DISABLED test_scaled_mm (__main__.DistMatrixOpsTestWithLocalTensor)
- Dominant language
- Python
- Stars
- 103k
- Forks
- 29.5k
- PR merge metrics
- PR metrics pending
Description
Platforms: linux
Disabling the DTensor `test_scaled_mm` test: it fails consistently on the B200 distributed job (`linux.dgx.b200.8`) with `RuntimeError: Invalid scaling configuration` from the `_scaled_mm` meta (`torch/_meta_registrations.py:7203`), on the test's sub-row-wise sharded-scale config.
Note: the failure is B200 (sm100) specific, but the disable-bot has no per-GPU-arch platform, so this scopes to `linux`. Tracking / root-cause discussion: #190086. Re-enable once that is resolved.
cc @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @pragupta @msaroufim @dcci @aditvenk @weifengpy @malfet @pytorch/pytorch-dev-infra @tianyu-l @XilunWu @SherlockNoMad @ppwwyyxx
Contributor guide
Research direction
Start with the DTensor test named __main__.DistMatrixOpsTestWithLocalTensor.test_scaled_mm and inspect the failure in torch/_meta_registrations.py:7203. Read tracking issue #190086 for the root-cause discussion and reproduce on linux.dgx.b200.8 if available. Done means the invalid scaling configuration is resolved and the test can be re-enabled on linux.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, testing
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 30/100