Cannot install pytorch3d in virutal environment

[briandudeman]

❓How should I install pytorch3d in wsl without a CPU in a venv

A module I need uses pytorch3d, and I have been trying to install it, but when I run pip install "git+https://github.com/facebookresearch/pytorch3d.git@stable" I get the following error:

This is through wsl in a virtual environment.

Thank you for any help!

[msplr]

Try with pip install --no-build-isolation git+https://github.com/facebookresearch/pytorch3d.git@stable.

This makes your venv available for the build context.

[briandudeman]

Sorry for the delay, but when I run that command I just get this error:

preceded by a super long log of other stuff that I think is relating to that same error

[bottler]

These two errors are different, and the second is from a later stage than the first. I think the new error is either an actual build error somewhere - which would be higher up in the output - or an OOM. You could try running with PYTORCH3D_NO_NINJA=1 to disable parallel building, which should make the former problem easier to spot and relieve the latter problem.

[Luh1124]

I have the same problem, the torch version is 2.7.1, cuda is 12.6, torch.cuda.is_available() = True

[briandudeman]

Hi @bottler. After running with PYTORCH3D_NO_NINJA=1, I get this error message:

I have already installed torch using pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

[bottler]

Can you try adding --no-build-isolation to the command? Also it's probably easier to clone the repo locally first.

[oooolga]

I'm encountering a similar issue during installation. I've tried adding the --no-build-isolation flag, both with and without cloning the repository locally, but the installation consistently hangs at the wheel-building step. Here’s the relevant output:

Processing /storage/home/oooolga/Projects/pytorch3d
  Preparing metadata (pyproject.toml) ... done
Collecting iopath (from pytorch3d==0.7.8)
  Using cached iopath-0.1.10-py3-none-any.whl
Requirement already satisfied: tqdm in /storage/home/oooolga/micromamba/envs/da3/lib/python3.10/site-packages (from iopath->pytorch3d==0.7.8) (4.67.1)
Requirement already satisfied: typing_extensions in /storage/home/oooolga/micromamba/envs/da3/lib/python3.10/site-packages (from iopath->pytorch3d==0.7.8) (4.15.0)
Collecting portalocker (from iopath->pytorch3d==0.7.8)
  Using cached portalocker-3.2.0-py3-none-any.whl.metadata (8.7 kB)
  Using cached portalocker-3.2.0-py3-none-any.whl (22 kB)
Building wheels for collected packages: pytorch3d
  Building wheel for pytorch3d (pyproject.toml) ...

[bottler]

Hangs for how long? With how much memory usage? Building pytorch3d can take a long time. MAX_JOBS can be used to control the parallelism. And -v may let you see compiler output.

[oooolga]

Hi @bottler, you were right. Waiting a bit longer was what I needed. Thanks for your advice!