Skip to content

tensorflow.md

Latest commit

 

History

History
104 lines (76 loc) · 3.47 KB

tensorflow.md

File metadata and controls

104 lines (76 loc) · 3.47 KB
Raw

从源代码编译

参考Build from source和How to build TensorFlow 2.0 on Ubuntu 18.04 (x86_64) with Bazelisk

下载 TensorFlow

git clone https://github.com/tensorflow/tensorflow.git
cd tensorflow
git checkout branch_name  # r1.15, r2.2, r2.3, etc.

安装 Bazel

使用 bazelisk 自动安装不同的版本

安装 bazelisk，参考 Installing Bazel using Bazelisk

  • npm install -g @bazel/bazelisk
    • 安装 npm: sudo apt install npm
  • go get github.com/bazelbuild/bazelisk
    • 安装 Go: sudo apt install golang
    • 需要将go加入PATH，如果没有 bazel 只有 bazelisk，需要 ln -s bazelisk bazel

生成 TensorFlow 需要的 bazel 版本配置文件，如果没有的话（一般是 0.26.1，可参考 configure.py 文件_TF_MIN_BAZEL_VERSION ）：echo "0.26.1" > .bazelversion

运行 bazel version 下载安装对应的 bazel

直接安装对应版本的 bazel

参考 Installing Bazel on Ubuntu

安装 python 和 pip 包

编译 TensorFlow r1.15 版本本人使用的 python 版本是 3.6

安装 pip 包，参考 Issue，注意安装正确的版本，不然编译会出现各种问题。

Python 3.6.9

TF-1.14:
pip install 'numpy<1.19.0'
pip install -U pip six wheel mock
pip install future
pip install keras_applications==1.0.4 --no-deps
pip install keras_preprocessing==1.0.2 --no-deps
pip install pandas

bazel version: bazel-0.25.2-installer-linux-x86_64.sh

TF-1.15 & TF-2.0:
pip install 'numpy<1.19.0'
pip install -U pip six wheel mock
pip install future
pip install keras_applications==1.0.8 --no-deps
pip install keras_preprocessing==1.1.2 --no-deps
pip install pandas

bazel version: bazel-0.26.1-installer-linux-x86_64.sh

编译 TensorFlow

配置环境

  • 使用虚拟环境，运行：python configure.py
  • 使用系统环境，运行：./configure

编译

  • bazel build //tensorflow/tools/pip_package:build_pip_package
  • debug信息编译时加上-c dbg

编译错误

  • 如果是"external/grpc/src/core/lib/gpr/log_linux.cc:43:13: error: ambiguating new declaration of 'long int gettid()'"，参考Issue，进行如下操作：
cd third_party
wget https://nomeroff.net.ua/tf/Rename-gettid-functions.patch
cd ..
vim tensorflow/workspace.bzl
# 找到tf_http_archive( name = "grpc",...
# 添加下面的patch_file一行
    tf_http_archive(
        name = "grpc",
        patch_file = clean_dep("//third_party:Rename-gettid-functions.patch"),
        sha256 = "67a6c26db56f345f7cee846e681db2c23f919eba46dd639b09462d1b6203d28c",
        strip_prefix = "grpc-4566c2a29ebec0835643b972eb99f4306c4234a3",
        system_build_file = clean_dep("//third_party/systemlibs:grpc.BUILD"),
        urls = [
            "https://storage.googleapis.com/mirror.tensorflow.org/github.com/grpc/grpc/archive/4566c2a29ebec0835643b972eb99f4306c4234a3.tar.gz",
            "https://github.com/grpc/grpc/archive/4566c2a29ebec0835643b972eb99f4306c4234a3.tar.gz",
        ],
    )
# 重新运行bazel

安装 TensorFlow

./bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
pip install /tmp/tensorflow_pkg/tensorflow-version-tags.whl