Prebuilt wheels provided via 3rd party repository

[MiroPsota]

Prebuilt wheels for PyTorch packages with custom ops

I've created a repository that can build PyTorch wheels with custom ops through the GitHub Actions pipeline and publish them using GitHub Releases. Check it out at https://github.com/MiroPsota/torch_packages_builder.

Since there are various ways how to use it, please refer to the repository README for more information.

If you prefer own build or can't trust a 3rd party repository, feel free to fork it and build any package/version/commit ID you desire yourself.

• No Support for Pip Cache: pip relies on http cache, and GitHub generates on-the-fly redirections for release links, so they are probably not playing nicely together. I recommend hosting it yourself.

PyTorch3D specific info:

Install using pip:

pip install --extra-index-url https://miropsota.github.io/torch_packages_builder pytorch3d==<version>+<OPTIONAL_commit_hash>pt<PyTorch_version><compute_platform>

Where <compute_platform> is, as in PyTorch, one of cpu, cu<CUDA_short_version> (e.g. cu121, cu118, cu102), or rocm<ROCM_version> (not supported right now).

For example, the newest pytorch3d tag (as of writing) v0.7.8, PyTorch 2.6.0 with CUDA 12.6:

pip install --extra-index-url https://miropsota.github.io/torch_packages_builder pytorch3d==0.7.8+pt2.6.0cu126

Look at releases section if there are any other combinations, I will probably build occasionally with new pytorch releases and versions/commits.

These wheels are built with PyTorch versions 1.11.0 to 2.6.0 and their respective compute platforms and supported operating systems. Please note an exception for cu102 on Windows (due to no VS 2017 on the GitHub windows-2019 runner) and the ROCm platform. The build is done using the available public GitHub runners, so they might not work on older OSes.

The Windows build of pytorch3d==0.7.8+pt2.6.0cu126 is currently not possible.

Although the wheels have been successfully built, I don't test them to work correctly.

If you've installed PyTorch with pip, there's no need to have CUDA installed on your system, as the PyTorch wheels for pip bundle CUDA.

[yueool]

谢谢你，自己编译好几次都失败了，还是用现成的吧，我好笨，仰望大佬

[randprint]

can you check the build logs for windows 2.2.2 cu118? I tried to install from your repo to fix the dll import problem i'm having with torch3d but it seems the dll that should be in Lib\site-packages\pytorch3d is missing, if i understand the response to this issue that I'm also having #1664

[MiroPsota]

If I understand it correctly, Just do as the last post says.

[randprint]

tried that out of frustration, -e expects some file to be found on the url, but it isnt part of any pytorch3d install instructions I've seen anywhere anyway.

[Jacky-MYQ]

Wo! amazing!! I try it successfully! And my env is py3.11, cuda12.1, ubuntu22.04. I sincerely appreciate it that does me a favor!!

[zhanghua7099]

Thanks for your great work! This helped me a lot.

[shewangmu]

呜呜呜，好人一生平安

[FurkanGozukara]

so awesome gonna test now

pip install --extra-index-url https://miropsota.github.io/torch_packages_builder pytorch3d==0.7.8+pt2.5.1cu124

[brianlan]

Works for me, thanks. @MiroPsota MiroPsota

[linden713]

Thanks A LOT! You really have saved our life!

[axbycc-mark]

Great! This deserves support from the official team plus a mention in the installation guide.

[lly007]

Great work！
I suggest pytorch3d officially put this in the readme.

[UpendraArun]

Very helpful! This should go on the official install readme.

[rajk97]

Thank you very much!

[cake287]

You are sent from the heavens

[XHarold]

Perfectly solved my problem, great work!

[dolienky]

You are a GOD!

[Yicheng-Cao]

OMG, thank you so much for your contribution. It took me about one day to build the wheel, but it kept failing. You don't know how much this has helped me.

[tuqizhao]

@MiroPsota Can you provide the torch 2.6.0 cuda 12.6 python 3.12 on windows version， Although is not stable now. Tks

[MiroPsota]

Unfortunately, it's not possible at the moment, even with the main branch.

[fcbfcb1998]

Awesome