Installing torch 1.7.1 installs torchvision 0.2.2 instead of 0.8.2

[christian-cahig]

🐛 Bug

When I install torch via conda (as instructed here), torchvision 0.2.2 is installed instead of torchvision 0.8.2.

To Reproduce

I ran the following command:

conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch

Expected behavior

From this table of corresponding versions of torch and torchvision, I assumed that installing torch 1.7.1 will install torchvision 0.8.2.

Environment

• PyTorch / torchvision Version: N/A (I am yet to install)
• OS: Microsoft Windows 10 Home
• How you installed PyTorch / torchvision (conda, pip, source): conda
• Python version: 3.8 (64-bit runtime)
• CUDA/cuDNN version: 11.0.221
• GPU models and configuration: GeForce RTX 2060

cc @peterjc123 @nbcsm @guyang3532 @maxluk @gunandrose4u @smartcat2010 @mszhanyi

[pmeier]

This usually happens if you are on a machine that we do not provide binaries for. See #2622 for more details.

[vfdev-5]

@christian-cahig thanks for reporting. I could not repro that on a windows machine

logs

conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch
Collecting package metadata: ...working... done
Solving environment: ...working... done

## Package Plan ##

  environment location: C:\tools\miniconda3\envs\test

  added / updated specs:
    - cudatoolkit=11.0
    - pytorch
    - torchaudio
    - torchvision


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    blas-1.0                   |              mkl           6 KB
    ca-certificates-2021.1.19  |       haa95532_0         160 KB
    certifi-2020.12.5          |   py38haa95532_0         144 KB
    cudatoolkit-11.0.221       |       h74a9793_0       941.9 MB
    freetype-2.10.4            |       hd328e21_0         490 KB
    intel-openmp-2020.2        |              254         2.0 MB
    jpeg-9b                    |       hb83a4c4_2         313 KB
    libpng-1.6.37              |       h2a8f88b_0         598 KB
    libtiff-4.1.0              |       h56a325e_1         1.1 MB
    libuv-1.40.0               |       he774522_0         332 KB
    lz4-c-1.9.3                |       h2bbff1b_0         140 KB
    mkl-2020.2                 |              256       170.7 MB
    mkl-service-2.3.0          |   py38h196d8e1_0          53 KB
    mkl_fft-1.2.1              |   py38h46781fe_0         139 KB
    mkl_random-1.1.1           |   py38h47e9c7a_0         288 KB
    ninja-1.10.2               |   py38h6d14046_0         273 KB
    numpy-1.19.2               |   py38hadc3359_0          21 KB
    numpy-base-1.19.2          |   py38ha3acd2a_0         4.9 MB
    olefile-0.46               |             py_0          33 KB
    openssl-1.1.1j             |       h2bbff1b_0         5.8 MB
    pillow-8.1.0               |   py38h4fa10fc_0         739 KB
    pip-21.0.1                 |   py38haa95532_0         2.0 MB
    python-3.8.5               |       h5fd99cc_1        18.7 MB
    pytorch-1.7.1              |py3.8_cuda110_cudnn8_0      1007.0 MB  pytorch
    setuptools-52.0.0          |   py38haa95532_0         936 KB
    six-1.15.0                 |     pyhd3eb1b0_0          13 KB
    sqlite-3.33.0              |       h2a8f88b_0         1.3 MB
    tk-8.6.10                  |       he774522_0         3.2 MB
    torchaudio-0.7.2           |             py38         2.7 MB  pytorch
    torchvision-0.8.2          |       py38_cu110         7.3 MB  pytorch
    typing_extensions-3.7.4.3  |     pyha847dfd_0          25 KB
    vc-14.2                    |       h21ff451_1           8 KB
    vs2015_runtime-14.27.29016 |       h5e58377_2         2.2 MB
    wheel-0.36.2               |     pyhd3eb1b0_0          31 KB
    wincertstore-0.2           |           py38_0          15 KB
    xz-5.2.5                   |       h62dcd97_0         333 KB
    zlib-1.2.11                |       h62dcd97_4         132 KB
    zstd-1.4.5                 |       h04227a9_0         837 KB
    ------------------------------------------------------------
                                           Total:        2.13 GB

uname -a
> MSYS_NT-10.0-17763

If you already installed cudatoolkit or other versions of pytorch, try to remove everything such that your env is clean. If it does not work, try to use pip

pip install numpy
pip install --pre torch torchvision torchaudio -f https://download.pytorch.org/whl/nightly/cu110/torch_nightly.html

HTH

[christian-cahig]

I was able to "fix" this after restarting my PC. I'll close this issue now. Thanks, @pmeier and @vfdev-5,