ARM linux CI jobs fail when using torch 2.9 rc wheels
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 5k
- Forks
- 1.2k
- Avg merge
- 2d 10h
- Merged PRs (30d)
- 581
Description
🐛 Describe the bug
ARM Linux CI jobs are failing on the release/1.0 branch when using prebuilt PyTorch + domain packages. This appears to be due to the PyTorch packages requiring glibcxx 3.4.30. The existing ARM Linux images we use have 3.4.29.
This is a known problem with Conda on Ubuntu 22.04 (see https://github.com/pytorch/pytorch/issues/105248) and there is a workaround in install_conda.sh:
https://github.com/pytorch/executorch/blob/181ed4dd2adffaeaf158da5bfb9192a0e6199f1f/.ci/docker/common/install_conda.sh#L59-L62
However, this workaround is not used on ARM Linux because it apparently breaks the CI jobs (introduced in https://github.com/pytorch/executorch/pull/12068):
https://github.com/pytorch/executorch/blob/181ed4dd2adffaeaf158da5bfb9192a0e6199f1f/.ci/docker/common/install_conda.sh#L74-L78
ImportError: /opt/conda/envs/py_3.10/bin/../lib/libstdc++.so.6: version `GLIBCXX_3.4.30' not found (required by /opt/conda/envs/py_3.10/lib/python3.10/site-packages/executorch/extension/pybindings/_portable_lib.cpython-310-aarch64-linux-gnu.so)
From https://github.com/pytorch/executorch/actions/runs/18111201043/job/51550387476
We need to figure out what the proper fix is, perhaps updating the libcxx version in the docker image build or somewhere else? As a short-term fix, I'm installing a newer version with Conda in https://github.com/pytorch/executorch/pull/14677.
Versions
release/1.0 - commit 3854f4b36
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with .ci/docker/common/install_conda.sh, comparing the workaround at lines 59-62 with the ARM Linux path at lines 74-78. Review the referenced GitHub Actions job and the short-term approach in pull request 14677 to identify a fix that does not break ARM CI. Done means ARM Linux jobs on release/1.0 can install and import the torch 2.9 RC wheels successfully.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- anaconda, docker, python, pytorch
- Domain
- ci-cd, devops, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 35/100