can't install latest version of torchvision using pip, even though I have the latest dependencies

[trevoravant]

🐛 Describe the bug

First of all, the issue I'm about to describe is similar to #4076, but I don't have opencv installed and I'm using pip. Also, I'm not sure if I should create this issue on torchvision or pip's Github page.

I'm trying to install torchvision with pip. I use Arch Linux, and I have a recent version of torch (1.10.2) installed through Arch's python-pytorch-cuda package. I also have python 3.10.2 installed. When I try to install torchvision using pip, I expect it to install the latest version: 0.11.1. However, it will only install version 0.2.2.post3.

$ pip install --user torchvision
Requirement already satisfied: torchvision in /home/trevor/.local/lib/python3.10/site-packages (0.2.2.post3)
Requirement already satisfied: torch in /usr/lib/python3.10/site-packages (from torchvision) (1.10.2)
Requirement already satisfied: six in /usr/lib/python3.10/site-packages (from torchvision) (1.16.0)
Requirement already satisfied: numpy in /home/trevor/.local/lib/python3.10/site-packages (from torchvision) (1.22.1)
Requirement already satisfied: pillow>=4.1.1 in /home/trevor/.local/lib/python3.10/site-packages (from torchvision) (9.0.0)
Requirement already satisfied: typing_extensions in /usr/lib/python3.10/site-packages (from torch->torchvision) (4.0.1)

I have also tested this on another machine without CUDA support (I installed torch through Arch's python-pytorch package) and I get the same result.

Versions

Collecting environment information...
PyTorch version: 1.10.2
Is debug build: False
CUDA used to build PyTorch: 11.5
ROCM used to build PyTorch: N/A

OS: Arch Linux (x86_64)
GCC version: (GCC) 11.1.0
Clang version: 13.0.0
CMake version: version 3.22.2
Libc version: glibc-2.33

Python version: 3.10.2 (main, Jan 15 2022, 19:56:27) [GCC 11.1.0] (64-bit runtime)
Python platform: Linux-5.16.3-arch1-1-x86_64-with-glibc2.33
Is CUDA available: True
CUDA runtime version: 11.5.119
GPU models and configuration: GPU 0: NVIDIA GeForce GTX 1080 Ti
Nvidia driver version: 495.46
cuDNN version: Probably one of the following:
/usr/lib/libcudnn.so.8.3.0
/usr/lib/libcudnn_adv_infer.so.8.3.0
/usr/lib/libcudnn_adv_train.so.8.3.0
/usr/lib/libcudnn_cnn_infer.so.8.3.0
/usr/lib/libcudnn_cnn_train.so.8.3.0
/usr/lib/libcudnn_ops_infer.so.8.3.0
/usr/lib/libcudnn_ops_train.so.8.3.0
HIP runtime version: N/A
MIOpen runtime version: N/A

Versions of relevant libraries:
[pip3] numpy==1.22.1
[pip3] torch==1.10.2
[pip3] torchvision==0.2.2.post3
[conda] Could not collect

[aryanraj2713]

add a requirement.txt file for to install libs.

[oke-aditya]

Aah this issue of torchvision 0.2.2 is very redundant. See #4273 #4665 #4648

Also CUDA 11.5 is not supported currently.

You can try with CUDA 11.3 as per instructions from docs https://pytorch.org/

pip3 install torch==1.10.2+cu113 torchvision==0.11.3+cu113 torchaudio==0.10.2+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

[trevoravant]

@oke-aditya Thanks but I don't think CUDA is the culprit because (as I mentioned in the original post) I get the same problem without CUDA. I've looked at the issues you linked to, but still can't make sense of this.

[NicolasHug]

@trevoravant could you please try using the official instructions at https://pytorch.org/get-started/locally/ ?

[pmeier]

I also have python 3.10.2 installed.

There are no stable binaries for Python 3.10 yet. However, there are nightlies available.

[trevoravant]

@NicolasHug @pmeier Thank you for the suggestions. I uninstalled the Arch Linux package that I used to install torch. I also uninstalled CUDA just to eliminate another variable. I tried to install both the stable and nightly CPU-only versions of torch and torchvision from pytorch.org. The stable version produces the error:

ERROR: Could not find a version that satisfies the requirement torch==1.10.2+cpu
ERROR: No matching distribution found for torch==1.10.2+cpu

the nightly version doesn't produce an error, but still only installs torchvision 0.2.2.post3.

So is this because I'm using Python 3.10.2?

[pmeier]

So is this because I'm using Python 3.10.2?

Yes. There is currently no version of PyTorch that supports Python 3.10. The next release in late February / early March will. I suggest you use something like pyenv or conda to create a virtual environment with Python < 3.10 and go from there.