Register hpu device to fake backend (#156076)

amathewc · pytorchmergebot · commit 6835ba1b3409 · 2025-06-23T16:08:08.000Z
## MOTIVATION This PR intends to add hpu ( Intel Gaudi) also to the list of devices that will be supported by the "fake" distributed backend and the process group that will be created. ## CHANGES - Add "hpu" to the list of devices @ankurneog, @EikanWang Pull Request resolved: #156076 Approved by: https://github.com/d4l3k, https://github.com/EikanWang, https://github.com/albanD
diff --git a/torch/testing/_internal/distributed/fake_pg.py b/torch/testing/_internal/distributed/fake_pg.py
@@ -25,4 +25,4 @@ def _create_fake_pg(prefix_store, rank, world_size, timeout):
     return FakeProcessGroup(rank, world_size)
 
 
-dist.Backend.register_backend("fake", _create_fake_pg, devices=["cpu", "cuda"])
+dist.Backend.register_backend("fake", _create_fake_pg, devices=["cpu", "cuda", "hpu"])

Original file line number	Diff line number	Diff line change
`@@ -25,4 +25,4 @@ def _create_fake_pg(prefix_store, rank, world_size, timeout):`
`25`	`25`	`return FakeProcessGroup(rank, world_size)`
`26`	`26`
`27`	`27`
`28`		`-dist.Backend.register_backend("fake", _create_fake_pg, devices=["cpu", "cuda"])`
	`28`	`+dist.Backend.register_backend("fake", _create_fake_pg, devices=["cpu", "cuda", "hpu"])`