tensorflow serving目录解读
Posted
tags:
篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了tensorflow serving目录解读相关的知识，希望对你有一定的参考价值。
tf_serving-----WORKSPACE|
-----tensorflow-serving/----BUILD
| |
| -----workspace.bzl
| |
| -----example/-------BUILD
| |
| --------imagenet_lsvrc_2015_synsets.txt
| |
| --------imagenet_metadata.txt
| |
| --------inception_client.cc
| |
| --------inception_client.py
| |
| --------inception_k8s.yaml
| |
| --------inception_saved_model.py
| |
| --------mnist_client.py
| |
| --------mnist_iput_data.py
| |
| --------mnist_saved_model.py
|
-----------tensorflow/-------BUILD
| |
| --------WORKSPACE
| |
| ----tensorflow/------BUILD
| |
| -------workspace.bzl
|
-----------tf_models/----WORKSPACE
|
-----official/
|
-----tutorials/
|
-----research/-------inception/-------------------WORKSPACE
|
-------------------inception/--------BUILD
|
---------inception_train.py
|
---------inception_model.py
|
---------inception_eval.py
|
---------inception_distributed_train.py
解读：
项目文件夹名称为"tf_serving",WORKSPACE文件放在"tf_serving"项目文件夹的根目录
*************************************************************************
tf_serving/WORKSPACE解读：
*************************************************************************
#声明工作空间名称，与项目名称一致
workspace(name = "tf_serving")
#声明本地仓库'tensorflow'的名称及路径
local_repository(name = "org_tensorflow", path = "tensorflow",)
#声明io_bazel_rules_closure
http_archive(……)
#把新的Tensorflow Serving依赖加到workspace.bzl
load("//tensorflow_serving:workspace.bzl","tf_serving_workspace")
tf_serving_workspace()
#指定bazel最低要求版本
load("@org_tensorflow//tensorflow::workspace.bzl", "check_version")
check_version("0.5.4")*************************************************************************
tf_serving/tensorflow_serving/workspace.bzl解读：
*************************************************************************
#tensorFlow Serving外部依赖加载到WORKSPACE文件
load('@org_tensorflow//tensorflow::workspace.bzl', 'tf_workspace')
#此处是所有Tensorflow Serving的外部依赖。
#workspace_dir是Tensorflow Serving repo的绝对路径，如果是作为submodule
#连接，路径形式应该是'__workspace_dir__ + "serving"'
def tf_serving_workspace():
native.new_local_repository(
name = "inception_model",
path = "tf_models/research/inception",
build_file = "tf_models/research/inception/inception/BUILD",
)
tf_workspace(path_prefix = "", tf_repo_name = "org_tensorflow")
#gRPC依赖
native.bind(
name = "libssl",
actual = "@boringssl//:ssl",
)
native.bind(
name = "zlib",
actual = "@zlib_archive//:zlib")*************************************************************************
tf_serving/tensorflow_serving/BUILD解读：
*************************************************************************
# Tensorflow serving描述 package( default_visibility=["//tensorflow_serving:internal"], ) licenses(["notice"]) exports_files(["LICENSE"]) #开放源代码标记 package_group( name = "internal", package = [ "//tensorflow_serving/...", ], ) filegroup( name = "all_files", srcs = glob( ["**/*"], exclude = [ "**/METADATA", "**/OWNERS", "g3doc/sitemap.md", ], ), )
*************************************************************************
tf_serving/tensorflow/tensorflow/workspace.bzl解读：
*************************************************************************
# 可以在WORKSPACE文件中加载的Tensorflow外部依赖 load(……) def _is_windows(): …… def _get_env_var(): …… # 从'native.bazel_version'中解析bazel版本字符串 def _parse_bazel_version(): …… # 检查正在使用指定版本的bazel def check_version(): …… # 支持将Tensorflow作为submodule的临时工作区 def _temp_workaround_http_archive_impl(): …… # 如果非零code退出，则执行指定参数的命令并调用'fail' def _execute_and_check_ret_code(): …… # 在仓库根目录应用补丁文件 def _apply_patch(): …… # 下载仓库，在根节点应用补丁 def _patched_http_archive_impl(): …… # 如果Tensorflow连接为submodule，path_prefix不再使用 # tf_repo_name正在考虑中 def tf_workspace(): ……
*************************************************************************
tf_serving/tensorflow/tensorflow/BUILD解读：
*************************************************************************
package(default_visibility = [":internal"])
licenses(["notice"])
exports_files([
"LICENSE",
"ACKNOWLEDGMENTS",
# leakr文件用于//third_party/cloud_tpu
"leakr_badwords.dic",
"leakr_badfiles.dic",
])
load("//tensorflow:tensorflow.bzl", "tf_cc_shared_object")
load("//tensorflow/core:platform/default/build_config.bzl",
"tf_additional_binary_deps",
)
# 各种config setting
config_setting()
package_group()
filegroup()
py_library()
filegroup(
name = "all_opensource_files",
data = [
":all_files",
"//tensorflow/c:all_files",
"//tensorflow/cc:all_files",
……],
visibility = [':__subpackages__'],
)
load("//third_party/mkl:build_defs.bzl",
"if_mkl",
)
filegroup(
name = "intel_binary_blob",
data = if_mkl(
[
"//third_party/mkl:intel_binary_blob",
],
),
)
filegroup(
name = "docs_src",
data = glob(["docs_src/**/*.md"]),
)
tf_cc_shared_object(
……
)以上是关于tensorflow serving目录解读的主要内容，如果未能解决你的问题，请参考以下文章
谷歌把TensorFlow送给世界，世界却用它人工合成色情片！
Tensorflow r1.12及tensorflow serving r1.12 GPU版本编译遇到的问题