I just installed torch via pip in a virtualenv.
(a simple pip install torch, using python 3.10, in a linux with cuda 12.6)
I checked the total venv disk size before and after the pip install torch.
Before it was ocupying around 600 Mb
After, it ocupies new 5.7Gb
So, all installation files get a total of 5Gb!
Probably there is plenty of cuda stuff too, but I already have all drivers before. It is really needed to download all that stuff?
There are other ways to get a smaller install while keep cuda support?
No, the PyTorch binaries with CUDA support should use around 2.3GB and will install the CUDA runtime dependencies from PyPI, which other packages can reuse.
Maybe you’ve installed other lib as well, torch along without torchtext is 2.3GB I think, like what @ptrblck said, I think for torchtext along it takes 1.1GB? Just leave enough space for your env, if you are new in ML or DL field, try to split one env as ‘one for all’ environment and depend on case to use which env. If you are working on linux, try docker, much easier for building a env and deploy.