Load cuda deps more aggressively #137059
Load cuda deps more aggressively #137059
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/137059
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit 7429bbb with merge base a6210fd ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
|
This PR needs a
|
36aaa2e to
31cfdfd
Compare
|
Updated with another case, s 8000 ince sometimes the error doesn't include the missing lib name: |
31cfdfd to
dd1b62d
Compare
dd1b62d to
c9d5142
Compare
778b01a to
78f0d40
Compare
|
@eqy can we get this merged in? or can you help find other reviewers? |
|
+1 |
9ed8bf8 to
c5be05c
Compare
c5be05c to
2a9c685
Compare
|
@keith what's the test plan for this change? |
|
similar to #149808 this is fixing something that isn't tested in CI. What's the recommendation for that case? I can easily whip up a bazel project that reproduces the issue, but I imagine that wouldn't be helpful besides a 1 time verification? FWIW we've been using this patch in prod since I submitted this change. |
Previously cuda deps were only loaded if loading the globals library failed with a cuda shared lib related error. It's possible the globals library to load successfully, but then for the torch native libraries to fail with a cuda error. This now handles both of these cases. Fixes pytorch#117350
2a9c685 to
7429bbb
Compare
|
My company also uses this patch and it works well to get pytorch functioning with a bazel hermetic python system. Wanted to chime in here incase that helps. |
|
@malfet any advice here 🙏🏻 |
Previously cuda deps were only loaded if loading the globals library
failed with a cuda shared lib related error. It's possible the globals
library to load successfully, but then for the torch native libraries to
fail with a cuda error. This now handles both of these cases.
Fixes #117350
cc @malfet @seemethere @ptrblck @msaroufim