[BE][8/N] Remove ShardedTensor from TP FSDP integration test and other tests depending on Sharded Linear#96254
[BE][8/N] Remove ShardedTensor from TP FSDP integration test and other tests depending on Sharded Linear#96254fduwjj wants to merge 3 commits intogh/fduwjj/81/basefrom
Conversation
[ghstack-poisoned]
fduwjj
requested review from
H-Huang,
awgu,
fegin,
kwen2501,
mrshenli,
rohan-varma,
wanchaol and
zhaojuanmao
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/96254
Note: Links to docs will display an error until the docs builds have been completed. ❌ 3 FailuresAs of commit 32f1858: NEW FAILURES - The following jobs have failed:
FLAKY - The following jobs failed but were likely due to flakiness present on master:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
huydhn
added
the
ciflow/periodic
|
Looks like the failure about dynamo benchmark is not related to this PR "No CUDA GPUs are available" |
We removed ShardedLinear in #95948 but it broke TP_FSDP integration test because it is using ShardedTensor in the test. Migrating using DTensor fixes the test. DTensor shards the bias too so that we need to change the test a little bit. [ghstack-poisoned]
fduwjj
changed the title
|
The new periodic multigpu failure https://hud.pytorch.org/pr/96254#11842511229 also look related, so I guess it's another test to be update |
…st and other tests depending on Sharded Linear" We removed ShardedLinear in #95948 but it broke TP_FSDP integration test because it is using ShardedTensor in the test. Migrating using DTensor fixes the test. DTensor shards the bias too so that we need to change the test a little bit. [ghstack-poisoned]
|
@huydhn Ahhh that's no wonder why I didn't see it. We need to remove that test too. We have one already for DTensor under test/distributed/. Removed that test too. |
huydhn
left a comment
huydhn
left a comment
There was a problem hiding this comment.
LGTM! You might see buck failure on periodic, but it's broken in trunk at the moment. So the failure is expected.
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Merge failedReason: 1 jobs have failed, first few of them are: periodic / linux-bionic-cuda11.7-py3.9-gcc7 / test (multigpu, 1, 1, linux.16xlarge.nvidia.gpu) Details for Dev Infra teamRaised by workflow job |
|
@pytorchbot merge -f "failing tests are not related to this PR." |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
This test has been deleted in #96254 Pull Request resolved: #96431 Approved by: https://github.com/clee2000, https://github.com/fduwjj
…r tests depending on Sharded Linear (#96254) We removed ShardedLinear in pytorch/pytorch#95948 but it broke TP_FSDP integration test because it is using ShardedTensor in the test. Migrating using DTensor fixes the test. DTensor shards the bias too so that we need to change the test a little bit. Pull Request resolved: pytorch/pytorch#96254 Approved by: https://github.com/huydhn
…r tests depending on Sharded Linear (#96254) We removed ShardedLinear in pytorch/pytorch#95948 but it broke TP_FSDP integration test because it is using ShardedTensor in the test. Migrating using DTensor fixes the test. DTensor shards the bias too so that we need to change the test a little bit. Pull Request resolved: pytorch/pytorch#96254 Approved by: https://github.com/huydhn
…r tests depending on Sharded Linear (pytorch#96254) We removed ShardedLinear in pytorch#95948 but it broke TP_FSDP integration test because it is using ShardedTensor in the test. Migrating using DTensor fixes the test. DTensor shards the bias too so that we need to change the test a little bit. Pull Request resolved: pytorch#96254 Approved by: https://github.com/huydhn
3 participants
Stack from ghstack (oldest at bottom):
We removed ShardedLinear in #95948 but it broke TP_FSDP integration test because it is using ShardedTensor in the test. Migrating using DTensor fixes the test. DTensor shards the bias too so that we need to change the test a little bit.