Cannot pip install torchvision==0.8.0+cu110

[rsdmse]

🐛 Bug

Cannot pip install torchvision==0.8.0+cu110

To Reproduce

Steps to reproduce the behavior:

1. Copy command from pytorch website.
2. pip install torch==1.7.0+cu110 torchvision==0.8.0+cu110 torchaudio===0.7.0 -f https://download.pytorch.org/whl/torch_stable.html
3. Error message as follows

ERROR: Could not find a version that satisfies the requirement torchvision==0.8.0+cu110 (from versions: 0.1.6, 0.1.7, 0.1.8, 0.1.9, 0.2.0, 0.2.1, 0.2.2, 0.2.2.post2, 0.2.2.post3, 0.3.0, 0.4.0, 0.4.0+cpu, 0.4.0+cu92, 0.4.1, 0.4.1+cpu, 0.4.1+cu100, 0.4.1+cu92, 0.4.2, 0.4.2+cpu, 0.4.2+cu100, 0.4.2+cu92, 0.5.0, 0.5.0+cpu, 0.5.0+cu100, 0.5.0+cu92, 0.6.0, 0.6.0+cpu, 0.6.0+cu101, 0.6.0+cu92, 0.6.1, 0.6.1+cpu, 0.6.1+cu101, 0.6.1+cu92, 0.7.0, 0.7.0+cpu, 0.7.0+cu101, 0.7.0+cu92, 0.8.0)
ERROR: No matching distribution found for torchvision==0.8.0+cu110

Expected behavior

0.8.0+cu110 should be listed in https://download.pytorch.org/whl/torch_stable.html but it's not.

[seemethere]

Transferred this issue to pytorch/vision since it has to do with pytorch/vision

[seemethere]

Currently working on this, will have to issue a v0.8.1 to resolve this though.

[rsdmse]

Thanks for your help. Just wanted to check - would it be ok to install the default torchvision 0.8.0 with torch 1.7.0+cu110?
And is there a way to install http://download.pytorch.org/whl/cu110/torchvision-0.8.0-cp37-cp37m-linux_x86_64.whl which seems to be 0.8.0 with CUDA 11?

[seemethere]

Yes you can install with pip install -f https://download.pytorch.org/whl/cu110/torch_stable.html torch==1.7.0+cu110 torchvision==0.8.0

[seemethere]

This should be resolved now, you can use:

pip install torch==1.7.0+cu110 torchvision==0.8.1+cu110 torchaudio===0.7.0 -f https://download.pytorch.org/whl/torch_stable.html

[pkuppens]

@seemethere , has this problem reappeared??

It worked for me in the past, but today, no luck.

(I do get a permission denied when I try to go to the download link https://download.pytorch.org/ directly:)

Context:
Working on Windows with Visual Studio Code, Jupyter Notebook, on a Python 3.9.0 based virtualenv.
I import numpy 1.19.3 first (because of fmod bug in Windows), and then try both versions you mention:

!pip install numpy==1.19.3
!pip install torch===1.7.0+cu110 torchvision===0.8.1+cu110 torchaudio===0.7.0 -f https://download.pytorch.org/whl/torch_stable.html
!pip install -f https://download.pytorch.org/whl/cu110/torch_stable.html torch==1.7.0+cu110 torchvision==0.8.0

Results in:

Requirement already satisfied: numpy==1.19.3 in e:\cart-tech\unetcart\.venv\lib\site-packages (1.19.3)
Looking in links: https://download.pytorch.org/whl/torch_stable.html
ERROR: Could not find a version that satisfies the requirement torch===1.7.0+cu110 (from versions: 0.1.2, 0.1.2.post1, 0.1.2.post2)
ERROR: No matching distribution found for torch===1.7.0+cu110
Looking in links: https://download.pytorch.org/whl/cu110/torch_stable.html
ERROR: Could not find a version that satisfies the requirement torch==1.7.0+cu110 (from versions: 0.1.2, 0.1.2.post1, 0.1.2.post2)
ERROR: No matching distribution found for torch==1.7.0+cu110

I also tried from a terminal inside the virtualenv.

(Related question, I'm on CUDA 11.1, is that OK/backward compatible?)

[fmassa]

Actually, @pkuppens this is a duplicate of #3046

From @vfdev-5 comment there:

yes, it is related to Python 3.9. The binaries for Pytorch, Torchvision and others are only available for Python 3.6, 3.7 and 3.8.

As such, I'm closing as a duplicate. We will be adding Python 3.9 binaries for PyTorch / torchvision soon, and #3041 is a step towards this direction.

[HGamalElDin]

Yes you can install with pip install -f https://download.pytorch.org/whl/cu110/torch_stable.html torch==1.7.0+cu110 torchvision==0.8.0

please how to append this to the requirments.txt file? as it always raises (ERROR: Invalid requirement: 'torch==1.7.1+cu101 torchvision==0.8.2+cu101 torchaudio===0.7.2' (from line 12 of E:\requirements.txt)

[vfdev-5]

@HGamalElDin try something like

pip install -r requirements.txt -f https://download.pytorch.org/whl/cu110/torch_stable.html 

[HGamalElDin]

@HGamalElDin try something like

pip install -r requirements.txt -f https://download.pytorch.org/whl/cu110/torch_stable.html 

thank you, this works for me:
Adding in the file,
--find-links https://download.pytorch.org/whl/torch_stable.html
torch==1.7.1+cu101
torchvision==0.8.2+cu101
torchaudio===0.7.2

[kumar-akshay324]

If the torchvision version has been updated, can the website be updated to reflect these changes please - https://pytorch.org/get-started/previous-versions/#linux-and-windows-9

[Ynjxsjmh]

I use nvcc to check my CUDA compiler version

$ /usr/local/cuda/bin/nvcc --version

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2020 NVIDIA Corporation
Built on Wed_Jul_22_19:09:09_PDT_2020
Cuda compilation tools, release 11.0, V11.0.221
Build cuda_11.0_bu.TC445_37.28845127_0

Since it's 11.0 so I just search CUDA 11.0 in https://pytorch.org/get-started/previous-versions/ page, it gives 4 results, for v1.7.0, there are conda and pip installation commands which give the torchvision not found error, hope this can be fixed

$ conda install pytorch==1.7.0 torchvision==0.8.0 torchaudio==0.7.0 cudatoolkit=11.0 -c pytorch
$ pip install torch==1.7.0+cu110 torchvision==0.8.0+cu110 torchaudio==0.7.0 -f https://download.pytorch.org/whl/torch_stable.html

[hiepbk]

Yes you can install with pip install -f https://download.pytorch.org/whl/cu110/torch_stable.html torch==1.7.0+cu110 torchvision==0.8.0

please how to append this to the requirments.txt file? as it always raises (ERROR: Invalid requirement: 'torch==1.7.1+cu101 torchvision==0.8.2+cu101 torchaudio===0.7.2' (from line 12 of E:\requirements.txt)

You can add --find-links https://download.pytorch.org/whl/torch_stable.html in your requirement.txt