Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Try to build CUDA wheels #106

Open
wants to merge 237 commits into
base: main
Choose a base branch
from
Open

Try to build CUDA wheels #106

wants to merge 237 commits into from

Conversation

frostedoyster
Copy link
Collaborator

No description provided.

Copy link

github-actions bot commented Feb 13, 2024

Here is a pre-built version of the code in this pull request: wheels.zip, you can install it locally by unzipping wheels.zip and using pip to install the file matching your system

.github/workflows/build-wheels.yml Outdated Show resolved Hide resolved
.github/workflows/build-jax-wheels.yml Outdated Show resolved Hide resolved
.github/workflows/build-torch-wheels.yml Outdated Show resolved Hide resolved
scripts/manylinux2014_aarch64/Dockerfile Outdated Show resolved Hide resolved
scripts/manylinux2014_aarch64/remove_unused_python.sh Outdated Show resolved Hide resolved

#RUN pip install torch==2.4.1+cu124 --extra-index-url https://download.pytorch.org/whl/cu124
#RUN pip install numpy cmake
RUN pip install torch==${PYTORCH_VERSION}+cu${CUDA_VER_NO_DOT} --extra-index-url https://download.pytorch.org/whl/cu${CUDA_VER_NO_DOT}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sphericart-jax/setup.py Outdated Show resolved Hide resolved
@nickjbrowning
Copy link
Collaborator

nickjbrowning commented Dec 18, 2024

@Luthaf

#106 (comment)

what do you mean regarding this? We need the torch cuda wheels specifically because we need access to the "c10/cuda" namespace, which doesn't seem to be available in cpu-torch, which is what gets pulled for build if we don't specify pip_extra_index_url.

@Luthaf
Copy link
Contributor

Luthaf commented Dec 18, 2024

which doesn't seem to be available in cpu-torch, which is what gets pulled for build if we don't specify pip_extra_index_url.

So that's the bit I don't understand. As far as I know, the linux-x64 wheels on PyPI contain all of libtorch_cuda, and we should be able to build against them. I would prefer it if all the things we put on PyPI are built against things that come from PyPI, since different torch build can have different ABI.

PyPI wheels might not contain anything CUDA related for linux-arm64 or windows-x64, and we will have to work around this, but for linux-x64 I'd like to build against PyPI wheel.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants