深度学习之pytorch编译

在训练自己的数据集的时候会遇到各种各样的问题，其中pytorch编译就是一个很麻烦的问题。

从github下载的zip文件里面不包含git库，只能在终端使用git clone下载，网络不稳定就会报很多错误。

我在下载的时候有四个地方没有克隆成功：

1. 正克隆到 '/home/**/pytorch/third_party/FP16'... fatal: unable to access 'https://github.com/Maratyszcza/FP16.git/': gnutls_handshake() failed: The TLS connection was non-properly terminated. fatal: 无法克隆 'https://github.com/Maratyszcza/FP16.git' 到子模组路径 '/home/**/pytorch/third_party/FP16' Failed to clone 'third_party/FP16'. Retry scheduled
2. 正克隆到 '/home/**/pytorch/third_party/FXdiv'... remote: Enumerating objects: 275, done. remote: Counting objects: 100% (53/53), done. remote: Compressing objects: 100% (27/27), done. error: RPC failed; curl 56 GnuTLS recv error (-54): Error in the pull function. fatal: The remote end hung up unexpectedly fatal: 过早的文件结束符（EOF） fatal: index-pack failed fatal: 无法克隆 'https://github.com/Maratyszcza/FXdiv.git' 到子模组路径 '/home/**/pytorch/third_party/FXdiv' Failed to clone 'third_party/FXdiv'. Retry scheduled
3. 正克隆到 '/home/***pytorch/third_party/fbgemm'... fatal: unable to access 'https://github.com/pytorch/fbgemm/': GnuTLS recv error (-54): Error in the pull function. fatal: 无法克隆 'https://github.com/pytorch/fbgemm' 到子模组路径 '/home/**/pytorch/third_party/fbgemm' Failed to clone 'third_party/fbgemm'. Retry scheduled
4. 正克隆到 '/home/**pytorch/third_party/kineto/libkineto/third_party/dynolog/third_party/cpr'... fatal: unable to access 'https://github.com/libcpr/cpr.git/': Operation timed out after 0 milliseconds with 0 out of 0 bytes received fatal: 无法克隆 'https://github.com/libcpr/cpr.git' 到子模组路径'/hom**/pytorch/thire/d_party/kineto/libkineto/third_party/dynolog/third_party/cpr' Failed to clone 'third_party/cpr'. Retry scheduled

以下是具体步骤：

1.单独克隆失败的子模块：

对于每个失败的子模块，手动克隆并放置到相应的目录中。例如，对于 FP16 子模块：

git clone https://github.com/Maratyszcza/FP16.git third_party/FP16

git clone https://github.com/Maratyszcza/FXdiv.git third_party/FXdiv git clone https://github.com/pytorch/fbgemm.git third_party/fbgemm git clone https://github.com/libcpr/cpr.git third_party/kineto/libkineto/third_party/dynolog/third_party/cpr

• 运行问题：目标路径 'third_party/FP16' 已经存在，并且不是一个空目录。

方法 1：删除目录并重新克隆

1. 删除现有目录：

   rm -rf FP16

2. 重新克隆：

   git clone https://github.com/Maratyszcza/FP16.git FP16

其余子模块也是一样的

rm -rf FXdiv

git clone https://github.com/Maratyszcza/FXdiv.git FXdiv

rm -rf fbgemm

git clone https://github.com/pytorch/fbgemm.git fbgemm

rm -rf kineto/libkineto/third_party/dynolog/third_party/cpr

git clone https://github.com/libcpr/cpr.git kineto/libkineto/third_part

方法 2：进入已有目录并继续克隆

1. 进入目录：

   cd third_party/FP16

2. 检查是否已初始化为 git 仓库：

   git status

2.检查并更新子模块：

• 在克隆完成后，进入 pytorch 主目录，并更新子模块：

cd /home/**/pytorch
git submodule update --init --recursive

• 如果网络不稳定，增加 Git 的超时时间可能有助于防止连接超时：

git config --global http.lowSpeedLimit 0
git config --global http.lowSpeedTime 999999