[WIP][2/N] Port 5 _composable distributed test to Intel GPU #159241
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/159241
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 9d14e4b with merge base 2dccff7 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@pytorchbot label "topic: not user facing" |
guangyey
reviewed
Jul 30, 2025
| @@ -58,6 +58,11 @@ | |||
| from torch.testing._internal.distributed.checkpoint_utils import with_temp_dir | |||
|
|
|||
|
|
|||
| device = torch.accelerator.current_accelerator() | |||
| device_module = torch.get_device_module(device) | |||
There was a problem hiding this comment.
Where is device_module used?
There was a problem hiding this comment.
Where is
device_moduleused?
will remove it.
guangyey
reviewed
Jul 30, 2025
| @@ -58,6 +58,11 @@ | |||
| from torch.testing._internal.distributed.checkpoint_utils import with_temp_dir | |||
|
|
|||
|
|
|||
| device = torch.accelerator.current_accelerator() | |||
There was a problem hiding this comment.
Suggested change
| device = torch.accelerator.current_accelerator() | |
| device_type = acc.type if (acc := torch.accelerator.current_accelerator()) else "cpu" |
6a5b8de to
4e028bf
Compare
4e028bf to
a055b23
Compare
|
@pytorchbot label "module: xpu" |
|
@pytorchbot label "triaged" |
guangyey
reviewed
Jul 30, 2025
| @@ -98,6 +98,8 @@ def _test_compile( | |||
| self.create_pg(device) | |||
| torch._dynamo.config.optimize_ddp = "python_reducer" | |||
| torch.manual_seed(123) | |||
| if device_type == "xpu": | |||
There was a problem hiding this comment.
If without this code change, which test case will fail due to the non-deterministic accuracy issue.
There was a problem hiding this comment.
test_compile_backward_only, test_compile_bf16, test_compile_fp16, test_compile_gpu, test_compile_gpu_ac, listed in intel/torch-xpu-ops#1668
If without this code change, which test case will fail due to the non-deterministic accuracy issue.
guangyey
reviewed
Jul 31, 2025
| for linear in model.linears: | ||
| fully_shard(linear) | ||
| fully_shard(model.linears) | ||
| replicate(model, device_id=torch.cuda.current_device()) | ||
| replicate(model, device_id=device_module.current_device()) |
There was a problem hiding this comment.
Suggested change
| replicate(model, device_id=device_module.current_device()) | |
| replicate(model, device_id=torch.accelerator.current_device_index()) |
guangyey
reviewed
Jul 31, 2025
| @@ -273,14 +283,14 @@ def forward(self, x: torch.Tensor): | |||
| return y | |||
|
|
|||
| self._init_pg() | |||
| torch.cuda.set_device(self.rank) | |||
| device_module.set_device(self.rank) | |||
There was a problem hiding this comment.
Suggested change
| device_module.set_device(self.rank) | |
| torch.accelerator.set_device_index(self.rank) |
guangyey
reviewed
Jul 31, 2025
| # DDP instance is attached in first pre forward | ||
| model_cuda(torch.randn(2, 2)) | ||
| replicate_ddp_weakref = replicate.state(model_cuda)._ddp_weakref() | ||
| self.assertEqual([0], replicate_ddp_weakref.device_ids) | ||
| # Pass in int as device_id | ||
| replicate(model_cuda2, device_id=int(torch.cuda.current_device())) | ||
| replicate(model_cuda2, device_id=int(device_module.current_device())) |
guangyey
reviewed
Jul 31, 2025
| @@ -233,13 +242,13 @@ def test_replicate_device_id(self): | |||
| # Should be None for CPU training | |||
| self.assertEqual(None, replicate_ddp_weakref.device_ids) | |||
|
|
|||
| replicate(model_cuda, device_id=torch.device(torch.cuda.current_device())) | |||
| replicate(model_cuda, device_id=torch.device(device_module.current_device())) | |||
guangyey
reviewed
Jul 31, 2025
| def test_replicate_ignore_module(self): | ||
| self._init_pg() | ||
| torch.cuda.set_device(self.rank) | ||
| device_module.set_device(self.rank) |
guangyey
reviewed
Jul 31, 2025
| torch.cuda.set_device(self.rank) | ||
| model = MyNet().cuda() | ||
| replicate(model, device_id=torch.cuda.current_device()) | ||
| device_module.set_device(self.rank) |
guangyey
reviewed
Jul 31, 2025
| replicate(model, device_id=torch.cuda.current_device()) | ||
| device_module.set_device(self.rank) | ||
| model = MyNet().to(device) | ||
| replicate(model, device_id=device_module.current_device()) |
guangyey
reviewed
Jul 31, 2025
Thanks, will modify it. |
a055b23 to
9d14e4b
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.
For #114850, we will port distributed tests to Intel GPU. This is the second PR for _composable cases, the first is #159118.
We could enable Intel GPU with following methods and try the best to keep the original code styles:
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @pragupta @gujinghui @EikanWang @fengyuan14 @guangyey