Skip to content

Commit 0ff4cc8

Browse files
committed
[dtensor] disable gpu tests in op db first
There seems to be some issue with the cuda tests where our CI aren't capturing those failures (probably because of lacking 4 GPUs in CI environment). Disabling it first and debug later see #92343 [ghstack-poisoned]
1 parent 0a404fd commit 0ff4cc8

File tree

1 file changed

+3
-1
lines changed

1 file changed

+3
-1
lines changed

test/distributed/_tensor/test_dtensor_ops.py

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -563,7 +563,9 @@ def wrapped(fn):
563563

564564

565565
OP_DB_WORLD_SIZE = 4
566-
DEVICE_TYPE = "cuda" if torch.cuda.is_available() and torch.cuda.device_count() >= OP_DB_WORLD_SIZE else "cpu"
566+
# DEVICE_TYPE = "cuda" if torch.cuda.is_available() and torch.cuda.device_count() >= OP_DB_WORLD_SIZE else "cpu"
567+
# TODO: debug cuda illegal memory access issue and re-enable cuda tests
568+
DEVICE_TYPE = "cpu"
567569

568570

569571
class TestDTensorOps(DTensorOpTestBase):

0 commit comments

Comments
 (0)