Pytorch安装的一些问题

Pytorch（conda）安装

Pytorch安装（conda）的一些问题

问题：
最近安装了一下pytorch，前面安装了tensorflow没有遇到什么问题，就是直接在pycharm中搜索tensorflow-gpu版本并安装，gpu可用。但是照猫画虎的安装pytorch之后torch可用，但是Gpu不可用，torch.cuda.is_available()返回False。

尝试解决方法
1.更新Cuda，从此处下载cuda版本：下载地址参考文章：参考。一般情况参考中遇到的问题你都要解决，包括visual studio安装失败，nsight xxx安装失败等。安装目录最好不建议默认（刚开始没敢改结果少了几G）
仍存在问题：安装了10.1版本参考他人不同查看cuda方法看淡方法不同：

nvcc -V
结果：
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:12:52_Pacific_Daylight_Time_2019
Cuda compilation tools, release 10.1, V10.1.243

nvidia-smi
结果：
NVIDIA-SMI 442.59       Driver Version: 442.59       CUDA Version: 10.2

最终并没有处理这个问题，（可能这两者反应的不是同一个问题，根本没有关联？）。
2. 更新Gpu 驱动，此处就不添加参考了。
3. 下载cudnn，下载地址：下载地址，将压缩包内相应的文件放入文件夹：https://developer.nvidia.com/cudnn下相应文件夹内覆盖。

最终上述的方法并没有解决gpu不可用的问题。
查看gpu是否支持：官网，最终发现本人1050ti没有在支持列表内，但是（tf可用？）这也在安装cuda的时候有体现，会warning，最终确认可以用。

常用方法：完成上述的gpu准备后，在官网选择自己自己相应的版本命令安装，（去掉“-c pytorch”命令参数，不然使用官网源速度慢）。但是我使用上述方法gpu仍不可用。

怀疑使用的torch是cpu版本（但是自己确实使用的非cpu版本安装命令），这个部分博客也说了这个问题，window下官网方法下载的是cpu版本，于时使用conda remove卸载原始的安装包，发现确实是cpu版本的。
卸载详情：

## Package Plan ##

  environment location: E:\anaconda

  removed specs:
    - pytorch


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    intel-openmp-2020.0        |              166         1.5 MB  https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
    mkl-2020.0                 |              166        98.9 MB  https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
    ------------------------------------------------------------
                                           Total:       100.5 MB

The following packages will be REMOVED:

  _pytorch_select-1.1.0-cpu
  ninja-1.7.2-0
  pytorch-1.0.1-cpu_py37h39a92a0_0
  torchvision-0.2.1-py_2

在此过程中更改了conda channel，再使用官网方法安装报错。

Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.

PackagesNotFoundError: The following packages are not available from current channels:

  - torchvision

Current channels:

  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/win-64
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/noarch
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/noarch
  - https://repo.anaconda.com/pkgs/main/win-64
  - https://repo.anaconda.com/pkgs/main/noarch
  - http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/win-64
  - http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/noarch

To search for alternate channels that may provide the conda package you're
looking for, navigate to

    https://anaconda.org

and use the search bar at the top of the page.

解决方法：参考
使用命令查找可用的torch

anaconda search -t conda torch

结果返回了几百行，win下可用的不多，win下gpu可用的更少，最终选择：

     pytorch/pytorch           |    1.4.0 | conda           | linux-64, osx-64, win-64 | py2.7_2, py27_cuda7.5.18_cudnn6.0.21hc114ab0_4, py2.7_0, py2.7_1, py3.5_cuda10.0.130_cudnn7.6.2_0, py35_cuda0.0_cudnn0.0_1, py3.7_cuda101_cudnn7_0, py35_cuda90_cudnn7he774522_1, py35_cuda0.0_cudnn0.0_2, py36_cuda9.0.176_cudnn7.0.3hdc18817_4, py3.7_cuda9.0.176_cudnn7.4.2_0, py3.7_cuda9.0.176_cudnn7.4.2_2, py37_cuda80_cudnn7he774522_1, py3.7_cuda9.2.148_cudnn7.6.2_0, py3.6_cuda9.0.176_cudnn7.5.1_0, py3.5_cuda9.0.176_cudnn7.4.2_2, py3.5_cuda10.0.130_cudnn7.4.1_1, py3.5_cuda9.0.176_cudnn7.4.2_0, py35_cuda9.0.176_cudnn7.0.3h5a7d906_4, py3.6_cuda9.0.176_cudnn7.4.2_2, py3.6_cuda9.0.176_cudnn7.4.2_0, py27_cuda8.0.61_cudnn7.0.3hb4df5cf_4, py3.5_cuda9.0.176_cudnn7.5.1_0, py3.6_cuda10.0.130_cudnn7.5.1_0, py2.7_cuda9.2.148_cudnn7.6.2_0, py3.6_0, py36_cuda9.1.85_cudnn7.0.5_nccl2_2, py3.8_cpu_0, py36_cuda9.1.85_cudnn7.1.2_1, py2.7_cuda8.0.61_cudnn7.1.2_0, py2.7_cuda10.1.243_cudnn7.6.3_0, py3.5_cuda100_cudnn7_1, py3.7_cuda10.0.130_cudnn7.6.2_0, py3.5_cuda8.0.61_cudnn7.1.2_2, py3.5_cuda8.0.61_cudnn7.1.2_1, py3.5_cuda8.0.61_cudnn7.1.2_0, py3.6_cuda9.0.176_cudnn7.4.1_1, py36_cuda9.1.85_cudnn7.0.5_2, py3.6_2, py3.8_cuda10.0.130_cudnn7.6.3_0, py36_cuda0.0_cudnn0.0h57b1bc9_4, py3.6_1, py3.6_cuda10.0.130_cudnn7.4.1_1, py27_cuda9.0.176_cudnn7.0.5_nccl2_2, py3.7_cuda10.0.130_cudnn7.4.2_2, py3.7_cuda10.0.130_cudnn7.4.2_0, py27_cuda0.0_cudnn0.0_2, py2.7_cuda10.0.130_cudnn7.6.2_0, py3.5_cuda10.0.130_cudnn7.5.1_0, py36_cuda9.0.176_cudnn7.0.5_nccl2_2, py3.7_cpu_0, py27_cuda9.0.176_cudnn7.1.2_1, py35_cuda8.0.61_cudnn7.0.3h4d8fc25_4, py3.7_cuda10.0.130_cudnn7.4.1_1, py3.5_cuda9.2.148_cudnn7.6.3_0, py36_cuda7.5.18_cudnn6.0.21h759af52_4, py3.6_cuda101_cudnn7_0, py35_cuda7.5.18_cudnn6.0.21h2ca90fe_4, py35_cuda9.0.176_cudnn7.0.5_nccl2_2, py2.7_cuda10.0.130_cudnn7.4.1_1, py35_cuda9.1.85_cudnn7.0.5_nccl2_2, py36_cuda9.0.176_cudnn7.1.2_1, py37_cuda9.2.148_cudnn7.1.4_1, py2.7_cuda10.0.130_cudnn7.5.1_0, py3.7_cuda8.0.61_cudnn7.1.2_2, py2.7_cuda9.0.176_cudnn7.4.1_1, py2.7_cuda10.0.130_cudnn7.4.2_2, py2.7_cuda10.0.130_cudnn7.4.2_0, py36_cuda8.0.61_cudnn7.0.5_2, py3.5_cuda9.0.176_cudnn7.4.1_1, py3.7_cpu_1, py3.7_cuda8.0.61_cudnn7.1.2_1, py3.7_cuda8.0.61_cudnn7.1.2_0, py3.7_cuda90_cudnn7_1, py2.7_cuda10.0.130_cudnn7.6.3_0, py27_cuda8.0.61_cudnn7.0.3hf383a3f_4, py3.5_cpu_1, py37_cuda90_cudnn7he774522_1, py3.5_cuda10.1.243_cudnn7.6.3_0, py35_cuda80_cudnn7he774522_1, py3.7_1, py3.7_0, py2.7_cpu_0, py3.6_cpu_1, py3.6_cpu_0, py27_cuda0.0_cudnn0.0he480db7_4, py27_cuda9.0.176_cudnn7.0.5_2, py3.6_cuda10.0.130_cudnn7.6.3_0, py3.6_cuda80_cudnn7_1, py27_cuda9.1.85_cudnn7.0.5_nccl2_2, py27_cuda9.0.176_cudnn7.0.3_nccl2h301e181_4, py3.7_cuda10.0.130_cudnn7.5.1_0, py3.6_cuda10.0.130_cudnn7.4.2_2, py3.6_cuda10.0.130_cudnn7.4.2_0, py27_cuda8.0.61_cudnn7.1.2_1, py3.6_cuda10.0.130_cudnn7.6.2_0, py36_cuda80_cudnn7he774522_1, py3.7_2, py35_py27__9.0.176_7.1.2_2, py3.6_cuda9.2.148_cudnn7.6.3_0, py36_cuda8.0.61_cudnn7.0.3h37a80b5_4, py35_cuda9.1.85_cudnn7.0.5_2, py3.7_cuda92_cudnn7_1, py3.7_cuda9.0.176_cudnn7.4.1_1, py3.7_cuda92_cudnn7_0, py27_cuda8.0.61_cudnn7.0.5_2, py27_cuda9.2.148_cudnn7.1.4_1, py3.6_cuda90_cudnn7_1, py27_cuda9.1.85_cudnn7.1.2_1, py2.7_cuda9.0.176_cudnn7.4.2_2, py2.7_cuda9.0.176_cudnn7.4.2_0, py3.5_cuda80_cudnn7_1, py3.8_cuda101_cudnn7_0, py3.5_cuda10.0.130_cudnn7.4.2_0, py35_cuda0.0_cudnn0.0hc53adbe_4, py3.7_cuda10.0.130_cudnn7.6.3_0, py3.5_cuda9.2.148_cudnn7.6.2_0, py3.5_1, py3.5_0, py3.5_2, py2.7_cuda8.0.61_cudnn7.1.2_2, py3.8_0, py36_cuda90_cudnn7he774522_1, py3.6_cuda10.1.243_cudnn7.6.3_0, py3.6_cuda92_cudnn7_0, py2.7_cuda9.0.176_cudnn7.5.1_0, py3.8_cuda9.2.148_cudnn7.6.3_0, py37_cuda0.0_cudnn0.0_1, py36_cuda9.0.176_cudnn7.0.5_2, py3.7_cuda9.2.148_cudnn7.6.3_0, py35_cuda9.2.148_cudnn7.1.4_1, py35_cuda9.0.176_cudnn7.0.3_nccl2h5f42aa5_4, py3.5_cpu_0, py36_cuda92_cudnn7he774522_1, py3.5_cuda101_cudnn7_0, py35_cuda8.0.61_cudnn7.0.3hb362f6e_4, py27__9.0.176_7.1.2_2, py36_cuda91_cudnn7he774522_1, py35_cuda8.0.61_cudnn7.0.5_2, py2.7_cuda8.0.61_cudnn7.1.2_1, py3.5_cuda10.0.130_cudnn7.4.2_2, py35_cuda91_cudnn7he774522_1, py3.6_cuda92_cudnn7_1, py36_py35_py27__9.0.176_7.1.2_2, py37_cuda92_cudnn7he774522_1, py3.7_cuda9.0.176_cudnn7.5.1_0, py27_cuda9.0.176_cudnn7.0.3hdbbd62b_4, py35_cuda9.0.176_cudnn7.0.5_2, py36_cuda9.0.176_cudnn7.0.3_nccl2h295ae03_4, py3.8_cuda92_cudnn7_0, py3.7_cuda80_cudnn7_1, py37_cuda8.0.61_cudnn7.1.2_1, py2.7_cuda9.2.148_cudnn7.6.3_0, py3.6_cuda9.2.148_cudnn7.6.2_0, py36_cuda9.2.148_cudnn7.1.4_1, py27_cuda0.0_cudnn0.0_1, py35_cuda9.1.85_cudnn7.1.2_1, py37_py36_py35_py27__9.0.176_7.1.2_2, py35_cuda92_cudnn7he774522_1, py3.6_cuda100_cudnn7_1, py3.5_cuda92_cudnn7_0, py3.5_cuda92_cudnn7_1, py3.7_cuda100_cudnn7_1, py35_cuda9.0.176_cudnn7.1.2_1, py3.6_cuda8.0.61_cudnn7.1.2_0, py3.6_cuda8.0.61_cudnn7.1.2_1, py3.6_cuda8.0.61_cudnn7.1.2_2, py3.7_cuda10.1.243_cudnn7.6.3_0, py36_cuda8.0.61_cudnn7.0.3hcf1d89b_4, py27_cuda9.1.85_cudnn7.0.5_2, py3.5_cuda10.0.130_cudnn7.6.3_0, py37_cuda9.0.176_cudnn7.1.2_1, py36_cuda8.0.61_cudnn7.1.2_3, py35_cuda8.0.61_cudnn7.1.2_1, py36_cuda8.0.61_cudnn7.1.2_1, py35_cuda8.0.61_cudnn7.1.2_3, py27_cuda8.0.61_cudnn7.1.2_3, py3.8_cuda10.1.243_cudnn7.6.3_0, py3.5_cuda90_cudnn7_1, py36_cuda0.0_cudnn0.0_1, py36_cuda0.0_cudnn0.0_2
                                          : PyTorch is an optimized tensor library for deep learning using GPUs and CPUs.

使用命令下载（死马当活马医，也不知道能不能成功）

 conda install -c https://conda.anaconda.org/pytorch pytorch

输出：

Collecting package metadata (current_repodata.json): done
Solving environment: -
The environment is inconsistent, please check the package plan carefully
The following packages are causing the inconsistency:

  - defaults/win-64::anaconda==custom=py37_1
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::astropy==4.0=py37he774522_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::bkcharts==0.2=py37_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::bokeh==2.0.0=py37_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::bottleneck==1.3.2=py37h2a96729_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/noarch::dask==2.12.0=py_0
  - defaults/win-64::gensim==3.8.0=py37hf9181ef_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::h5py==2.10.0=py37h5e291fa_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::imageio==2.6.1=py37_0
  - conda-forge/noarch::keras-applications==1.0.8=py_1
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/noarch::keras-preprocessing==1.1.0=py_1
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::matplotlib==3.1.3=py37_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::matplotlib-base==3.1.3=py37h64f37c6_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::mkl_fft==1.0.15=py37h14836fe_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::mkl_random==1.1.0=py37h675688f_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::numba==0.48.0=py37h47e9c7a_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::numexpr==2.7.1=py37h25d0782_0
  - defaults/win-64::numpy==1.16.5=py37h19fb1c0_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/noarch::opt_einsum==3.1.0=py_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::pandas==1.0.1=py37h47e9c7a_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::patsy==0.5.1=py37_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::pytables==3.5.2=py37h1da0976_1
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::pytest-arraydiff==0.3=py37h39e3cac_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/noarch::pytest-astropy==0.8.0=py_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/noarch::pytest-doctestplus==0.5.0=py_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::pywavelets==1.1.1=py37he774522_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::scikit-image==0.16.2=py37h47e9c7a_0
  - defaults/win-64::scikit-learn==0.21.3=py37h6288b17_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::scipy==1.4.1=py37h9439919_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/noarch::seaborn==0.10.0=py_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::statsmodels==0.11.0=py37he774522_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/noarch::tensorboard==2.1.0=py3_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::tensorflow==2.1.0=gpu_py37h7db9008_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::tensorflow-base==2.1.0=gpu_py37h55f5790_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/noarch::tensorflow-estimator==2.1.0=pyhd54b08b_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::tensorflow-gpu==2.1.0=h0d30ee6_0
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64::_anaconda_depends==2019.03=py37_0
done

## Package Plan ##

  environment location: E:\anaconda

  added / updated specs:
    - pytorch


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    numpy-1.16.5               |   py37h19fb1c0_0          49 KB  https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
    pytorch-1.4.0              |py3.7_cuda101_cudnn7_0       472.8 MB  pytorch
    scikit-learn-0.22.1        |   py37h6288b17_0         4.7 MB  https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
    ------------------------------------------------------------
                                           Total:       477.5 MB

The following NEW packages will be INSTALLED:

  ninja              anaconda/pkgs/free/win-64::ninja-1.7.2-0
  pytorch            pytorch/win-64::pytorch-1.4.0-py3.7_cuda101_cudnn7_0

The following packages will be UPDATED:

  scikit-learn       pkgs/main::scikit-learn-0.21.3-py37h6~ --> anaconda/pkgs/main::scikit-learn-0.22.1-py37h6288b17_0

The following packages will be SUPERSEDED by a higher-priority channel:

  numpy                                           pkgs/main --> anaconda/pkgs/main
Preparing transaction: done
Verifying transaction: done
Executing transaction: done

最终测试：

查看tf是否受影响：

问题
可能大多数开发者都在linux下吧，但是win下安装的也很少提到安装的只是cpu版本，因此一直以为是gpu环境问题，浪费了很长时间，也有很多安装的并没有测试gpu是否可用，也有再显卡设置里设置高性能显卡优先使用就解决了返回false的问题，但是确实我各种方法下自动给安装的都是cpu版本，所以最终的方法还是应该查找所有可用的pytorch版本，查看相应版本的环境，依据环境下载相应版本。