Always try to preload pypi cuda deps. #133963
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/133963
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit c4188d5 with merge base 24b695a ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
oraluben
force-pushed
the
preload-pypi-if-possible
branch
from
409bdf6 to
02f2270
Compare
soulitzer
added
the
triaged
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
|
ping @malfet |
|
@pytorchbot label "topic: not user facing" |
oraluben
force-pushed
the
preload-pypi-if-possible
branch
from
6112c47 to
daaadd0
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #101314
Fixes #121207
Could detect #138324 before release
Fixes the import error when using redhat-like system with a system-wide cuda runtime.
Redhat-like have seperate platlib and purelib dir and breaks the rpath search, while if a system-wide cuda runtime presents, importing
libtorch_global_deps.sowill not fail because it only depends onlibcudart.soandlibnvToolsExt.soso the pypi path is not triggered, then it will fail likely because of a version mismatch of libcudnn and libnccl.This PR always try to preload pypi nvidia librarys if anysite-packages/nvidiapresents, and fixes the issue.Risk: unexpected (e.g. partiacally installed pypi nvidia packages)site-packages/nvidiafolder present will lead to import failure. This PR also change the behavior that all paths insys.pathwill be walked through, searching for/nvidia/<libs>. It seems unnecessary to me, but I do see some references from old issue e.g. #92096. Now it only searched forsite-packages/nvidia. (Debian might need to take care of thedist-packagesdir)We use python's import mechanism to detect and find the pypi packages, this can handle various case, including purelib/platlib, and different install location from easy install.
@malfet @atalman