I am new to PyTorch and wanted to install it on my Windows 10 64-bit workstation. I proceeded to use the online tool to generate the following command in my case:
After issuing this command, I obtained the following cryptic failure response:
Looking in indexes: …, https://download.pytorch.org/whl/cu117
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch
However, looking manually in the torch link list, I found the following distribution file that seems to fit the bill in my case:
torch-1.13.1+cu117-cp39-cp39-win_amd64.whl
Can anyone explain why this happened in my case? Along these lines, I have a question to ask: When issuing the pip3 install command above, is my workstation checked out with respect to components (CUDA, Python) and their versions (12.0.0 and 3.11.1) before the install takes place? And if so, when will PyTorch become compatible with my versions since I do want the latest for these general components?
Any assistance here for this beginner would be much appreciated…
You are pointing to a Python3.9 build in torch-1.13.1+cu117-cp39-cp39-win_amd64.whl while you are using Python3.11, which is not supported for Windows.
At least I cannot find any 1.13.1+cu117 build with cp311 for Windows in the linked URL.
No and yes. PyTorch will not use your locally installed CUDA toolkit unless you are building from source or a custom CUDA extension. You will only need a properly installed and working NVIDIA driver to install and execute the PyTorch binaries. Thus the pip install command doesn’t care which CUDA toolkit you are using and will install its own runtime dependencies.
However, the used Python version is important and pip will fail as seen in your case.
Thanks for the needed clarifying information (especially about what “cp39” means in the distribution file name)! I will switch to Python 3.9 that was part of my previous MS Visual Studio installation and try again (will need to change the Path environment variable first). I will let you know if I am still in trouble…
Could you try to download these wheels directly and install them locally via pip install ./torch-1.13.1....?
If this works it seems your pip environment has trouble finding packages and you might need to update pip itself. If this fails it could mean that your current environment is using e.g. another Python version.
Dear Patrick & Hari:
Just wanted to pop in here to let you know that I have successfully been able to install PyTorch on my Windows platform. The crux of the matter was the version of Python. I was trying to use the latest version (3.11.1) and this will not work; it might be the same for version 3.10 (?). What does work is version 3.9.7. I did have to do a direct download and manual install of the .whl files, but this was because of an unstable Internet connection that failed for the large .whl file containing torch (~2.2 GB). Hope this helps…
chaosdoc