::string &output_path,
const std::string &output_file);
int read_proto_conf(const std::string &conf_full_path,
google::protobuf::Message *conf);
int read_proto_conf(const std::string &conf_path,
const std::string &conf_file,
google::protobuf::Message *conf);
int write_proto_conf(google::protobuf::Message *message,
const std::string &output_path,
const std::string &output_file);
} // namespace configure
} // namespace paddle_serving
......
core/cube/cube-server/src/server.cpp
浏览文件 @ 47ae3ad2
......@@ -12,9 +12,9 @@
// See the License for the specific language governing permissions and
// limitations under the License.
#include "core/cube/cube-server/include/cube/server.h"
#include <brpc/server.h>
#include "core/cube/cube-server/include/cube/framework.h"
#include "core/cube/cube-server/include/cube/server.h"
namespace rec {
namespace mcube {
......
core/general-client/include/general_model.h
浏览文件 @ 47ae3ad2
......@@ -84,19 +84,14 @@ class PredictorClient {
PredictorRes& predict_res, // NOLINT
const int& pid);
std::vector<std::vector<float>> predict(
const std::vector<std::vector<float>>& float_feed,
const std::vector<std::string>& float_feed_name,
const std::vector<std::vector<int64_t>>& int_feed,
const std::vector<std::string>& int_feed_name,
const std::vector<std::string>& fetch_name);
std::vector<std::vector<std::vector<float>>> batch_predict(
int batch_predict(
const std::vector<std::vector<std::vector<float>>>& float_feed_batch,
const std::vector<std::string>& float_feed_name,
const std::vector<std::vector<std::vector<int64_t>>>& int_feed_batch,
const std::vector<std::string>& int_feed_name,
const std::vector<std::string>& fetch_name);
const std::vector<std::string>& fetch_name,
PredictorRes& predict_res_batch, // NOLINT
const int& pid);
private:
PredictorApi _api;
......
core/general-client/src/general_model.cpp
浏览文件 @ 47ae3ad2
......@@ -264,26 +264,22 @@ int PredictorClient::predict(const std::vector<std::vector<float>> &float_feed,
return 0;
}
std::vector<std::vector<std::vector<float>>> PredictorClient::batch_predict(
int PredictorClient::batch_predict(
const std::vector<std::vector<std::vector<float>>> &float_feed_batch,
const std::vector<std::string> &float_feed_name,
const std::vector<std::vector<std::vector<int64_t>>> &int_feed_batch,
const std::vector<std::string> &int_feed_name,
const std::vector<std::string> &fetch_name) {
const std::vector<std::string> &fetch_name,
PredictorRes &predict_res_batch,
const int &pid) {
int batch_size = std::max(float_feed_batch.size(), int_feed_batch.size());
std::vector<std::vector<std::vector<float>>> fetch_result_batch;
if (fetch_name.size() == 0) {
return fetch_result_batch;
}
predict_res_batch._int64_map.clear();
predict_res_batch._float_map.clear();
Timer timeline;
int64_t preprocess_start = timeline.TimeStampUS();
fetch_result_batch.resize(batch_size);
int fetch_name_num = fetch_name.size();
for (int bi = 0; bi < batch_size; bi++) {
fetch_result_batch[bi].resize(fetch_name_num);
}
_api.thrd_clear();
_predictor = _api.fetch_predictor("general_model");
......@@ -371,20 +367,36 @@ std::vector<std::vector<std::vector<float>>> PredictorClient::batch_predict(
} else {
client_infer_end = timeline.TimeStampUS();
postprocess_start = client_infer_end;
for (auto &name : fetch_name) {
predict_res_batch._int64_map[name].resize(batch_size);
predict_res_batch._float_map[name].resize(batch_size);
}
for (int bi = 0; bi < batch_size; bi++) {
for (auto &name : fetch_name) {
int idx = _fetch_name_to_idx[name];
int len = res.insts(bi).tensor_array(idx).data_size();
VLOG(2) << "fetch name: " << name;
VLOG(2) << "tensor data size: " << len;
fetch_result_batch[bi][idx].resize(len);
VLOG(2)
<< "fetch name " << name << " index " << idx << " first data "
<< *(const float *)res.insts(bi).tensor_array(idx).data(0).c_str();
/*
TBA
*/
if (_fetch_name_to_type[name] == 0) {
int len = res.insts(bi).tensor_array(idx).int64_data_size();
VLOG(2) << "fetch tensor : " << name << " type: int64 len : " << len;
predict_res_batch._int64_map[name][bi].resize(len);
VLOG(2) << "fetch name " << name << " index " << idx << " first data "
<< res.insts(bi).tensor_array(idx).int64_data(0);
for (int i = 0; i < len; ++i) {
predict_res_batch._int64_map[name][bi][i] =
res.insts(bi).tensor_array(idx).int64_data(i);
}
} else if (_fetch_name_to_type[name] == 1) {
int len = res.insts(bi).tensor_array(idx).float_data_size();
VLOG(2) << "fetch tensor : " << name
<< " type: float32 len : " << len;
predict_res_batch._float_map[name][bi].resize(len);
VLOG(2) << "fetch name " << name << " index " << idx << " first data "
<< res.insts(bi).tensor_array(idx).float_data(0);
for (int i = 0; i < len; ++i) {
predict_res_batch._float_map[name][bi][i] =
res.insts(bi).tensor_array(idx).float_data(i);
}
}
}
}
postprocess_end = timeline.TimeStampUS();
......@@ -393,6 +405,7 @@ std::vector<std::vector<std::vector<float>>> PredictorClient::batch_predict(
if (FLAGS_profile_client) {
std::ostringstream oss;
oss << "PROFILE\t"
<< "pid:" << pid << "\t"
<< "prepro_0:" << preprocess_start << " "
<< "prepro_1:" << preprocess_end << " "
<< "client_infer_0:" << client_infer_start << " "
......@@ -411,7 +424,7 @@ std::vector<std::vector<std::vector<float>>> PredictorClient::batch_predict(
fprintf(stderr, "%s\n", oss.str().c_str());
}
return fetch_result_batch;
return 0;
}
} // namespace general_model
......
core/general-client/src/general_model_main.cpp
浏览文件 @ 47ae3ad2
......@@ -17,18 +17,18 @@
#include "core/general-client/include/general_model.h"
using namespace std;
using namespace std; // NOLINT
using baidu::paddle_serving::general_model::PredictorClient;
using baidu::paddle_serving::general_model::FetchedMap;
int main(int argc, char * argv[]) {
PredictorClient * client = new PredictorClient();
int main(int argc, char* argv[]) {
PredictorClient* client = new PredictorClient();
client->init("inference.conf");
client->set_predictor_conf("./", "predictor.conf");
client->create_predictor();
std::vector<std::vector<float> > float_feed;
std::vector<std::vector<int64_t> > int_feed;
std::vector<std::vector<float>> float_feed;
std::vector<std::vector<int64_t>> int_feed;
std::vector<std::string> float_feed_name;
std::vector<std::string> int_feed_name = {"words", "label"};
std::vector<std::string> fetch_name = {"cost", "acc", "prediction"};
......@@ -53,13 +53,14 @@ int main(int argc, char * argv[]) {
cin >> label;
int_feed.push_back({label});
FetchedMap result;
client->predict(
float_feed, float_feed_name,
int_feed, int_feed_name, fetch_name,
&result);
client->predict(float_feed,
float_feed_name,
int_feed,
int_feed_name,
fetch_name,
&result);
cout << label << "\t" << result["prediction"][1] << endl;
......
core/general-client/src/pybind_general_model.cpp
浏览文件 @ 47ae3ad2
......@@ -90,12 +90,16 @@ PYBIND11_MODULE(serving_client, m) {
const std::vector<std::vector<std::vector<int64_t>>>
&int_feed_batch,
const std::vector<std::string> &int_feed_name,
const std::vector<std::string> &fetch_name) {
const std::vector<std::string> &fetch_name,
PredictorRes &predict_res_batch,
const int &pid) {
return self.batch_predict(float_feed_batch,
float_feed_name,
int_feed_batch,
int_feed_name,
fetch_name);
fetch_name,
predict_res_batch,
pid);
});
}
......
core/general-server/op/general_copy_op.cpp
浏览文件 @ 47ae3ad2
......@@ -69,8 +69,7 @@ int GeneralCopyOp::inference() {
for (int i = 0; i < out->size(); ++i) {
int64_t *src_ptr = static_cast<int64_t *>(in->at(i).data.data());
out->at(i).data.Resize(
out->at(i).lod[0].back() * sizeof(int64_t));
out->at(i).data.Resize(out->at(i).lod[0].back() * sizeof(int64_t));
out->at(i).shape = {out->at(i).lod[0].back(), 1};
int64_t *tgt_ptr = static_cast<int64_t *>(out->at(i).data.data());
for (int j = 0; j < out->at(i).lod[0].back(); ++j) {
......
core/general-server/op/general_copy_op.h
浏览文件 @ 47ae3ad2
......@@ -24,23 +24,22 @@
#include "paddle_inference_api.h" // NOLINT
#endif
#include <string>
#include "core/predictor/framework/resource.h"
#include "core/general-server/op/general_infer_helper.h"
#include "core/general-server/general_model_service.pb.h"
#include "core/general-server/op/general_infer_helper.h"
#include "core/predictor/framework/resource.h"
namespace baidu {
namespace paddle_serving {
namespace serving {
class GeneralCopyOp :
public baidu::paddle_serving::predictor::OpWithChannel<GeneralBlob> {
class GeneralCopyOp
: public baidu::paddle_serving::predictor::OpWithChannel<GeneralBlob> {
public:
typedef std::vector<paddle::PaddleTensor> TensorVector;
DECLARE_OP(GeneralCopyOp);
int inference();
};
} // namespace serving
......
core/general-server/op/general_infer_op.h
浏览文件 @ 47ae3ad2
......@@ -39,7 +39,6 @@ class GeneralInferOp
DECLARE_OP(GeneralInferOp);
int inference();
};
} // namespace serving
......
core/general-server/op/general_reader_op.cpp
浏览文件 @ 47ae3ad2
......@@ -188,8 +188,7 @@ int GeneralReaderOp::inference() {
for (int j = 0; j < batch_size; ++j) {
int elem_num = req->insts(j).tensor_array(i).int64_data_size();
for (int k = 0; k < elem_num; ++k) {
dst_ptr[offset + k] =
req->insts(j).tensor_array(i).int64_data(k);
dst_ptr[offset + k] = req->insts(j).tensor_array(i).int64_data(k);
}
if (out->at(i).lod.size() == 1) {
offset = out->at(i).lod[0][j + 1];
......@@ -203,8 +202,7 @@ int GeneralReaderOp::inference() {
for (int j = 0; j < batch_size; ++j) {
int elem_num = req->insts(j).tensor_array(i).float_data_size();
for (int k = 0; k < elem_num; ++k) {
dst_ptr[offset + k] =
req->insts(j).tensor_array(i).float_data(k);
dst_ptr[offset + k] = req->insts(j).tensor_array(i).float_data(k);
}
if (out->at(i).lod.size() == 1) {
offset = out->at(i).lod[0][j + 1];
......
core/general-server/op/general_reader_op.h
浏览文件 @ 47ae3ad2
......@@ -24,24 +24,23 @@
#include "paddle_inference_api.h" // NOLINT
#endif
#include <string>
#include "core/predictor/framework/resource.h"
#include "core/general-server/op/general_infer_helper.h"
#include "core/general-server/general_model_service.pb.h"
#include "core/general-server/load_general_model_service.pb.h"
#include "core/general-server/op/general_infer_helper.h"
#include "core/predictor/framework/resource.h"
namespace baidu {
namespace paddle_serving {
namespace serving {
class GeneralReaderOp : public baidu::paddle_serving::predictor::OpWithChannel<
GeneralBlob> {
class GeneralReaderOp
: public baidu::paddle_serving::predictor::OpWithChannel<GeneralBlob> {
public:
typedef std::vector<paddle::PaddleTensor> TensorVector;
DECLARE_OP(GeneralReaderOp);
int inference();
};
} // namespace serving
......
core/general-server/op/general_response_op.cpp
浏览文件 @ 47ae3ad2
......@@ -122,8 +122,7 @@ int GeneralResponseOp::inference() {
} else {
for (int j = 0; j < batch_size; ++j) {
FetchInst *fetch_p = res->mutable_insts(j);
fetch_p->mutable_tensor_array(var_idx)->add_int64_data(
data_ptr[0]);
fetch_p->mutable_tensor_array(var_idx)->add_int64_data(data_ptr[0]);
}
}
}
......@@ -143,16 +142,15 @@ int GeneralResponseOp::inference() {
if (var_size == batch_size) {
for (int j = 0; j < batch_size; ++j) {
for (int k = j * cap; k < (j + 1) * cap; ++k) {
FetchInst * fetch_p = res->mutable_insts(j);
FetchInst *fetch_p = res->mutable_insts(j);
fetch_p->mutable_tensor_array(var_idx)->add_float_data(
data_ptr[k]);
}
}
} else {
for (int j = 0; j < batch_size; ++j) {
FetchInst * fetch_p = res->mutable_insts(j);
fetch_p->mutable_tensor_array(var_idx)->add_float_data(
data_ptr[0]);
FetchInst *fetch_p = res->mutable_insts(j);
fetch_p->mutable_tensor_array(var_idx)->add_float_data(data_ptr[0]);
}
}
}
......
core/general-server/op/general_response_op.h
浏览文件 @ 47ae3ad2
......@@ -39,7 +39,6 @@ class GeneralResponseOp
DECLARE_OP(GeneralResponseOp);
int inference();
};
} // namespace serving
......
core/general-server/op/general_text_reader_op.cpp
浏览文件 @ 47ae3ad2
......@@ -12,11 +12,11 @@
// See the License for the specific language governing permissions and
// limitations under the License.
#include "core/general-server/op/general_text_reader_op.h"
#include <algorithm>
#include <iostream>
#include <memory>
#include <sstream>
#include "core/general-server/op/general_text_reader_op.h"
#include "core/predictor/framework/infer.h"
#include "core/predictor/framework/memory.h"
#include "core/util/include/timer.h"
......@@ -32,7 +32,6 @@ using baidu::paddle_serving::predictor::general_model::Request;
using baidu::paddle_serving::predictor::general_model::FeedInst;
using baidu::paddle_serving::predictor::PaddleGeneralModelConfig;
int GeneralTextReaderOp::inference() {
// reade request from client
const Request *req = dynamic_cast<const Request *>(get_request_message());
......@@ -132,11 +131,9 @@ int GeneralTextReaderOp::inference() {
int64_t *dst_ptr = static_cast<int64_t *>(out->at(i).data.data());
int offset = 0;
for (int j = 0; j < batch_size; ++j) {
for (int k = 0;
k < req->insts(j).tensor_array(i).int_data_size();
for (int k = 0; k < req->insts(j).tensor_array(i).int_data_size();
++k) {
dst_ptr[offset + k] =
req->insts(j).tensor_array(i).int_data(k);
dst_ptr[offset + k] = req->insts(j).tensor_array(i).int_data(k);
}
if (out->at(i).lod.size() == 1) {
offset = out->at(i).lod[0][j + 1];
......@@ -148,11 +145,9 @@ int GeneralTextReaderOp::inference() {
float *dst_ptr = static_cast<float *>(out->at(i).data.data());
int offset = 0;
for (int j = 0; j < batch_size; ++j) {
for (int k = 0;
k < req->insts(j).tensor_array(i).int_data_size();
for (int k = 0; k < req->insts(j).tensor_array(i).int_data_size();
++k) {
dst_ptr[offset + k] =
req->insts(j).tensor_array(i).int_data(k);
dst_ptr[offset + k] = req->insts(j).tensor_array(i).int_data(k);
}
if (out->at(i).lod.size() == 1) {
offset = out->at(i).lod[0][j + 1];
......
core/general-server/op/general_text_reader_op.h
浏览文件 @ 47ae3ad2
......@@ -24,17 +24,17 @@
#include "paddle_inference_api.h" // NOLINT
#endif
#include <string>
#include "core/predictor/framework/resource.h"
#include "core/general-server/op/general_infer_helper.h"
#include "core/general-server/general_model_service.pb.h"
#include "core/general-server/load_general_model_service.pb.h"
#include "core/general-server/op/general_infer_helper.h"
#include "core/predictor/framework/resource.h"
namespace baidu {
namespace paddle_serving {
namespace serving {
class GeneralTextReaderOp :
public baidu::paddle_serving::predictor::OpWithChannel<GeneralBlob> {
class GeneralTextReaderOp
: public baidu::paddle_serving::predictor::OpWithChannel<GeneralBlob> {
public:
typedef std::vector<paddle::PaddleTensor> TensorVector;
......
core/general-server/op/general_text_response_op.h
浏览文件 @ 47ae3ad2
......@@ -40,7 +40,6 @@ class GeneralTextResponseOp
DECLARE_OP(GeneralTextResponseOp);
int inference();
};
} // namespace serving
......
core/general-server/proto/general_model_service.proto
浏览文件 @ 47ae3ad2
......@@ -28,13 +28,9 @@ message Tensor {
repeated int32 shape = 6;
};
message FeedInst {
repeated Tensor tensor_array = 1;
};
message FeedInst { repeated Tensor tensor_array = 1; };
message FetchInst {
repeated Tensor tensor_array = 1;
};
message FetchInst { repeated Tensor tensor_array = 1; };
message Request {
repeated FeedInst insts = 1;
......
core/kvdb/include/kvdb/kvdb_impl.h
浏览文件 @ 47ae3ad2
......@@ -27,11 +27,11 @@
// limitations under the License.
#pragma once
#include <chrono>
#include <chrono> // NOLINT
#include <functional>
#include <memory>
#include <unordered_map>
#include <vector>
#include <functional>
class AbstractKVDB;
class FileReader;
class ParamDict;
......@@ -65,7 +65,7 @@ class FileReader {
std::string data;
FILE *stream = nullptr;
const int max_buffer = 256;
char buffer[max_buffer];
char buffer[max_buffer]; // NOLINT
cmd.append(" 2>&1");
stream = popen(cmd.c_str(), "r");
if (stream) {
......@@ -76,7 +76,8 @@ class FileReader {
return data;
};
std::string cmd = "md5sum " + this->filename_;
// TODO: throw exception if error occurs during execution of shell command
// NOLINT TODO: throw exception if error occurs during execution of shell
// command
std::string md5val = getCmdOut(cmd);
this->time_stamp_ = md5val == this->last_md5_val_
? this->time_stamp_
......@@ -93,7 +94,7 @@ class FileReader {
return this->time_stamp_;
}
inline virtual ~FileReader(){};
inline virtual ~FileReader() {}
private:
std::string filename_;
......@@ -128,7 +129,7 @@ class ParamDict {
virtual ~ParamDict();
private:
std::function<std::pair<Key, Value>(std::string)> read_func_;
std::function<std::pair<Key, Value>(std::string)> read_func_; // NOLINT
std::vector<FileReaderPtr> file_reader_lst_;
AbsKVDBPtr front_db, back_db;
};
......@@ -139,5 +140,5 @@ class ParamDictMgr {
void InsertParamDict(std::string, ParamDictPtr);
private:
std::unordered_map<std::string, ParamDictPtr> ParamDictMap;
std::unordered_map<std::string, ParamDictPtr> ParamDictMap; // NOLINT
};
core/kvdb/include/kvdb/paddle_rocksdb.h
浏览文件 @ 47ae3ad2
......@@ -25,7 +25,7 @@
class RocksDBWrapper {
public:
RocksDBWrapper(std::string db_name);
RocksDBWrapper(std::string db_name); // NOLINT
std::string Get(std::string key);
bool Put(std::string key, std::string value);
......@@ -33,6 +33,7 @@ class RocksDBWrapper {
static std::shared_ptr<RocksDBWrapper> RocksDBWrapperFactory(
std::string db_name = "SparseMatrix");
void Close();
private:
rocksdb::DB *db_;
std::string db_name_;
......
core/kvdb/src/mock_param_dict_impl.cpp
浏览文件 @ 47ae3ad2
......@@ -16,7 +16,7 @@
#include <fstream>
#include <iterator>
#include <sstream>
#include <thread>
#include <thread> // NOLINT
#include "core/kvdb/include/kvdb/rocksdb_impl.h"
std::vector<FileReaderPtr> ParamDict::GetDictReaderLst() {
......@@ -33,8 +33,10 @@ void ParamDict::SetFileReaderLst(std::vector<std::string> lst) {
std::vector<float> ParamDict::GetSparseValue(std::string feasign,
std::string slot) {
auto BytesToFloat = [](uint8_t* byte_array) { return *((float*)byte_array); };
// TODO: the concatation of feasign and slot is TBD.
auto BytesToFloat = [](uint8_t* byte_array) {
return *((float*)byte_array); // NOLINT
};
// NOLINT TODO: the concatation of feasign and slot is TBD.
std::string result = front_db->Get(feasign + slot);
std::vector<float> value;
if (result == "NOT_FOUND") return value;
......@@ -87,7 +89,7 @@ bool ParamDict::InsertSparseValue(std::string feasign,
value.push_back(raw_values_ptr[i]);
}
back_db->Set(key, value);
// TODO: change stateless to stateful
// NOLINT TODO: change stateless to stateful
return true;
}
......@@ -140,5 +142,4 @@ void ParamDict::CreateKVDB() {
this->back_db->CreateDB();
}
ParamDict::~ParamDict() {
}
ParamDict::~ParamDict() {}
core/kvdb/src/paddle_rocksdb.cpp
浏览文件 @ 47ae3ad2
......@@ -51,7 +51,7 @@ void RocksDBWrapper::SetDBName(std::string db_name) {
void RocksDBWrapper::Close() {
if (db_ != nullptr) {
db_->Close();
delete(db_);
delete (db_);
db_ = nullptr;
}
}
......
core/kvdb/src/rockskvdb_impl.cpp
浏览文件 @ 47ae3ad2
......@@ -32,12 +32,8 @@ void RocksKVDB::Set(std::string key, std::string value) {
return;
}
void RocksKVDB::Close() {
this->db_->Close();
}
void RocksKVDB::Close() { this->db_->Close(); }
std::string RocksKVDB::Get(std::string key) { return this->db_->Get(key); }
RocksKVDB::~RocksKVDB() {
this->db_->Close();
}
RocksKVDB::~RocksKVDB() { this->db_->Close(); }
core/pdcodegen/src/pdcodegen.cpp
浏览文件 @ 47ae3ad2
......@@ -15,14 +15,14 @@
#include <list>
#include "boost/algorithm/string.hpp"
#include "boost/scoped_ptr.hpp"
#include "core/pdcodegen/pds_option.pb.h"
#include "core/pdcodegen/plugin/strutil.h"
#include "core/pdcodegen/plugin/substitute.h"
#include "google/protobuf/compiler/code_generator.h"
#include "google/protobuf/compiler/plugin.h"
#include "google/protobuf/descriptor.h"
#include "google/protobuf/io/printer.h"
#include "google/protobuf/io/zero_copy_stream.h"
#include "core/pdcodegen/pds_option.pb.h"
#include "core/pdcodegen/plugin/strutil.h"
#include "core/pdcodegen/plugin/substitute.h"
using std::string;
using google::protobuf::Descriptor;
using google::protobuf::FileDescriptor;
......@@ -115,7 +115,8 @@ class PdsCodeGenerator : public CodeGenerator {
printer.Print("#include \"core/predictor/common/inner_common.h\"\n");
printer.Print("#include \"core/predictor/framework/service.h\"\n");
printer.Print("#include \"core/predictor/framework/manager.h\"\n");
printer.Print("#include \"core/predictor/framework/service_manager.h\"\n");
printer.Print(
"#include \"core/predictor/framework/service_manager.h\"\n");
}
if (generate_stub) {
printer.Print("#include <baidu/rpc/parallel_channel.h>\n");
......@@ -845,7 +846,8 @@ class PdsCodeGenerator : public CodeGenerator {
printer.Print("#include \"core/predictor/common/inner_common.h\"\n");
printer.Print("#include \"core/predictor/framework/service.h\"\n");
printer.Print("#include \"core/predictor/framework/manager.h\"\n");
printer.Print("#include \"core/predictor/framework/service_manager.h\"\n");
printer.Print(
"#include \"core/predictor/framework/service_manager.h\"\n");
}
if (generate_stub) {
printer.Print("#include <brpc/parallel_channel.h>\n");
......
core/predictor/common/inner_common.h
浏览文件 @ 47ae3ad2
......@@ -52,9 +52,9 @@
#include "glog/raw_logging.h"
#include "core/configure/general_model_config.pb.h"
#include "core/configure/include/configure_parser.h"
#include "core/configure/server_configure.pb.h"
#include "core/configure/general_model_config.pb.h"
#include "core/predictor/common/constant.h"
#include "core/predictor/common/types.h"
......
core/predictor/framework/dag_view.cpp
浏览文件 @ 47ae3ad2
......@@ -45,7 +45,7 @@ int DagView::init(Dag* dag, const std::string& service_name) {
<< "at:" << si;
return ERR_MEM_ALLOC_FAILURE;
}
VLOG(2) << "stage[" << si << "] name: " << stage->full_name;
VLOG(2) << "stage[" << si << "] name: " << stage->full_name;
VLOG(2) << "stage[" << si << "] node size: " << stage->nodes.size();
vstage->full_name = service_name + NAME_DELIMITER + stage->full_name;
uint32_t node_size = stage->nodes.size();
......@@ -74,7 +74,7 @@ int DagView::init(Dag* dag, const std::string& service_name) {
LOG(WARNING) << "Failed init op, type:" << node->type;
return ERR_INTERNAL_FAILURE;
}
op->set_full_name(service_name + NAME_DELIMITER + node->full_name);
vnode->conf = node;
vnode->op = op;
......@@ -85,9 +85,9 @@ int DagView::init(Dag* dag, const std::string& service_name) {
VLOG(2) << "set op pre name: \n"
<< "current op name: " << vstage->nodes.back()->op->op_name()
<< " previous op name: "
<< _view[si-1]->nodes.back()->op->op_name();
<< _view[si - 1]->nodes.back()->op->op_name();
vstage->nodes.back()->op->set_pre_node_name(
_view[si-1]->nodes.back()->op->op_name());
_view[si - 1]->nodes.back()->op->op_name());
}
_view.push_back(vstage);
}
......
core/predictor/framework/factory.h
浏览文件 @ 47ae3ad2
......@@ -16,8 +16,8 @@
#include <map>
#include <string>
#include <utility>
#include "glog/raw_logging.h"
#include "core/predictor/common/inner_common.h"
#include "glog/raw_logging.h"
namespace baidu {
namespace paddle_serving {
namespace predictor {
......
core/predictor/framework/resource.cpp
浏览文件 @ 47ae3ad2
......@@ -197,13 +197,10 @@ int Resource::general_model_initialize(const std::string& path,
for (int i = 0; i < feed_var_num; ++i) {
_config->_feed_name[i] = model_config.feed_var(i).name();
_config->_feed_alias_name[i] = model_config.feed_var(i).alias_name();
VLOG(2) << "feed var[" << i << "]: "
<< _config->_feed_name[i];
VLOG(2) << "feed var[" << i << "]: "
<< _config->_feed_alias_name[i];
VLOG(2) << "feed var[" << i << "]: " << _config->_feed_name[i];
VLOG(2) << "feed var[" << i << "]: " << _config->_feed_alias_name[i];
_config->_feed_type[i] = model_config.feed_var(i).feed_type();
VLOG(2) << "feed type[" << i << "]: "
<< _config->_feed_type[i];
VLOG(2) << "feed type[" << i << "]: " << _config->_feed_type[i];
if (model_config.feed_var(i).is_lod_tensor()) {
VLOG(2) << "var[" << i << "] is lod tensor";
......
core/predictor/framework/resource.h
浏览文件 @ 47ae3ad2
......@@ -13,10 +13,10 @@
// limitations under the License.
#pragma once
#include <map>
#include <memory>
#include <string>
#include <vector>
#include <map>
#include "core/cube/cube-api/include/cube_api.h"
#include "core/kvdb/include/kvdb/paddle_rocksdb.h"
#include "core/predictor/common/inner_common.h"
......@@ -36,15 +36,15 @@ class PaddleGeneralModelConfig {
public:
std::vector<std::string> _feed_name;
std::vector<std::string> _feed_alias_name;
std::vector<int> _feed_type; // 0 int64, 1 float
std::vector<bool> _is_lod_feed; // true lod tensor
std::vector<int> _feed_type; // 0 int64, 1 float
std::vector<bool> _is_lod_feed; // true lod tensor
std::vector<bool> _is_lod_fetch; // whether a fetch var is lod_tensor
std::vector<int> _capacity; // capacity for each tensor
/*
feed_shape_ for feeded variable
feed_shape_[i][j] represents the jth dim for ith input Tensor
if is_lod_feed_[i] == False, feed_shape_[i][0] = -1
*/
std::vector<int> _capacity; // capacity for each tensor
/*
feed_shape_ for feeded variable
feed_shape_[i][j] represents the jth dim for ith input Tensor
if is_lod_feed_[i] == False, feed_shape_[i][0] = -1
*/
std::vector<std::vector<int>> _feed_shape;
std::vector<std::string> _fetch_name;
......
core/predictor/src/pdserving.cpp
浏览文件 @ 47ae3ad2
......@@ -99,8 +99,8 @@ static void g_change_server_port() {
if (read_proto_conf(FLAGS_inferservice_path.c_str(),
FLAGS_inferservice_file.c_str(),
&conf) != 0) {
VLOG(2) << "failed to load configure[" << FLAGS_inferservice_path
<< "," << FLAGS_inferservice_file << "].";
VLOG(2) << "failed to load configure[" << FLAGS_inferservice_path << ","
<< FLAGS_inferservice_file << "].";
return;
}
uint32_t port = conf.port();
......@@ -157,8 +157,7 @@ int main(int argc, char** argv) {
mkdir(FLAGS_log_dir.c_str(), 0777);
ret = stat(FLAGS_log_dir.c_str(), &st_buf);
if (ret != 0) {
VLOG(2) << "Log path " << FLAGS_log_dir
<< " not exist, and create fail";
VLOG(2) << "Log path " << FLAGS_log_dir << " not exist, and create fail";
return -1;
}
}
......
core/predictor/unittest/test_message_op.h
浏览文件 @ 47ae3ad2
......@@ -15,8 +15,8 @@
#pragma once
#include <gtest/gtest.h>
#include "core/predictor/framework/channel.h"
#include "core/predictor/op/op.h"
#include "core/predictor/msg_data.pb.h"
#include "core/predictor/op/op.h"
namespace baidu {
namespace paddle_serving {
......
core/predictor/unittest/test_server_manager.cpp
浏览文件 @ 47ae3ad2
......@@ -13,7 +13,7 @@
// limitations under the License.
#include "core/predictor/unittest/test_server_manager.h" // TestServerManager
#include <gflags/gflags.h> // FLAGS
#include <gflags/gflags.h> // FLAGS
#include <string>
#include "core/predictor/framework/server.h" // ServerManager
......
core/sdk-cpp/include/common.h
浏览文件 @ 47ae3ad2
......@@ -53,9 +53,9 @@
#include "json2pb/json_to_pb.h"
#endif
#include "core/configure/general_model_config.pb.h"
#include "core/configure/include/configure_parser.h"
#include "core/configure/sdk_configure.pb.h"
#include "core/configure/general_model_config.pb.h"
#include "core/sdk-cpp/include/utils.h"
......
core/sdk-cpp/include/config_manager.h
浏览文件 @ 47ae3ad2
......@@ -32,9 +32,9 @@ class EndpointConfigManager {
EndpointConfigManager()
: _last_update_timestamp(0), _current_endpointmap_id(1) {}
int create(const std::string & sdk_desc_str);
int create(const std::string& sdk_desc_str);
int load(const std::string & sdk_desc_str);
int load(const std::string& sdk_desc_str);
int create(const char* path, const char* file);
......
core/sdk-cpp/include/factory.h
浏览文件 @ 47ae3ad2
......@@ -16,9 +16,9 @@
#include <map>
#include <string>
#include <utility>
#include "glog/raw_logging.h"
#include "core/sdk-cpp/include/common.h"
#include "core/sdk-cpp/include/stub_impl.h"
#include "glog/raw_logging.h"
namespace baidu {
namespace paddle_serving {
......
core/sdk-cpp/include/predictor_sdk.h
浏览文件 @ 47ae3ad2
......@@ -31,7 +31,7 @@ class PredictorApi {
int register_all();
int create(const std::string & sdk_desc_str);
int create(const std::string& sdk_desc_str);
int create(const char* path, const char* file);
......
core/sdk-cpp/proto/general_model_service.proto
浏览文件 @ 47ae3ad2
......@@ -28,13 +28,9 @@ message Tensor {
repeated int32 shape = 6;
};
message FeedInst {
repeated Tensor tensor_array = 1;
};
message FeedInst { repeated Tensor tensor_array = 1; };
message FetchInst {
repeated Tensor tensor_array = 1;
};
message FetchInst { repeated Tensor tensor_array = 1; };
message Request {
repeated FeedInst insts = 1;
......
core/sdk-cpp/src/endpoint.cpp
浏览文件 @ 47ae3ad2
......@@ -35,8 +35,7 @@ int Endpoint::initialize(const EndpointInfo& ep_info) {
return -1;
}
_variant_list.push_back(var);
VLOG(2) << "Succ create variant: " << vi
<< ", endpoint:" << _endpoint_name;
VLOG(2) << "Succ create variant: " << vi << ", endpoint:" << _endpoint_name;
}
return 0;
......
core/sdk-cpp/src/predictor_sdk.cpp
浏览文件 @ 47ae3ad2
......@@ -30,7 +30,7 @@ int PredictorApi::register_all() {
return 0;
}
int PredictorApi::create(const std::string & api_desc_str) {
int PredictorApi::create(const std::string& api_desc_str) {
VLOG(2) << api_desc_str;
if (register_all() != 0) {
LOG(ERROR) << "Failed do register all!";
......
core/sdk-cpp/src/variant.cpp
浏览文件 @ 47ae3ad2
......@@ -54,7 +54,7 @@ int Variant::initialize(const EndpointInfo& ep_info,
if (_stub_map.size() > 0) {
VLOG(2) << "Initialize variants from VariantInfo"
<< ", stubs count: " << _stub_map.size();
<< ", stubs count: " << _stub_map.size();
return 0;
}
......
core/util/CMakeLists.txt
浏览文件 @ 47ae3ad2
include(src/CMakeLists.txt)
add_library(utils ${util_srcs})
core/util/include/timer.h
浏览文件 @ 47ae3ad2
......@@ -15,7 +15,6 @@ limitations under the License. */
#pragma once
#include <stdlib.h>
namespace baidu {
namespace paddle_serving {
......
core/util/src/CMakeLists.txt
浏览文件 @ 47ae3ad2
FILE(GLOB srcs ${CMAKE_CURRENT_LIST_DIR}/*.cc)
LIST(APPEND util_srcs ${srcs})
doc/COMPILE.md
浏览文件 @ 47ae3ad2
# 如何编译PaddleServing
### 编译环境设置
- os: CentOS 6u3
- gcc: 4.8.2及以上
- go: 1.9.2及以上
- git：2.17.1及以上
- cmake：3.2.2及以上
- python：2.7.2及以上
### 获取代码
``` python
......@@ -9,6 +17,12 @@ git submodule update --init --recursive
### 编译Server部分
#### PYTHONROOT设置
``` shell
# 例如python的路径为/usr/bin/python，可以设置PYTHONROOT
export PYTHONROOT=/usr/
```
#### 集成CPU版本Paddle Inference Library
``` shell
cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ -DPYTHON_LIBRARIES=$PYTHONROOT/lib/libpython2.7.so -DPYTHON_EXECUTABLE=/home/users/dongdaxiang/software/baidu/third-party/python/bin/python -DCLIENT_ONLY=OFF ..
......@@ -28,3 +42,5 @@ cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ -DPYTHON_LIBRARIES=$PY
make -j10
```
### 安装wheel包
无论是client端还是server端，编译完成后，安装python/dist/下的whl包即可
doc/CONTRIBUTE.md
浏览文件 @ 47ae3ad2
......@@ -152,4 +152,3 @@ GLOG_minloglevel=1 bin/serving
2 -ERROR
3 - FATAL (Be careful as FATAL log will generate a coredump)
doc/IMDB_GO_CLIENT.md
浏览文件 @ 47ae3ad2
......@@ -193,6 +193,3 @@ total num: 25000
acc num: 22014
acc: 0.88056
```
doc/NEW_OPERATOR.md
浏览文件 @ 47ae3ad2
......@@ -143,6 +143,3 @@ self.op_dict = {
"general_dist_kv": "GeneralDistKVOp"
}
```
doc/SERVER_DAG.md
浏览文件 @ 47ae3ad2
......@@ -54,10 +54,3 @@ op_seq_maker.add_op(dist_kv_op)
op_seq_maker.add_op(general_infer_op)
op_seq_maker.add_op(general_response_op)
```
paddle_inference/inferencer-fluid-cpu/include/fluid_cpu_engine.h
浏览文件 @ 47ae3ad2
......@@ -21,8 +21,8 @@
#include <vector>
#include "core/configure/include/configure_parser.h"
#include "core/configure/inferencer_configure.pb.h"
#include "paddle_inference_api.h" // NOLINT
#include "core/predictor/framework/infer.h"
#include "paddle_inference_api.h" // NOLINT
namespace baidu {
namespace paddle_serving {
......@@ -336,7 +336,7 @@ class SigmoidModel {
return -1;
}
VLOG(2) << "load sigmoid_b [" << _sigmoid_b._params[0] << "] ["
<< _sigmoid_b._params[1] << "].";
<< _sigmoid_b._params[1] << "].";
_exp_max_input = exp_max;
_exp_min_input = exp_min;
return 0;
......@@ -373,7 +373,7 @@ class SigmoidFluidModel {
clone_model.reset(new SigmoidFluidModel());
clone_model->_sigmoid_core = _sigmoid_core;
clone_model->_fluid_core = _fluid_core->Clone();
return std::move(clone_model);
return std::move(clone_model); // NOLINT
}
public:
......@@ -459,7 +459,7 @@ class FluidCpuWithSigmoidCore : public FluidFamilyCore {
}
protected:
std::unique_ptr<SigmoidFluidModel> _core;
std::unique_ptr<SigmoidFluidModel> _core; // NOLINT
};
class FluidCpuNativeDirWithSigmoidCore : public FluidCpuWithSigmoidCore {
......
paddle_inference/inferencer-fluid-gpu/include/fluid_gpu_engine.h
浏览文件 @ 47ae3ad2
......@@ -23,8 +23,8 @@
#include <vector>
#include "core/configure/include/configure_parser.h"
#include "core/configure/inferencer_configure.pb.h"
#include "paddle_inference_api.h" // NOLINT
#include "core/predictor/framework/infer.h"
#include "paddle_inference_api.h" // NOLINT
DECLARE_int32(gpuid);
......@@ -334,13 +334,13 @@ class SigmoidModel {
return -1;
}
VLOG(2) << "load sigmoid_w [" << _sigmoid_w._params[0] << "] ["
<< _sigmoid_w._params[1] << "].";
<< _sigmoid_w._params[1] << "].";
if (0 != _sigmoid_b.init(2, 1, sigmoid_b_file) || 0 != _sigmoid_b.load()) {
LOG(ERROR) << "load params sigmoid_b failed.";
return -1;
}
VLOG(2) << "load sigmoid_b [" << _sigmoid_b._params[0] << "] ["
<< _sigmoid_b._params[1] << "].";
<< _sigmoid_b._params[1] << "].";
_exp_max_input = exp_max;
_exp_min_input = exp_min;
return 0;
......
python/examples/bert/batching.py 0 → 100644
浏览文件 @ 47ae3ad2
#coding:utf-8
# Copyright (c) 2019 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.
"""Mask, padding and batching."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
def prepare_batch_data(insts,
total_token_num,
max_seq_len=128,
pad_id=None,
cls_id=None,
sep_id=None,
mask_id=None,
return_input_mask=True,
return_max_len=True,
return_num_token=False):
"""
1. generate Tensor of data
2. generate Tensor of position
3. generate self attention mask, [shape: batch_size * max_len * max_len]
"""
batch_src_ids = [inst[0] for inst in insts]
batch_sent_ids = [inst[1] for inst in insts]
batch_pos_ids = [inst[2] for inst in insts]
labels_list = []
# compatible with squad, whose example includes start/end positions,
# or unique id
for i in range(3, len(insts[0]), 1):
labels = [inst[i] for inst in insts]
labels = np.array(labels).astype("int64").reshape([-1, 1])
labels_list.append(labels)
out = batch_src_ids
# Second step: padding
src_id, self_input_mask = pad_batch_data(
out, pad_idx=pad_id, max_seq_len=max_seq_len, return_input_mask=True)
pos_id = pad_batch_data(
batch_pos_ids,
pad_idx=pad_id,
max_seq_len=max_seq_len,
return_pos=False,
return_input_mask=False)
sent_id = pad_batch_data(
batch_sent_ids,
pad_idx=pad_id,
max_seq_len=max_seq_len,
return_pos=False,
return_input_mask=False)
return_list = [src_id, pos_id, sent_id, self_input_mask] + labels_list
return return_list if len(return_list) > 1 else return_list[0]
def pad_batch_data(insts,
pad_idx=0,
max_seq_len=128,
return_pos=False,
return_input_mask=False,
return_max_len=False,
return_num_token=False,
return_seq_lens=False):
"""
Pad the instances to the max sequence length in batch, and generate the
corresponding position data and input mask.
"""
return_list = []
#max_len = max(len(inst) for inst in insts)
max_len = max_seq_len
# Any token included in dict can be used to pad, since the paddings' loss
# will be masked out by weights and make no effect on parameter gradients.
inst_data = np.array([
list(inst) + list([pad_idx] * (max_len - len(inst))) for inst in insts
])
return_list += [inst_data.astype("int64").reshape([-1, max_len, 1])]
# position data
if return_pos:
inst_pos = np.array([
list(range(0, len(inst))) + [pad_idx] * (max_len - len(inst))
for inst in insts
])
return_list += [inst_pos.astype("int64").reshape([-1, max_len, 1])]
if return_input_mask:
# This is used to avoid attention on paddings.
input_mask_data = np.array(
[[1] * len(inst) + [0] * (max_len - len(inst)) for inst in insts])
input_mask_data = np.expand_dims(input_mask_data, axis=-1)
return_list += [input_mask_data.astype("float32")]
if return_max_len:
return_list += [max_len]
if return_num_token:
num_token = 0
for inst in insts:
num_token += len(inst)
return_list += [num_token]
if return_seq_lens:
seq_lens = np.array([len(inst) for inst in insts])
return_list += [seq_lens.astype("int64").reshape([-1, 1])]
return return_list if len(return_list) > 1 else return_list[0]
python/examples/bert/benchmark.py
浏览文件 @ 47ae3ad2
# -*- coding: utf-8 -*-
#
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
......@@ -11,55 +13,59 @@
# 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 unicode_literals, absolute_import
import os
import sys
import time
from paddle_serving_client import Client
from paddle_serving_client.metric import auc
from paddle_serving_client.utils import MultiThreadRunner
import time
from bert_client import BertService
from paddle_serving_client.utils import benchmark_args
from batching import pad_batch_data
import tokenization
import requests
import json
from bert_reader import BertReader
args = benchmark_args()
def predict(thr_id, resource):
bc = BertService(
model_name="bert_chinese_L-12_H-768_A-12",
max_seq_len=20,
do_lower_case=True)
bc.load_client(resource["conf_file"], resource["server_endpoint"])
thread_num = resource["thread_num"]
file_list = resource["filelist"]
line_id = 0
result = []
label_list = []
dataset = []
for fn in file_list:
fin = open(fn)
for line in fin:
if line_id % thread_num == thr_id - 1:
dataset.append(line.strip())
line_id += 1
fin.close()
start = time.time()
fetch = ["pooled_output"]
for inst in dataset:
fetch_map = bc.run_general([[inst]], fetch)
result.append(fetch_map["pooled_output"])
end = time.time()
return [result, label_list, [end - start]]
def single_func(idx, resource):
fin = open("data-c.txt")
if args.request == "rpc":
reader = BertReader(vocab_file="vocab.txt", max_seq_len=20)
config_file = './serving_client_conf/serving_client_conf.prototxt'
fetch = ["pooled_output"]
client = Client()
client.load_client_config(args.model)
client.connect([resource["endpoint"][idx % 4]])
if __name__ == '__main__':
conf_file = sys.argv[1]
data_file = sys.argv[2]
thread_num = sys.argv[3]
resource = {}
resource["conf_file"] = conf_file
resource["server_endpoint"] = ["127.0.0.1:9292"]
resource["filelist"] = [data_file]
resource["thread_num"] = int(thread_num)
start = time.time()
for line in fin:
feed_dict = reader.process(line)
result = client.predict(feed=feed_dict, fetch=fetch)
end = time.time()
elif args.request == "http":
start = time.time()
header = {"Content-Type": "application/json"}
for line in fin:
#dict_data = {"words": "this is for output ", "fetch": ["pooled_output"]}
dict_data = {"words": line, "fetch": ["pooled_output"]}
r = requests.post(
'http://{}/bert/prediction'.format(resource["endpoint"][0]),
data=json.dumps(dict_data),
headers=header)
end = time.time()
return [[end - start]]
thread_runner = MultiThreadRunner()
result = thread_runner.run(predict, int(sys.argv[3]), resource)
print("total time {} s".format(sum(result[-1]) / len(result[-1])))
if __name__ == '__main__':
multi_thread_runner = MultiThreadRunner()
endpoint_list = [
"127.0.0.1:9494", "127.0.0.1:9495", "127.0.0.1:9496", "127.0.0.1:9497"
]
#endpoint_list = endpoint_list + endpoint_list + endpoint_list
#result = multi_thread_runner.run(single_func, args.thread, {"endpoint":endpoint_list})
result = single_func(0, {"endpoint": endpoint_list})
print(result)
python/examples/bert/bert_client.py
浏览文件 @ 47ae3ad2
# coding:utf-8
# pylint: disable=doc-string-missing
import os
import sys
import numpy as np
......@@ -9,6 +10,9 @@ import time
from paddlehub.common.logger import logger
import socket
from paddle_serving_client import Client
from paddle_serving_client.utils import MultiThreadRunner
from paddle_serving_client.utils import benchmark_args
args = benchmark_args()
_ver = sys.version_info
is_py2 = (_ver[0] == 2)
......@@ -122,36 +126,30 @@ class BertService():
return fetch_map_batch
def test():
def single_func(idx, resource):
bc = BertService(
model_name='bert_chinese_L-12_H-768_A-12',
max_seq_len=20,
show_ids=False,
do_lower_case=True)
server_addr = ["127.0.0.1:9292"]
config_file = './serving_client_conf/serving_client_conf.prototxt'
fetch = ["pooled_output"]
server_addr = [resource["endpoint"][idx]]
bc.load_client(config_file, server_addr)
batch_size = 1
batch = []
for line in sys.stdin:
if batch_size == 1:
result = bc.run_general([[line.strip()]], fetch)
print(result)
else:
if len(batch) < batch_size:
batch.append([line.strip()])
else:
result = bc.run_batch_general(batch, fetch)
batch = []
for r in result:
print(r)
if len(batch) > 0:
result = bc.run_batch_general(batch, fetch)
batch = []
for r in result:
print(r)
start = time.time()
fin = open("data-c.txt")
for line in fin:
result = bc.run_general([[line.strip()]], fetch)
end = time.time()
return [[end - start]]
if __name__ == '__main__':
test()
multi_thread_runner = MultiThreadRunner()
result = multi_thread_runner.run(single_func, args.thread, {
"endpoint": [
"127.0.0.1:9494", "127.0.0.1:9495", "127.0.0.1:9496",
"127.0.0.1:9497"
]
})
python/examples/bert/bert_reader.py 0 → 100644
浏览文件 @ 47ae3ad2
# 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 batching import pad_batch_data
import tokenization
class BertReader():
def __init__(self, vocab_file="", max_seq_len=128):
self.vocab_file = vocab_file
self.tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file)
self.max_seq_len = max_seq_len
self.vocab = self.tokenizer.vocab
self.pad_id = self.vocab["[PAD]"]
self.cls_id = self.vocab["[CLS]"]
self.sep_id = self.vocab["[SEP]"]
self.mask_id = self.vocab["[MASK]"]
def pad_batch(self, token_ids, text_type_ids, position_ids):
batch_token_ids = [token_ids]
batch_text_type_ids = [text_type_ids]
batch_position_ids = [position_ids]
padded_token_ids, input_mask = pad_batch_data(
batch_token_ids,
max_seq_len=self.max_seq_len,
pad_idx=self.pad_id,
return_input_mask=True)
padded_text_type_ids = pad_batch_data(
batch_text_type_ids,
max_seq_len=self.max_seq_len,
pad_idx=self.pad_id)
padded_position_ids = pad_batch_data(
batch_position_ids,
max_seq_len=self.max_seq_len,
pad_idx=self.pad_id)
return padded_token_ids, padded_position_ids, padded_text_type_ids, input_mask
def process(self, sent):
text_a = tokenization.convert_to_unicode(sent)
tokens_a = self.tokenizer.tokenize(text_a)
if len(tokens_a) > self.max_seq_len - 2:
tokens_a = tokens_a[0:(self.max_seq_len - 2)]
tokens = []
text_type_ids = []
tokens.append("[CLS]")
text_type_ids.append(0)
for token in tokens_a:
tokens.append(token)
text_type_ids.append(0)
token_ids = self.tokenizer.convert_tokens_to_ids(tokens)
position_ids = list(range(len(token_ids)))
p_token_ids, p_pos_ids, p_text_type_ids, input_mask = \
self.pad_batch(token_ids, text_type_ids, position_ids)
feed_result = {
"input_ids": p_token_ids.reshape(-1).tolist(),
"position_ids": p_pos_ids.reshape(-1).tolist(),
"segment_ids": p_text_type_ids.reshape(-1).tolist(),
"input_mask": input_mask.reshape(-1).tolist()
}
return feed_result
python/examples/bert/bert_web_service.py 0 → 100644
浏览文件 @ 47ae3ad2
# 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_server_gpu.web_service import WebService
from bert_reader import BertReader
import sys
import os
class BertService(WebService):
def load(self):
self.reader = BertReader(vocab_file="vocab.txt", max_seq_len=20)
def preprocess(self, feed={}, fetch=[]):
feed_res = self.reader.process(feed["words"].encode("utf-8"))
return feed_res, fetch
bert_service = BertService(name="bert")
bert_service.load()
bert_service.load_model_config(sys.argv[1])
gpu_ids = os.environ["CUDA_VISIBLE_DEVICES"]
gpus = [int(x) for x in gpu_ids.split(",")]
bert_service.set_gpus(gpus)
bert_service.prepare_server(workdir="workdir", port=9494, device="gpu")
bert_service.run_server()
python/examples/bert/prepare_model.py
浏览文件 @ 47ae3ad2
......@@ -11,14 +11,16 @@
# 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 paddlehub as hub
import paddle.fluid as fluid
import sys
import paddle_serving_client.io as serving_io
model_name = "bert_chinese_L-12_H-768_A-12"
module = hub.Module(model_name)
inputs, outputs, program = module.context(trainable=True, max_seq_len=20)
inputs, outputs, program = module.context(
trainable=True, max_seq_len=int(sys.argv[1]))
place = fluid.core_avx.CPUPlace()
exe = fluid.Executor(place)
input_ids = inputs["input_ids"]
......@@ -34,10 +36,12 @@ feed_var_names = [
target_vars = [pooled_output, sequence_output]
serving_io.save_model("serving_server_model", "serving_client_conf", {
"input_ids": input_ids,
"position_ids": position_ids,
"segment_ids": segment_ids,
"input_mask": input_mask,
}, {"pooled_output": pooled_output,
"sequence_output": sequence_output}, program)
serving_io.save_model(
"bert_seq{}_model".format(sys.argv[1]),
"bert_seq{}_client".format(sys.argv[1]), {
"input_ids": input_ids,
"position_ids": position_ids,
"segment_ids": segment_ids,
"input_mask": input_mask,
}, {"pooled_output": pooled_output,
"sequence_output": sequence_output}, program)
python/examples/bert/tokenization.py 0 → 100644
浏览文件 @ 47ae3ad2
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""Tokenization classes."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import io
import unicodedata
import six
import sentencepiece as spm
import pickle
def convert_to_unicode(text): # pylint: disable=doc-string-with-all-args
"""Converts `text` to Unicode (if it's not already), assuming utf-8 input."""
if six.PY3:
if isinstance(text, str):
return text
elif isinstance(text, bytes):
return text.decode("utf-8", "ignore")
else:
raise ValueError("Unsupported string type: %s" % (type(text)))
elif six.PY2:
if isinstance(text, str):
return text.decode("utf-8", "ignore")
elif isinstance(text, unicode): # noqa
return text
else:
raise ValueError("Unsupported string type: %s" % (type(text)))
else:
raise ValueError("Not running on Python2 or Python 3?")
def printable_text(text): # pylint: disable=doc-string-with-all-args
"""Returns text encoded in a way suitable for print or `tf.logging`."""
# These functions want `str` for both Python2 and Python3, but in one case
# it's a Unicode string and in the other it's a byte string.
if six.PY3:
if isinstance(text, str):
return text
elif isinstance(text, bytes):
return text.decode("utf-8", "ignore")
else:
raise ValueError("Unsupported string type: %s" % (type(text)))
elif six.PY2:
if isinstance(text, str):
return text
elif isinstance(text, unicode): # noqa
return text.encode("utf-8")
else:
raise ValueError("Unsupported string type: %s" % (type(text)))
else:
raise ValueError("Not running on Python2 or Python 3?")
def load_vocab(vocab_file): # pylint: disable=doc-string-with-all-args, doc-string-with-returns
"""Loads a vocabulary file into a dictionary."""
vocab = collections.OrderedDict()
fin = io.open(vocab_file, "r", encoding="UTF-8")
for num, line in enumerate(fin):
items = convert_to_unicode(line.strip()).split("\t")
if len(items) > 2:
break
token = items[0]
index = items[1] if len(items) == 2 else num
token = token.strip()
vocab[token] = int(index)
fin.close()
return vocab
def convert_by_vocab(vocab, items):
"""Converts a sequence of [tokens|ids] using the vocab."""
output = []
for item in items:
output.append(vocab[item])
return output
def convert_tokens_to_ids(vocab, tokens):
return convert_by_vocab(vocab, tokens)
def convert_ids_to_tokens(inv_vocab, ids):
return convert_by_vocab(inv_vocab, ids)
def whitespace_tokenize(text):
"""Runs basic whitespace cleaning and splitting on a peice of text."""
text = text.strip()
if not text:
return []
tokens = text.split()
return tokens
class FullTokenizer(object):
"""Runs end-to-end tokenziation."""
def __init__(self,
vocab_file,
do_lower_case=True,
use_sentence_piece_vocab=False):
self.vocab = load_vocab(vocab_file)
self.inv_vocab = {v: k for k, v in self.vocab.items()}
self.basic_tokenizer = BasicTokenizer(do_lower_case=do_lower_case)
self.use_sentence_piece_vocab = use_sentence_piece_vocab
self.wordpiece_tokenizer = WordpieceTokenizer(
vocab=self.vocab,
use_sentence_piece_vocab=self.use_sentence_piece_vocab)
def tokenize(self, text):
split_tokens = []
for token in self.basic_tokenizer.tokenize(text):
for sub_token in self.wordpiece_tokenizer.tokenize(token):
split_tokens.append(sub_token)
return split_tokens
def convert_tokens_to_ids(self, tokens):
return convert_by_vocab(self.vocab, tokens)
def convert_ids_to_tokens(self, ids):
return convert_by_vocab(self.inv_vocab, ids)
class CharTokenizer(object):
"""Runs end-to-end tokenziation."""
def __init__(self, vocab_file, do_lower_case=True):
self.vocab = load_vocab(vocab_file)
self.inv_vocab = {v: k for k, v in self.vocab.items()}
self.wordpiece_tokenizer = WordpieceTokenizer(vocab=self.vocab)
def tokenize(self, text):
split_tokens = []
for token in text.lower().split(" "):
for sub_token in self.wordpiece_tokenizer.tokenize(token):
split_tokens.append(sub_token)
return split_tokens
def convert_tokens_to_ids(self, tokens):
return convert_by_vocab(self.vocab, tokens)
def convert_ids_to_tokens(self, ids):
return convert_by_vocab(self.inv_vocab, ids)
class WSSPTokenizer(object): # pylint: disable=doc-string-missing
def __init__(self, vocab_file, sp_model_dir, word_dict, ws=True,
lower=True):
self.vocab = load_vocab(vocab_file)
self.inv_vocab = {v: k for k, v in self.vocab.items()}
self.ws = ws
self.lower = lower
self.dict = pickle.load(open(word_dict, 'rb'))
self.sp_model = spm.SentencePieceProcessor()
self.window_size = 5
self.sp_model.Load(sp_model_dir)
def cut(self, chars): # pylint: disable=doc-string-missing
words = []
idx = 0
while idx < len(chars):
matched = False
for i in range(self.window_size, 0, -1):
cand = chars[idx:idx + i]
if cand in self.dict:
words.append(cand)
matched = True
break
if not matched:
i = 1
words.append(chars[idx])
idx += i
return words
def tokenize(self, text, unk_token="[UNK]"): # pylint: disable=doc-string-missing
text = convert_to_unicode(text)
if self.ws:
text = [s for s in self.cut(text) if s != ' ']
else:
text = text.split(' ')
if self.lower:
text = [s.lower() for s in text]
text = ' '.join(text)
tokens = self.sp_model.EncodeAsPieces(text)
in_vocab_tokens = []
for token in tokens:
if token in self.vocab:
in_vocab_tokens.append(token)
else:
in_vocab_tokens.append(unk_token)
return in_vocab_tokens
def convert_tokens_to_ids(self, tokens):
return convert_by_vocab(self.vocab, tokens)
def convert_ids_to_tokens(self, ids):
return convert_by_vocab(self.inv_vocab, ids)
class BasicTokenizer(object):
"""Runs basic tokenization (punctuation splitting, lower casing, etc.)."""
def __init__(self, do_lower_case=True):
"""Constructs a BasicTokenizer.
Args:
do_lower_case: Whether to lower case the input.
"""
self.do_lower_case = do_lower_case
def tokenize(self, text): # pylint: disable=doc-string-with-all-args, doc-string-with-returns
"""Tokenizes a piece of text."""
text = convert_to_unicode(text)
text = self._clean_text(text)
# This was added on November 1st, 2018 for the multilingual and Chinese
# models. This is also applied to the English models now, but it doesn't
# matter since the English models were not trained on any Chinese data
# and generally don't have any Chinese data in them (there are Chinese
# characters in the vocabulary because Wikipedia does have some Chinese
# words in the English Wikipedia.).
text = self._tokenize_chinese_chars(text)
orig_tokens = whitespace_tokenize(text)
split_tokens = []
for token in orig_tokens:
if self.do_lower_case:
token = token.lower()
token = self._run_strip_accents(token)
split_tokens.extend(self._run_split_on_punc(token))
output_tokens = whitespace_tokenize(" ".join(split_tokens))
return output_tokens
def _run_strip_accents(self, text):
"""Strips accents from a piece of text."""
text = unicodedata.normalize("NFD", text)
output = []
for char in text:
cat = unicodedata.category(char)
if cat == "Mn":
continue
output.append(char)
return "".join(output)
def _run_split_on_punc(self, text):
"""Splits punctuation on a piece of text."""
chars = list(text)
i = 0
start_new_word = True
output = []
while i < len(chars):
char = chars[i]
if _is_punctuation(char):
output.append([char])
start_new_word = True
else:
if start_new_word:
output.append([])
start_new_word = False
output[-1].append(char)
i += 1
return ["".join(x) for x in output]
def _tokenize_chinese_chars(self, text):
"""Adds whitespace around any CJK character."""
output = []
for char in text:
cp = ord(char)
if self._is_chinese_char(cp):
output.append(" ")
output.append(char)
output.append(" ")
else:
output.append(char)
return "".join(output)
def _is_chinese_char(self, cp):
"""Checks whether CP is the codepoint of a CJK character."""
# This defines a "chinese character" as anything in the CJK Unicode block:
# https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_(Unicode_block)
#
# Note that the CJK Unicode block is NOT all Japanese and Korean characters,
# despite its name. The modern Korean Hangul alphabet is a different block,
# as is Japanese Hiragana and Katakana. Those alphabets are used to write
# space-separated words, so they are not treated specially and handled
# like the all of the other languages.
if ((cp >= 0x4E00 and cp <= 0x9FFF) or #
(cp >= 0x3400 and cp <= 0x4DBF) or #
(cp >= 0x20000 and cp <= 0x2A6DF) or #
(cp >= 0x2A700 and cp <= 0x2B73F) or #
(cp >= 0x2B740 and cp <= 0x2B81F) or #
(cp >= 0x2B820 and cp <= 0x2CEAF) or
(cp >= 0xF900 and cp <= 0xFAFF) or #
(cp >= 0x2F800 and cp <= 0x2FA1F)): #
return True
return False
def _clean_text(self, text):
"""Performs invalid character removal and whitespace cleanup on text."""
output = []
for char in text:
cp = ord(char)
if cp == 0 or cp == 0xfffd or _is_control(char):
continue
if _is_whitespace(char):
output.append(" ")
else:
output.append(char)
return "".join(output)
class WordpieceTokenizer(object):
"""Runs WordPiece tokenziation."""
def __init__(self,
vocab,
unk_token="[UNK]",
max_input_chars_per_word=100,
use_sentence_piece_vocab=False):
self.vocab = vocab
self.unk_token = unk_token
self.max_input_chars_per_word = max_input_chars_per_word
self.use_sentence_piece_vocab = use_sentence_piece_vocab
def tokenize(self, text): # pylint: disable=doc-string-with-all-args
"""Tokenizes a piece of text into its word pieces.
This uses a greedy longest-match-first algorithm to perform tokenization
using the given vocabulary.
For example:
input = "unaffable"
output = ["un", "##aff", "##able"]
Args:
text: A single token or whitespace separated tokens. This should have
already been passed through `BasicTokenizer.
Returns:
A list of wordpiece tokens.
"""
text = convert_to_unicode(text)
output_tokens = []
for token in whitespace_tokenize(text):
chars = list(token)
if len(chars) > self.max_input_chars_per_word:
output_tokens.append(self.unk_token)
continue
is_bad = False
start = 0
sub_tokens = []
while start < len(chars):
end = len(chars)
cur_substr = None
while start < end:
substr = "".join(chars[start:end])
if start == 0 and self.use_sentence_piece_vocab:
substr = u'\u2581' + substr
if start > 0 and not self.use_sentence_piece_vocab:
substr = "##" + substr
if substr in self.vocab:
cur_substr = substr
break
end -= 1
if cur_substr is None:
is_bad = True
break
sub_tokens.append(cur_substr)
start = end
if is_bad:
output_tokens.append(self.unk_token)
else:
output_tokens.extend(sub_tokens)
return output_tokens
def _is_whitespace(char):
"""Checks whether `chars` is a whitespace character."""
# \t, \n, and \r are technically contorl characters but we treat them
# as whitespace since they are generally considered as such.
if char == " " or char == "\t" or char == "\n" or char == "\r":
return True
cat = unicodedata.category(char)
if cat == "Zs":
return True
return False
def _is_control(char):
"""Checks whether `chars` is a control character."""
# These are technically control characters but we count them as whitespace
# characters.
if char == "\t" or char == "\n" or char == "\r":
return False
cat = unicodedata.category(char)
if cat.startswith("C"):
return True
return False
def _is_punctuation(char):
"""Checks whether `chars` is a punctuation character."""
cp = ord(char)
# We treat all non-letter/number ASCII as punctuation.
# Characters such as "^", "$", and "`" are not in the Unicode
# Punctuation class but we treat them as punctuation anyways, for
# consistency.
if ((cp >= 33 and cp <= 47) or (cp >= 58 and cp <= 64) or
(cp >= 91 and cp <= 96) or (cp >= 123 and cp <= 126)):
return True
cat = unicodedata.category(char)
if cat.startswith("P"):
return True
return False
python/examples/criteo_ctr/args.py
浏览文件 @ 47ae3ad2
# 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()
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/criteo_reader.py
浏览文件 @ 47ae3ad2
# 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.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
......@@ -23,8 +43,9 @@ class CriteoDataset(dg.MultiSlotDataGenerator):
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_])
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):
......@@ -32,16 +53,17 @@ class CriteoDataset(dg.MultiSlotDataGenerator):
for fname in filelist:
with open(fname.strip(), "r") as fin:
for line in fin:
dense_feature, sparse_feature, label = self._process_line(line)
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():
......@@ -54,6 +76,7 @@ class CriteoDataset(dg.MultiSlotDataGenerator):
return data_iter
if __name__ == "__main__":
criteo_dataset = CriteoDataset()
criteo_dataset.setup(int(sys.argv[1]))
......
python/examples/criteo_ctr/local_train.py
浏览文件 @ 47ae3ad2
# 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
......@@ -17,15 +32,17 @@ def train():
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)]
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 = dnn_model(
nn_input, sparse_input_ids, label,
args.embedding_size, args.sparse_feature_dim)
nn_input, sparse_input_ids, label, args.embedding_size,
args.sparse_feature_dim)
optimizer = fluid.optimizer.SGD(learning_rate=1e-4)
optimizer.minimize(loss)
......@@ -36,16 +53,17 @@ def train():
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)
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")))]
whole_filelist = [
"raw_data/part-%d" % x for x in range(len(os.listdir("raw_data")))
]
dataset.set_filelist(whole_filelist[:thread_num])
dataset.load_into_memory()
......@@ -53,8 +71,7 @@ def train():
epochs = 1
for i in range(epochs):
exe.train_from_dataset(
program=fluid.default_main_program(),
dataset=dataset, debug=True)
program=fluid.default_main_program(), dataset=dataset, debug=True)
print("epoch {} finished".format(i))
import paddle_serving_client.io as server_io
......@@ -63,9 +80,9 @@ def train():
feed_var_dict["sparse_{}".format(i)] = sparse
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", "ctr_client_conf", feed_var_dict,
fetch_var_dict, fluid.default_main_program())
if __name__ == '__main__':
train()
python/examples/criteo_ctr/network_conf.py
浏览文件 @ 47ae3ad2
# 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 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()))
param_attr=fluid.ParamAttr(
name="SparseFeatFactors",
initializer=fluid.initializer.Uniform()))
return fluid.layers.sequence_pool(input=emb, pool_type='sum')
def mlp_input_tensor(emb_sums, dense_tensor):
......@@ -21,18 +37,30 @@ def dnn_model(dense_input, sparse_inputs, label,
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]))))
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_sums = list(map(embedding_layer, sparse_inputs))
......
python/examples/criteo_ctr/test_client.py
浏览文件 @ 47ae3ad2
# 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 paddle
import sys
......@@ -13,9 +28,12 @@ batch = 1
buf_size = 100
dataset = criteo.CriteoDataset()
dataset.setup(1000001)
test_filelists = ["{}/part-%d".format(sys.argv[2]) % x
for x in range(len(os.listdir(sys.argv[2])))]
reader = dataset.infer_reader(test_filelists[len(test_filelists)-40:], batch, buf_size)
test_filelists = [
"{}/part-%d".format(sys.argv[2]) % x
for x in range(len(os.listdir(sys.argv[2])))
]
reader = dataset.infer_reader(test_filelists[len(test_filelists) - 40:], batch,
buf_size)
label_list = []
prob_list = []
......@@ -25,4 +43,3 @@ for data in reader():
feed_dict["sparse_{}".format(i - 1)] = data[0][i]
fetch_map = client.predict(feed=feed_dict, fetch=["prob"])
print(fetch_map)
python/examples/criteo_ctr/test_server.py
浏览文件 @ 47ae3ad2
# 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
......
python/examples/fit_a_line/README.md
浏览文件 @ 47ae3ad2
......@@ -19,4 +19,4 @@ python -m paddle_serving_server.web_serve --model uci_housing_model/ --thread 10
Prediction through http post
``` shell
curl -H "Content-Type:application/json" -X POST -d '{"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://127.0.0.1:9393/uci/prediction
```
\ No newline at end of file
```
python/examples/fit_a_line/benchmark.py
浏览文件 @ 47ae3ad2
......@@ -11,6 +11,8 @@
# 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
from paddle_serving_client.utils import MultiThreadRunner
from paddle_serving_client.utils import benchmark_args
......@@ -21,28 +23,35 @@ import requests
args = benchmark_args()
def single_func(idx, resource):
if args.request == "rpc":
client = Client()
client.load_client_config(args.model)
client.connect([args.endpoint])
train_reader = paddle.batch(paddle.reader.shuffle(
paddle.dataset.uci_housing.train(), buf_size=500), batch_size=1)
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.uci_housing.train(), buf_size=500),
batch_size=1)
start = time.time()
for data in train_reader():
fetch_map = client.predict(feed={"x": data[0][0]}, fetch=["price"])
end = time.time()
return [[end - start]]
elif args.request == "http":
train_reader = paddle.batch(paddle.reader.shuffle(
paddle.dataset.uci_housing.train(), buf_size=500), batch_size=1)
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.uci_housing.train(), buf_size=500),
batch_size=1)
start = time.time()
for data in train_reader():
r = requests.post('http://{}/uci/prediction'.format(args.endpoint),
data = {"x": data[0]})
r = requests.post(
'http://{}/uci/prediction'.format(args.endpoint),
data={"x": data[0]})
end = time.time()
return [[end - start]]
multi_thread_runner = MultiThreadRunner()
result = multi_thread_runner.run(single_func, args.thread, {})
print(result)
python/examples/fit_a_line/local_train.py
浏览文件 @ 47ae3ad2
# 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
import paddle.fluid as fluid
train_reader = paddle.batch(paddle.reader.shuffle(
paddle.dataset.uci_housing.train(), buf_size=500), batch_size=16)
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.uci_housing.train(), buf_size=500),
batch_size=16)
test_reader = paddle.batch(paddle.reader.shuffle(
paddle.dataset.uci_housing.test(), buf_size=500), batch_size=16)
test_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.uci_housing.test(), buf_size=500),
batch_size=16)
x = fluid.data(name='x', shape=[None, 13], dtype='float32')
y = fluid.data(name='y', shape=[None, 1], dtype='float32')
......@@ -26,11 +45,9 @@ import paddle_serving_client.io as serving_io
for pass_id in range(30):
for data_train in train_reader():
avg_loss_value, = exe.run(
fluid.default_main_program(),
feed=feeder.feed(data_train),
fetch_list=[avg_loss])
serving_io.save_model(
"uci_housing_model", "uci_housing_client",
{"x": x}, {"price": y_predict}, fluid.default_main_program())
avg_loss_value, = exe.run(fluid.default_main_program(),
feed=feeder.feed(data_train),
fetch_list=[avg_loss])
serving_io.save_model("uci_housing_model", "uci_housing_client", {"x": x},
{"price": y_predict}, fluid.default_main_program())
python/examples/fit_a_line/test_client.py
浏览文件 @ 47ae3ad2
# 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 paddle
import sys
client = Client()
client.load_client_config(sys.argv[1])
client.connect(["127.0.0.1:9393"])
test_reader = paddle.batch(paddle.reader.shuffle(
paddle.dataset.uci_housing.test(), buf_size=500), batch_size=1)
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/test_server.py
浏览文件 @ 47ae3ad2
# 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
......
python/examples/imdb/benchmark.py
浏览文件 @ 47ae3ad2
......@@ -11,6 +11,7 @@
# 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 time
......@@ -22,6 +23,7 @@ from paddle_serving_client.utils import benchmark_args
args = benchmark_args()
def single_func(idx, resource):
imdb_dataset = IMDBDataset()
imdb_dataset.load_resource(args.vocab)
......@@ -40,18 +42,21 @@ def single_func(idx, resource):
fin = open(fn)
for line in fin:
word_ids, label = imdb_dataset.get_words_and_label(line)
fetch_map = client.predict(feed={"words": word_ids},
fetch=["prediction"])
fetch_map = client.predict(
feed={"words": word_ids}, fetch=["prediction"])
elif args.request == "http":
for fn in filelist:
fin = open(fn)
for line in fin:
word_ids, label = imdb_dataset.get_words_and_label(line)
r = requests.post("http://{}/imdb/prediction".format(args.endpoint),
data={"words": word_ids})
r = requests.post(
"http://{}/imdb/prediction".format(args.endpoint),
data={"words": word_ids,
"fetch": ["prediction"]})
end = time.time()
return [[end - start]]
multi_thread_runner = MultiThreadRunner()
result = multi_thread_runner.run(single_func, args.thread, {})
print(result)
python/examples/imdb/imdb_reader.py
浏览文件 @ 47ae3ad2
......@@ -11,6 +11,7 @@
# 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 os
......@@ -18,6 +19,7 @@ import paddle
import re
import paddle.fluid.incubate.data_generator as dg
class IMDBDataset(dg.MultiSlotDataGenerator):
def load_resource(self, dictfile):
self._vocab = {}
......@@ -42,7 +44,7 @@ class IMDBDataset(dg.MultiSlotDataGenerator):
send = '|'.join(line.split('|')[:-1]).lower().replace("<br />",
" ").strip()
label = [int(line.split('|')[-1])]
words = [x for x in self._pattern.split(send) if x and x != " "]
feas = [
self._vocab[x] if x in self._vocab else self._unk_id for x in words
......@@ -56,9 +58,11 @@ class IMDBDataset(dg.MultiSlotDataGenerator):
for line in fin:
feas, label = self.get_words_and_label(line)
yield feas, label
import paddle
batch_iter = paddle.batch(
paddle.reader.shuffle(local_iter, buf_size=buf_size),
paddle.reader.shuffle(
local_iter, buf_size=buf_size),
batch_size=batch)
return batch_iter
......@@ -66,13 +70,15 @@ class IMDBDataset(dg.MultiSlotDataGenerator):
def memory_iter():
for i in range(1000):
yield self.return_value
def data_iter():
feas, label = self.get_words_and_label(line)
yield ("words", feas), ("label", label)
return data_iter
if __name__ == "__main__":
imdb = IMDBDataset()
imdb.load_resource("imdb.vocab")
imdb.run_from_stdin()
python/examples/imdb/imdb_web_service_demo.sh
浏览文件 @ 47ae3ad2
......@@ -3,4 +3,3 @@ tar -xzf imdb_service.tar.gz
wget --no-check-certificate https://fleet.bj.bcebos.com/text_classification_data.tar.gz
tar -zxvf text_classification_data.tar.gz
python text_classify_service.py serving_server_model/ workdir imdb.vocab
python/examples/imdb/local_train.py
浏览文件 @ 47ae3ad2
......@@ -11,6 +11,7 @@
# 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
import paddle
......
python/examples/imdb/nets.py
浏览文件 @ 47ae3ad2
# 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, doc-string-with-all-args, doc-string-with-returns
import sys
import time
import numpy as np
......@@ -13,10 +28,9 @@ def bow_net(data,
hid_dim=128,
hid_dim2=96,
class_dim=2):
"""
bow net
"""
emb = fluid.layers.embedding(input=data, size=[dict_dim, emb_dim], is_sparse=True)
""" bow net. """
emb = fluid.layers.embedding(
input=data, size=[dict_dim, emb_dim], is_sparse=True)
bow = fluid.layers.sequence_pool(input=emb, pool_type='sum')
bow_tanh = fluid.layers.tanh(bow)
fc_1 = fluid.layers.fc(input=bow_tanh, size=hid_dim, act="tanh")
......@@ -37,10 +51,9 @@ def cnn_net(data,
hid_dim2=96,
class_dim=2,
win_size=3):
"""
conv net
"""
emb = fluid.layers.embedding(input=data, size=[dict_dim, emb_dim], is_sparse=True)
""" conv net. """
emb = fluid.layers.embedding(
input=data, size=[dict_dim, emb_dim], is_sparse=True)
conv_3 = fluid.nets.sequence_conv_pool(
input=emb,
......@@ -67,9 +80,7 @@ def lstm_net(data,
hid_dim2=96,
class_dim=2,
emb_lr=30.0):
"""
lstm net
"""
""" lstm net. """
emb = fluid.layers.embedding(
input=data,
size=[dict_dim, emb_dim],
......@@ -103,9 +114,7 @@ def gru_net(data,
hid_dim2=96,
class_dim=2,
emb_lr=400.0):
"""
gru net
"""
""" gru net. """
emb = fluid.layers.embedding(
input=data,
size=[dict_dim, emb_dim],
......
python/examples/imdb/test_client.py
浏览文件 @ 47ae3ad2
......@@ -11,6 +11,7 @@
# 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
from imdb_reader import IMDBDataset
import sys
......@@ -31,4 +32,3 @@ for line in sys.stdin:
fetch = ["acc", "cost", "prediction"]
fetch_map = client.predict(feed=feed, fetch=fetch)
print("{} {}".format(fetch_map["prediction"][1], label[0]))
python/examples/imdb/test_client_batch.py
浏览文件 @ 47ae3ad2
......@@ -11,6 +11,7 @@
# 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
......
python/examples/imdb/text_classify_service.py
浏览文件 @ 47ae3ad2
......@@ -11,17 +11,20 @@
# 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_server.web_service import WebService
from imdb_reader import IMDBDataset
import sys
class IMDBService(WebService):
def prepare_dict(self, args={}):
if len(args) == 0:
exit(-1)
self.dataset = IMDBDataset()
self.dataset.load_resource(args["dict_file_path"])
def preprocess(self, feed={}, fetch=[]):
if "words" not in feed:
exit(-1)
......@@ -29,8 +32,9 @@ class IMDBService(WebService):
res_feed["words"] = self.dataset.get_words_only(feed["words"])[0]
return res_feed, fetch
imdb_service = IMDBService(name="imdb")
imdb_service.load_model_config(sys.argv[1])
imdb_service.prepare_server(workdir=sys.argv[2], port=9393, device="cpu")
imdb_service.prepare_dict({"dict_file_path":sys.argv[3]})
imdb_service.prepare_dict({"dict_file_path": sys.argv[3]})
imdb_service.run_server()
python/examples/util/get_acc.py
浏览文件 @ 47ae3ad2
# 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 os
......
python/paddle_serving_client/__init__.py
浏览文件 @ 47ae3ad2
......@@ -199,6 +199,7 @@ class Client(object):
float_feed_names = []
fetch_names = []
counter = 0
batch_size = len(feed_batch)
for feed in feed_batch:
int_slot = []
float_slot = []
......@@ -221,15 +222,21 @@ class Client(object):
if key in self.fetch_names_:
fetch_names.append(key)
result_batch = self.client_handle_.batch_predict(
result_batch = self.result_handle_
res = self.client_handle_.batch_predict(
float_slot_batch, float_feed_names, int_slot_batch, int_feed_names,
fetch_names)
fetch_names, result_batch, self.pid)
result_map_batch = []
for result in result_batch:
for index in range(batch_size):
result_map = {}
for i, name in enumerate(fetch_names):
result_map[name] = result[i]
if self.fetch_names_to_type_[name] == int_type:
result_map[name] = result_batch.get_int64_by_name(name)[
index]
elif self.fetch_names_to_type_[name] == float_type:
result_map[name] = result_batch.get_float_by_name(name)[
index]
result_map_batch.append(result_map)
return result_map_batch
......
python/paddle_serving_client/io/__init__.py
浏览文件 @ 47ae3ad2
......@@ -11,6 +11,7 @@
# 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.fluid import Executor
from paddle.fluid.compiler import CompiledProgram
......@@ -22,6 +23,7 @@ from paddle.fluid.io import save_inference_model
from ..proto import general_model_config_pb2 as model_conf
import os
def save_model(server_model_folder,
client_config_folder,
feed_var_dict,
......@@ -32,8 +34,12 @@ def save_model(server_model_folder,
feed_var_names = [feed_var_dict[x].name for x in feed_var_dict]
target_vars = fetch_var_dict.values()
save_inference_model(server_model_folder, feed_var_names,
target_vars, executor, main_program=main_program)
save_inference_model(
server_model_folder,
feed_var_names,
target_vars,
executor,
main_program=main_program)
config = model_conf.GeneralModelConfig()
......@@ -67,7 +73,7 @@ def save_model(server_model_folder,
fetch_var.fetch_type = 0
if fetch_var_dict[key].dtype == core.VarDesc.VarType.FP32:
fetch_var.fetch_type = 1
fetch_var.fetch_type = 1
if fetch_var.is_lod_tensor:
fetch_var.shape.extend([-1])
......@@ -82,15 +88,15 @@ def save_model(server_model_folder,
cmd = "mkdir -p {}".format(client_config_folder)
os.system(cmd)
with open("{}/serving_client_conf.prototxt".format(client_config_folder), "w") as fout:
with open("{}/serving_client_conf.prototxt".format(client_config_folder),
"w") as fout:
fout.write(str(config))
with open("{}/serving_server_conf.prototxt".format(server_model_folder), "w") as fout:
with open("{}/serving_server_conf.prototxt".format(server_model_folder),
"w") as fout:
fout.write(str(config))
with open("{}/serving_client_conf.stream.prototxt".format(client_config_folder), "wb") as fout:
with open("{}/serving_client_conf.stream.prototxt".format(
client_config_folder), "wb") as fout:
fout.write(config.SerializeToString())
with open("{}/serving_server_conf.stream.prototxt".format(server_model_folder), "wb") as fout:
with open("{}/serving_server_conf.stream.prototxt".format(
server_model_folder), "wb") as fout:
fout.write(config.SerializeToString())
python/paddle_serving_client/metric/acc.py
浏览文件 @ 47ae3ad2
......@@ -11,6 +11,8 @@
# 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
def acc(prob, label, threshold):
# we support prob is the probability for label to be one
......@@ -21,5 +23,3 @@ def acc(prob, label, threshold):
if (prob - threshold) * (label - prob) > 0:
right += 1
return float(right) / total
python/paddle_serving_client/metric/auc.py
浏览文件 @ 47ae3ad2
......@@ -11,6 +11,8 @@
# 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, doc-string-with-all-args, doc-string-with-returns
def tied_rank(x):
"""
......@@ -24,21 +26,22 @@ def tied_rank(x):
score : list of numbers
The tied rank f each element in x
"""
sorted_x = sorted(zip(x,range(len(x))))
sorted_x = sorted(zip(x, range(len(x))))
r = [0 for k in x]
cur_val = sorted_x[0][0]
last_rank = 0
for i in range(len(sorted_x)):
if cur_val != sorted_x[i][0]:
cur_val = sorted_x[i][0]
for j in range(last_rank, i):
r[sorted_x[j][1]] = float(last_rank+1+i)/2.0
for j in range(last_rank, i):
r[sorted_x[j][1]] = float(last_rank + 1 + i) / 2.0
last_rank = i
if i==len(sorted_x)-1:
for j in range(last_rank, i+1):
r[sorted_x[j][1]] = float(last_rank+i+2)/2.0
if i == len(sorted_x) - 1:
for j in range(last_rank, i + 1):
r[sorted_x[j][1]] = float(last_rank + i + 2) / 2.0
return r
def auc(actual, posterior):
"""
Computes the area under the receiver-operater characteristic (AUC)
......@@ -56,10 +59,9 @@ def auc(actual, posterior):
The mean squared error between actual and posterior
"""
r = tied_rank(posterior)
num_positive = len([0 for x in actual if x==1])
num_negative = len(actual)-num_positive
sum_positive = sum([r[i] for i in range(len(r)) if actual[i]==1])
auc = ((sum_positive - num_positive*(num_positive+1)/2.0) /
(num_negative*num_positive))
num_positive = len([0 for x in actual if x == 1])
num_negative = len(actual) - num_positive
sum_positive = sum([r[i] for i in range(len(r)) if actual[i] == 1])
auc = ((sum_positive - num_positive * (num_positive + 1) / 2.0) /
(num_negative * num_positive))
return auc
python/paddle_serving_client/utils/__init__.py
浏览文件 @ 47ae3ad2
......@@ -11,18 +11,26 @@
# 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
import subprocess
import argparse
from multiprocessing import Pool
def benchmark_args():
parser = argparse.ArgumentParser("benchmark")
parser.add_argument("--thread", type=int, default=10, help="concurrecy")
parser.add_argument("--model", type=str, default="", help="model for evaluation")
parser.add_argument("--endpoint", type=str, default="127.0.0.1:9292", help="endpoint of server")
parser.add_argument("--request", type=str, default="rpc", help="mode of service")
parser.add_argument(
"--model", type=str, default="", help="model for evaluation")
parser.add_argument(
"--endpoint",
type=str,
default="127.0.0.1:9292",
help="endpoint of server")
parser.add_argument(
"--request", type=str, default="rpc", help="mode of service")
return parser.parse_args()
......
python/paddle_serving_server/serve.py
浏览文件 @ 47ae3ad2
......@@ -19,16 +19,26 @@ Usage:
"""
import argparse
def parse_args():
def parse_args(): # pylint: disable=doc-string-missing
parser = argparse.ArgumentParser("serve")
parser.add_argument("--thread", type=int, default=10, help="Concurrency of server")
parser.add_argument("--model", type=str, default="", help="Model for serving")
parser.add_argument("--port", type=int, default=9292, help="Port the server")
parser.add_argument("--workdir", type=str, default="workdir", help="Working dir of current service")
parser.add_argument("--device", type=str, default="cpu", help="Type of device")
parser.add_argument(
"--thread", type=int, default=10, help="Concurrency of server")
parser.add_argument(
"--model", type=str, default="", help="Model for serving")
parser.add_argument(
"--port", type=int, default=9292, help="Port the server")
parser.add_argument(
"--workdir",
type=str,
default="workdir",
help="Working dir of current service")
parser.add_argument(
"--device", type=str, default="cpu", help="Type of device")
return parser.parse_args()
def start_standard_model():
def start_standard_model(): # pylint: disable=doc-string-missing
args = parse_args()
thread_num = args.thread
model = args.model
......@@ -59,5 +69,6 @@ def start_standard_model():
server.prepare_server(workdir=workdir, port=port, device=device)
server.run_server()
if __name__ == "__main__":
start_standard_model()
python/paddle_serving_server/web_serve.py
浏览文件 @ 47ae3ad2
......@@ -21,19 +21,31 @@ import argparse
from multiprocessing import Pool, Process
from .web_service import WebService
def parse_args():
def parse_args(): # pylint: disable=doc-string-missing
parser = argparse.ArgumentParser("web_serve")
parser.add_argument("--thread", type=int, default=10, help="Concurrency of server")
parser.add_argument("--model", type=str, default="", help="Model for serving")
parser.add_argument("--port", type=int, default=9292, help="Port the server")
parser.add_argument("--workdir", type=str, default="workdir", help="Working dir of current service")
parser.add_argument("--device", type=str, default="cpu", help="Type of device")
parser.add_argument("--name", type=str, default="default", help="Default service name")
parser.add_argument(
"--thread", type=int, default=10, help="Concurrency of server")
parser.add_argument(
"--model", type=str, default="", help="Model for serving")
parser.add_argument(
"--port", type=int, default=9292, help="Port the server")
parser.add_argument(
"--workdir",
type=str,
default="workdir",
help="Working dir of current service")
parser.add_argument(
"--device", type=str, default="cpu", help="Type of device")
parser.add_argument(
"--name", type=str, default="default", help="Default service name")
return parser.parse_args()
if __name__ == "__main__":
args = parse_args()
service = WebService(name=args.name)
service.load_model_config(args.model)
service.prepare_server(workdir=args.workdir, port=args.port, device=args.device)
service.prepare_server(
workdir=args.workdir, port=args.port, device=args.device)
service.run_server()
python/paddle_serving_server/web_service.py
浏览文件 @ 47ae3ad2
......@@ -12,11 +12,14 @@
# See the License for the specific language governing permissions and
# limitations under the License.
#!flask/bin/python
# pylint: disable=doc-string-missing
from flask import Flask, request, abort
from multiprocessing import Pool, Process
from paddle_serving_server import OpMaker, OpSeqMaker, Server
from paddle_serving_client import Client
class WebService(object):
def __init__(self, name="default_service"):
self.name = name
......@@ -38,7 +41,7 @@ class WebService(object):
server.set_num_threads(16)
server.load_model_config(self.model_config)
server.prepare_server(
workdir=self.workdir, port=self.port+1, device=self.device)
workdir=self.workdir, port=self.port + 1, device=self.device)
server.run_server()
def prepare_server(self, workdir="", port=9393, device="cpu"):
......@@ -51,8 +54,9 @@ class WebService(object):
client_service = Client()
client_service.load_client_config(
"{}/serving_server_conf.prototxt".format(self.model_config))
client_service.connect(["127.0.0.1:{}".format(self.port+1)])
client_service.connect(["127.0.0.1:{}".format(self.port + 1)])
service_name = "/" + self.name + "/prediction"
@app_instance.route(service_name, methods=['POST'])
def get_prediction():
if not request.json:
......@@ -60,16 +64,24 @@ class WebService(object):
if "fetch" not in request.json:
abort(400)
feed, fetch = self.preprocess(request.json, request.json["fetch"])
if "fetch" in feed:
del feed["fetch"]
fetch_map = client_service.predict(feed=feed, fetch=fetch)
fetch_map = self.postprocess(feed=request.json, fetch=fetch, fetch_map=fetch_map)
fetch_map = self.postprocess(
feed=request.json, fetch=fetch, fetch_map=fetch_map)
return fetch_map
app_instance.run(host="127.0.0.1", port=self.port, threaded=False, processes=1)
app_instance.run(host="0.0.0.0",
port=self.port,
threaded=False,
processes=1)
def run_server(self):
import socket
localIP = socket.gethostbyname(socket.gethostname())
print("web service address:")
print("http://{}:{}/{}/prediction".format(localIP, self.port, self.name))
print("http://{}:{}/{}/prediction".format(localIP, self.port,
self.name))
p_rpc = Process(target=self._launch_rpc_service)
p_web = Process(target=self._launch_web_service)
p_rpc.start()
......@@ -82,4 +94,3 @@ class WebService(object):
def postprocess(self, feed={}, fetch=[], fetch_map={}):
return fetch_map
python/paddle_serving_server_gpu/__init__.py
浏览文件 @ 47ae3ad2
......@@ -11,6 +11,7 @@
# 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
from .proto import server_configure_pb2 as server_sdk
......@@ -21,6 +22,28 @@ import socket
import paddle_serving_server_gpu as paddle_serving_server
from version import serving_server_version
from contextlib import closing
import argparse
def serve_args():
parser = argparse.ArgumentParser("serve")
parser.add_argument(
"--thread", type=int, default=10, help="Concurrency of server")
parser.add_argument(
"--model", type=str, default="", help="Model for serving")
parser.add_argument(
"--port", type=int, default=9292, help="Port of the starting gpu")
parser.add_argument(
"--workdir",
type=str,
default="workdir",
help="Working dir of current service")
parser.add_argument(
"--device", type=str, default="gpu", help="Type of device")
parser.add_argument("--gpu_ids", type=str, default="", help="gpu ids")
parser.add_argument(
"--name", type=str, default="default", help="Default service name")
return parser.parse_args()
class OpMaker(object):
......@@ -126,7 +149,8 @@ class Server(object):
self.model_config_path = model_config_path
self.engine.name = "general_model"
self.engine.reloadable_meta = model_config_path + "/fluid_time_file"
#self.engine.reloadable_meta = model_config_path + "/fluid_time_file"
self.engine.reloadable_meta = self.workdir + "/fluid_time_file"
os.system("touch {}".format(self.engine.reloadable_meta))
self.engine.reloadable_type = "timestamp_ne"
self.engine.runtime_thread_num = 0
......@@ -154,6 +178,7 @@ class Server(object):
self.infer_service_conf.services.extend([infer_service])
def _prepare_resource(self, workdir):
self.workdir = workdir
if self.resource_conf == None:
with open("{}/{}".format(workdir, self.general_model_config_fn),
"w") as fout:
......@@ -217,6 +242,7 @@ class Server(object):
if not self.check_port(port):
raise SystemExit("Prot {} is already used".format(port))
self.set_port(port)
self._prepare_resource(workdir)
self._prepare_engine(self.model_config_path, device)
self._prepare_infer_service(port)
......
python/paddle_serving_server_gpu/serve.py
浏览文件 @ 47ae3ad2
......@@ -17,36 +17,23 @@ Usage:
Example:
python -m paddle_serving_server.serve --model ./serving_server_model --port 9292
"""
import os
import argparse
from multiprocessing import Pool, Process
from paddle_serving_server_gpu import serve_args
def parse_args():
parser = argparse.ArgumentParser("serve")
parser.add_argument(
"--thread", type=int, default=10, help="Concurrency of server")
parser.add_argument(
"--model", type=str, default="", help="Model for serving")
parser.add_argument(
"--port", type=int, default=9292, help="Port the server")
parser.add_argument(
"--workdir",
type=str,
default="workdir",
help="Working dir of current service")
parser.add_argument(
"--device", type=str, default="gpu", help="Type of device")
parser.add_argument("--gpuid", type=int, default=0, help="Index of GPU")
return parser.parse_args()
def start_standard_model():
args = parse_args()
def start_gpu_card_model(gpuid, args): # pylint: disable=doc-string-missing
gpuid = int(gpuid)
device = "gpu"
port = args.port
if gpuid == -1:
device = "cpu"
elif gpuid >= 0:
port = args.port + gpuid
thread_num = args.thread
model = args.model
port = args.port
workdir = args.workdir
device = args.device
gpuid = args.gpuid
workdir = "{}_{}".format(args.workdir, gpuid)
if model == "":
print("You must specify your serving model")
......@@ -69,9 +56,33 @@ def start_standard_model():
server.load_model_config(model)
server.prepare_server(workdir=workdir, port=port, device=device)
server.set_gpuid(gpuid)
if gpuid >= 0:
server.set_gpuid(gpuid)
server.run_server()
def start_multi_card(args): # pylint: disable=doc-string-missing
gpus = ""
if args.gpu_ids == "":
gpus = os.environ["CUDA_VISIBLE_DEVICES"]
else:
gpus = args.gpu_ids.split(",")
if len(gpus) <= 0:
start_gpu_card_model(-1)
else:
gpu_processes = []
for i, gpu_id in enumerate(gpus):
p = Process(
target=start_gpu_card_model, args=(
i,
args, ))
gpu_processes.append(p)
for p in gpu_processes:
p.start()
for p in gpu_processes:
p.join()
if __name__ == "__main__":
start_standard_model()
args = serve_args()
start_multi_card(args)
python/paddle_serving_server_gpu/web_serve.py
浏览文件 @ 47ae3ad2
......@@ -17,35 +17,23 @@ Usage:
Example:
python -m paddle_serving_server.web_serve --model ./serving_server_model --port 9292
"""
import argparse
import os
from multiprocessing import Pool, Process
from .web_service import WebService
def parse_args():
parser = argparse.ArgumentParser("web_serve")
parser.add_argument(
"--thread", type=int, default=10, help="Concurrency of server")
parser.add_argument(
"--model", type=str, default="", help="Model for serving")
parser.add_argument(
"--port", type=int, default=9292, help="Port the server")
parser.add_argument(
"--workdir",
type=str,
default="workdir",
help="Working dir of current service")
parser.add_argument(
"--device", type=str, default="cpu", help="Type of device")
parser.add_argument(
"--name", type=str, default="default", help="Default service name")
return parser.parse_args()
import paddle_serving_server_gpu as serving
from paddle_serving_server_gpu import serve_args
if __name__ == "__main__":
args = parse_args()
service = WebService(name=args.name)
service.load_model_config(args.model)
service.prepare_server(
args = serve_args()
web_service = WebService(name=args.name)
web_service.load_model_config(args.model)
gpu_ids = []
if args.gpu_ids == "":
if "CUDA_VISIBLE_DEVICES" in os.environ:
gpu_ids = os.environ["CUDA_VISIBLE_DEVICES"]
if len(gpu_ids) > 0:
gpus = [int(x) for x in gpu_ids.split(",")]
web_service.set_gpus(gpus)
web_service.prepare_server(
workdir=args.workdir, port=args.port, device=args.device)
service.run_server()
python/paddle_serving_server_gpu/web_service.py
浏览文件 @ 47ae3ad2
......@@ -12,49 +12,92 @@
# See the License for the specific language governing permissions and
# limitations under the License.
#!flask/bin/python
# pylint: disable=doc-string-missing
from flask import Flask, request, abort
from multiprocessing import Pool, Process
from paddle_serving_server_gpu import OpMaker, OpSeqMaker, Server
import paddle_serving_server_gpu as serving
from paddle_serving_client import Client
from .serve import start_multi_card
import time
import random
class WebService(object):
def __init__(self, name="default_service"):
self.name = name
self.gpus = []
self.rpc_service_list = []
def load_model_config(self, model_config):
self.model_config = model_config
def _launch_rpc_service(self):
op_maker = OpMaker()
def set_gpus(self, gpus):
self.gpus = gpus
def default_rpc_service(self,
workdir="conf",
port=9292,
gpuid=0,
thread_num=10):
device = "gpu"
if gpuid == -1:
device = "cpu"
op_maker = serving.OpMaker()
read_op = op_maker.create('general_reader')
general_infer_op = op_maker.create('general_infer')
general_response_op = op_maker.create('general_response')
op_seq_maker = OpSeqMaker()
op_seq_maker = serving.OpSeqMaker()
op_seq_maker.add_op(read_op)
op_seq_maker.add_op(general_infer_op)
op_seq_maker.add_op(general_response_op)
server = Server()
server = serving.Server()
server.set_op_sequence(op_seq_maker.get_op_sequence())
server.set_num_threads(16)
server.set_gpuid = self.gpuid
server.set_num_threads(thread_num)
server.load_model_config(self.model_config)
server.prepare_server(
workdir=self.workdir, port=self.port + 1, device=self.device)
server.run_server()
if gpuid >= 0:
server.set_gpuid(gpuid)
server.prepare_server(workdir=workdir, port=port, device=device)
return server
def _launch_rpc_service(self, service_idx):
self.rpc_service_list[service_idx].run_server()
def prepare_server(self, workdir="", port=9393, device="gpu", gpuid=0):
self.workdir = workdir
self.port = port
self.device = device
self.gpuid = gpuid
if len(self.gpus) == 0:
# init cpu service
self.rpc_service_list.append(
self.default_rpc_service(
self.workdir, self.port + 1, -1, thread_num=10))
else:
for i, gpuid in enumerate(self.gpus):
self.rpc_service_list.append(
self.default_rpc_service(
"{}_{}".format(self.workdir, i),
self.port + 1 + i,
gpuid,
thread_num=10))
def _launch_web_service(self):
def _launch_web_service(self, gpu_num):
app_instance = Flask(__name__)
client_service = Client()
client_service.load_client_config(
"{}/serving_server_conf.prototxt".format(self.model_config))
client_service.connect(["127.0.0.1:{}".format(self.port + 1)])
client_list = []
if gpu_num > 1:
gpu_num = 0
for i in range(gpu_num):
client_service = Client()
client_service.load_client_config(
"{}/serving_server_conf.prototxt".format(self.model_config))
client_service.connect(["127.0.0.1:{}".format(self.port + i + 1)])
client_list.append(client_service)
time.sleep(1)
service_name = "/" + self.name + "/prediction"
@app_instance.route(service_name, methods=['POST'])
......@@ -64,12 +107,12 @@ class WebService(object):
if "fetch" not in request.json:
abort(400)
feed, fetch = self.preprocess(request.json, request.json["fetch"])
fetch_map = client_service.predict(feed=feed, fetch=fetch)
fetch_map = client_list[0].predict(feed=feed, fetch=fetch)
fetch_map = self.postprocess(
feed=request.json, fetch=fetch, fetch_map=fetch_map)
return fetch_map
app_instance.run(host="127.0.0.1",
app_instance.run(host="0.0.0.0",
port=self.port,
threaded=False,
processes=1)
......@@ -80,12 +123,21 @@ class WebService(object):
print("web service address:")
print("http://{}:{}/{}/prediction".format(localIP, self.port,
self.name))
p_rpc = Process(target=self._launch_rpc_service)
p_web = Process(target=self._launch_web_service)
p_rpc.start()
rpc_processes = []
for idx in range(len(self.rpc_service_list)):
p_rpc = Process(target=self._launch_rpc_service, args=(idx, ))
rpc_processes.append(p_rpc)
for p in rpc_processes:
p.start()
p_web = Process(
target=self._launch_web_service, args=(len(self.gpus), ))
p_web.start()
for p in rpc_processes:
p.join()
p_web.join()
p_rpc.join()
def preprocess(self, feed={}, fetch=[]):
return feed, fetch
......
python/setup.py.in
浏览文件 @ 47ae3ad2
......@@ -77,4 +77,3 @@ setup(
],
license='Apache 2.0',
keywords=('paddle-serving serving-client deployment industrial easy-to-use'))
python/setup.py.server.in
浏览文件 @ 47ae3ad2
......@@ -38,7 +38,7 @@ max_version, mid_version, min_version = python_version()
REQUIRED_PACKAGES = [
'six >= 1.10.0', 'protobuf >= 3.1.0',
'paddle_serving_client', 'flask'
'paddle_serving_client', 'flask >= 1.1.1'
]
if not find_package("paddlepaddle") and not find_package("paddlepaddle-gpu"):
......
tools/Dockerfile.ci 0 → 100644
浏览文件 @ 47ae3ad2
FROM centos:7.3.1611
RUN yum -y install wget >/dev/null \
&& yum -y install gcc gcc-c++ make glibc-static which >/dev/null \
&& yum -y install git openssl-devel curl-devel bzip2-devel python-devel >/dev/null \
&& wget https://cmake.org/files/v3.2/cmake-3.2.0-Linux-x86_64.tar.gz >/dev/null \
&& tar xzf cmake-3.2.0-Linux-x86_64.tar.gz \
&& mv cmake-3.2.0-Linux-x86_64 /usr/local/cmake3.2.0 \
&& echo 'export PATH=/usr/local/cmake3.2.0/bin:$PATH' >> /root/.bashrc \
&& rm cmake-3.2.0-Linux-x86_64.tar.gz \
&& wget https://dl.google.com/go/go1.14.linux-amd64.tar.gz >/dev/null \
&& tar xzf go1.14.linux-amd64.tar.gz \
&& mv go /usr/local/go \
&& echo 'export GOROOT=/usr/local/go' >> /root/.bashrc \
&& echo 'export PATH=/usr/local/go/bin:$PATH' >> /root/.bashrc \
&& rm go1.14.linux-amd64.tar.gz \
&& 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 \
&& pip install google protobuf setuptools wheel flask >/dev/null \
&& rm get-pip.py \
&& wget http://nixos.org/releases/patchelf/patchelf-0.10/patchelf-0.10.tar.bz2 \
&& yum -y install bzip2 >/dev/null \
&& tar -jxf patchelf-0.10.tar.bz2 \
&& cd patchelf-0.10 \
&& ./configure --prefix=/usr \
&& make >/dev/null && make install >/dev/null \
&& cd .. \
&& rm -rf patchelf-0.10* \
&& yum -y update >/dev/null \
&& yum -y install dnf >/dev/null \
&& yum -y install dnf-plugins-core >/dev/null \
&& dnf copr enable alonid/llvm-3.8.0 -y \
&& dnf install llvm-3.8.0 clang-3.8.0 compiler-rt-3.8.0 -y \
&& echo 'export PATH=/opt/llvm-3.8.0/bin:$PATH' >> /root/.bashrc
tools/cpp_examples/demo-client/paddle_serving_client.egg-info/SOURCES.txt
浏览文件 @ 47ae3ad2
......@@ -3,4 +3,4 @@ paddle_serving_client.egg-info/PKG-INFO
paddle_serving_client.egg-info/SOURCES.txt
paddle_serving_client.egg-info/dependency_links.txt
paddle_serving_client.egg-info/not-zip-safe
paddle_serving_client.egg-info/top_level.txt
\ No newline at end of file
paddle_serving_client.egg-info/top_level.txt
tools/cpp_examples/demo-client/src/general_model.cpp
浏览文件 @ 47ae3ad2
......@@ -12,8 +12,8 @@
// See the License for the specific language governing permissions and
// limitations under the License.
#include "general_model.h" // NOLINT
#include <fstream>
#include "general_model.h"
#include "sdk-cpp/builtin_format.pb.h"
#include "sdk-cpp/include/common.h"
#include "sdk-cpp/include/predictor_sdk.h"
......@@ -28,7 +28,7 @@ namespace baidu {
namespace paddle_serving {
namespace general_model {
void PredictorClient::init(const std::string & conf_file) {
void PredictorClient::init(const std::string &conf_file) {
_conf_file = conf_file;
std::ifstream fin(conf_file);
if (!fin) {
......@@ -65,9 +65,8 @@ void PredictorClient::init(const std::string & conf_file) {
}
}
void PredictorClient::set_predictor_conf(
const std::string & conf_path,
const std::string & conf_file) {
void PredictorClient::set_predictor_conf(const std::string &conf_path,
const std::string &conf_file) {
_predictor_path = conf_path;
_predictor_conf = conf_file;
}
......@@ -80,53 +79,51 @@ int PredictorClient::create_predictor() {
_api.thrd_initialize();
}
void PredictorClient::predict(
const std::vector<std::vector<float> > & float_feed,
const std::vector<std::string> & float_feed_name,
const std::vector<std::vector<int64_t> > & int_feed,
const std::vector<std::string> & int_feed_name,
const std::vector<std::string> & fetch_name,
FetchedMap * fetch_result) {
void PredictorClient::predict(const std::vector<std::vector<float>> &float_feed,
const std::vector<std::string> &float_feed_name,
const std::vector<std::vector<int64_t>> &int_feed,
const std::vector<std::string> &int_feed_name,
const std::vector<std::string> &fetch_name,
FetchedMap *fetch_result) {
_api.thrd_clear();
_predictor = _api.fetch_predictor("general_model");
Request req;
std::vector<Tensor *> tensor_vec;
FeedInst * inst = req.add_insts();
for (auto & name : float_feed_name) {
FeedInst *inst = req.add_insts();
for (auto &name : float_feed_name) {
tensor_vec.push_back(inst->add_tensor_array());
}
for (auto & name : int_feed_name) {
for (auto &name : int_feed_name) {
tensor_vec.push_back(inst->add_tensor_array());
}
int vec_idx = 0;
for (auto & name : float_feed_name) {
for (auto &name : float_feed_name) {
int idx = _feed_name_to_idx[name];
Tensor * tensor = tensor_vec[idx];
Tensor *tensor = tensor_vec[idx];
for (int j = 0; j < _shape[idx].size(); ++j) {
tensor->add_shape(_shape[idx][j]);
}
tensor->set_elem_type(1);
for (int j = 0; j < float_feed[vec_idx].size(); ++j) {
tensor->add_data(
(char *)(&(float_feed[vec_idx][j])), sizeof(float));
tensor->add_data((char *)(&(float_feed[vec_idx][j])), // NOLINT
sizeof(float));
}
vec_idx++;
}
vec_idx = 0;
for (auto & name : int_feed_name) {
for (auto &name : int_feed_name) {
int idx = _feed_name_to_idx[name];
Tensor * tensor = tensor_vec[idx];
Tensor *tensor = tensor_vec[idx];
for (int j = 0; j < _shape[idx].size(); ++j) {
tensor->add_shape(_shape[idx][j]);
}
tensor->set_elem_type(0);
for (int j = 0; j < int_feed[vec_idx].size(); ++j) {
tensor->add_data(
(char *)(&(int_feed[vec_idx][j])), sizeof(int64_t));
tensor->add_data((char *)(&(int_feed[vec_idx][j])), // NOLINT
sizeof(int64_t));
}
vec_idx++;
}
......@@ -139,13 +136,13 @@ void PredictorClient::predict(
LOG(ERROR) << "failed call predictor with req: " << req.ShortDebugString();
exit(-1);
} else {
for (auto & name : fetch_name) {
for (auto &name : fetch_name) {
int idx = _fetch_name_to_idx[name];
int len = res.insts(0).tensor_array(idx).data_size();
(*fetch_result)[name].resize(len);
for (int i = 0; i < len; ++i) {
(*fetch_result)[name][i] = *(const float *)
res.insts(0).tensor_array(idx).data(i).c_str();
(*fetch_result)[name][i] =
*(const float *)res.insts(0).tensor_array(idx).data(i).c_str();
}
}
}
......@@ -154,12 +151,12 @@ void PredictorClient::predict(
}
void PredictorClient::predict_with_profile(
const std::vector<std::vector<float> > & float_feed,
const std::vector<std::string> & float_feed_name,
const std::vector<std::vector<int64_t> > & int_feed,
const std::vector<std::string> & int_feed_name,
const std::vector<std::string> & fetch_name,
FetchedMap * fetch_result) {
const std::vector<std::vector<float>> &float_feed,
const std::vector<std::string> &float_feed_name,
const std::vector<std::vector<int64_t>> &int_feed,
const std::vector<std::string> &int_feed_name,
const std::vector<std::string> &fetch_name,
FetchedMap *fetch_result) {
return;
}
......
tools/cpp_examples/demo-client/src/general_model.h
浏览文件 @ 47ae3ad2
......@@ -18,9 +18,9 @@
#include <unistd.h>
#include <fstream>
#include <map>
#include <string>
#include <vector>
#include <map>
#include "sdk-cpp/builtin_format.pb.h"
#include "sdk-cpp/general_model_service.pb.h"
......@@ -37,44 +37,40 @@ namespace general_model {
typedef std::map<std::string, std::vector<float>> FetchedMap;
class PredictorClient {
public:
PredictorClient() {}
~PredictorClient() {}
void init(const std::string & client_conf);
void set_predictor_conf(
const std::string& conf_path,
const std::string& conf_file);
void init(const std::string& client_conf);
void set_predictor_conf(const std::string& conf_path,
const std::string& conf_file);
int create_predictor();
void predict(
const std::vector<std::vector<float> > & float_feed,
const std::vector<std::string> & float_feed_name,
const std::vector<std::vector<int64_t> > & int_feed,
const std::vector<std::string> & int_feed_name,
const std::vector<std::string> & fetch_name,
FetchedMap * result_map);
void predict(const std::vector<std::vector<float>>& float_feed,
const std::vector<std::string>& float_feed_name,
const std::vector<std::vector<int64_t>>& int_feed,
const std::vector<std::string>& int_feed_name,
const std::vector<std::string>& fetch_name,
FetchedMap* result_map);
void predict_with_profile(
const std::vector<std::vector<float> > & float_feed,
const std::vector<std::string> & float_feed_name,
const std::vector<std::vector<int64_t> > & int_feed,
const std::vector<std::string> & int_feed_name,
const std::vector<std::string> & fetch_name,
FetchedMap * result_map);
void predict_with_profile(const std::vector<std::vector<float>>& float_feed,
const std::vector<std::string>& float_feed_name,
const std::vector<std::vector<int64_t>>& int_feed,
const std::vector<std::string>& int_feed_name,
const std::vector<std::string>& fetch_name,
FetchedMap* result_map);
private:
PredictorApi _api;
Predictor * _predictor;
Predictor* _predictor;
std::string _predictor_conf;
std::string _predictor_path;
std::string _conf_file;
std::map<std::string, int> _feed_name_to_idx;
std::map<std::string, int> _fetch_name_to_idx;
std::map<std::string, std::string> _fetch_name_to_var_name;
std::vector<std::vector<int> > _shape;
std::vector<std::vector<int>> _shape;
};
} // namespace general_model
......
tools/cpp_examples/demo-client/src/general_model_main.cpp
浏览文件 @ 47ae3ad2
......@@ -15,20 +15,20 @@
#include <fstream>
#include <vector>
#include "general_model.h"
#include "general_model.h" // NOLINT
using namespace std;
using namespace std; // NOLINT
using baidu::paddle_serving::general_model::PredictorClient;
using baidu::paddle_serving::general_model::FetchedMap;
int main(int argc, char * argv[]) {
PredictorClient * client = new PredictorClient();
int main(int argc, char* argv[]) {
PredictorClient* client = new PredictorClient();
client->init("inference.conf");
client->set_predictor_conf("./", "predictor.conf");
client->create_predictor();
std::vector<std::vector<float> > float_feed;
std::vector<std::vector<int64_t> > int_feed;
std::vector<std::vector<float>> float_feed;
std::vector<std::vector<int64_t>> int_feed;
std::vector<std::string> float_feed_name;
std::vector<std::string> int_feed_name = {"words", "label"};
std::vector<std::string> fetch_name = {"cost", "acc", "prediction"};
......@@ -53,13 +53,14 @@ int main(int argc, char * argv[]) {
cin >> label;
int_feed.push_back({label});
FetchedMap result;
client->predict(
float_feed, float_feed_name,
int_feed, int_feed_name, fetch_name,
&result);
client->predict(float_feed,
float_feed_name,
int_feed,
int_feed_name,
fetch_name,
&result);
cout << label << "\t" << result["prediction"][1] << endl;
......
tools/cpp_examples/demo-client/src/load_general_model.cpp
浏览文件 @ 47ae3ad2
......@@ -18,14 +18,14 @@
#include <fstream>
#include "core/sdk-cpp/builtin_format.pb.h"
#include "core/sdk-cpp/load_general_model_service.pb.h"
#include "core/sdk-cpp/include/common.h"
#include "core/sdk-cpp/include/predictor_sdk.h"
#include "core/sdk-cpp/load_general_model_service.pb.h"
using baidu::paddle_serving::sdk_cpp::Predictor;
using baidu::paddle_serving::sdk_cpp::PredictorApi;
using baidu::paddle_serving::predictor::
load_general_model_service::RequestAndResponse;
using baidu::paddle_serving::predictor::load_general_model_service::
RequestAndResponse;
int create_req(RequestAndResponse& req) { // NOLINT
req.set_a(1);
......
tools/serving_build.sh 0 → 100644
浏览文件 @ 47ae3ad2
此差异已折叠。
tools/serving_check_style.sh 0 → 100644
浏览文件 @ 47ae3ad2
此差异已折叠。
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