frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-21 05:34:18 +08:00
Code Issues Packages Projects Releases Wiki Activity

Revert "Standardize on error types for distributed errors. (#107651)"

Browse Source 
This reverts commit 0e2317479b3cb987e1f3230876654f156bd11a09.

Reverted https://github.com/pytorch/pytorch/pull/107651 on behalf of https://github.com/huydhn due to Sorry for reverting your change, but it is failing inductor test in trunk for one of its model moco ([comment](https://github.com/pytorch/pytorch/pull/107651#issuecomment-1696578138))
This commit is contained in:
PyTorch MergeBot
2023-08-28 23:58:33 +00:00
parent cd4f74fb2e
commit d4ff06ec84
28 changed files with 240 additions and 265 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
torch/testing/_internal/distributed/distributed_test.py
View File

@ -713,7 +713,7 @@ class DistributedTest:
self.assertEqual(dist.get_backend(group_id), backend_str)
else:
with self.assertRaisesRegex(
ValueError, "Invalid process group specified"
RuntimeError, "Invalid process group specified"
):
dist.get_backend(group_id)
@ -970,7 +970,7 @@ class DistributedTest:
group, group_id, rank = self._init_global_test()
with self.assertRaisesRegex(
ValueError,
RuntimeError,
"The new group's rank should be within the the world_size set by init_process_group",
):
dist.new_subgroups_by_enumeration(
@ -1492,7 +1492,7 @@ class DistributedTest:
if rank == 0:
rank_to_GPU = init_multigpu_helper(dist.get_world_size(), BACKEND)
device_id = rank_to_GPU[rank][0]
with self.assertRaisesRegex(ValueError, "^Invalid ``op``"):
with self.assertRaisesRegex(RuntimeError, "^Invalid ``op``"):
send_tensor = _build_tensor(rank + 1, device_id=device_id)
send_op = dist.P2POp(dist.broadcast, send_tensor, 1)
dist.batch_isend_irecv([send_op])
@ -1504,7 +1504,7 @@ class DistributedTest:
self._barrier()
rank = dist.get_rank()
if rank == 0:
with self.assertRaisesRegex(ValueError, "^Invalid ``p2p_op_list``"):
with self.assertRaisesRegex(RuntimeError, "^Invalid ``p2p_op_list``"):
dist.batch_isend_irecv([1, 2])
# NCCL Batch SEND RECV Mixed Backend Error
@ -1519,7 +1519,7 @@ class DistributedTest:
group_nccl = dist.new_group(ranks=[0, 1], backend="nccl")
if rank == 0:
with self.assertRaisesRegex(
ValueError, "All ops need to use the same group"
RuntimeError, "All ops need to use the same group"
):
send_tensor = _build_tensor(rank + 1)
send_op_gloo = dist.P2POp(dist.isend, send_tensor, 1, group_gloo)
@ -2772,7 +2772,7 @@ class DistributedTest:
group, group_id, rank = self._init_global_test()
for unsupported_op in unsupported_ops:
with self.assertRaisesRegex(
ValueError, "all_reduce does not support"
RuntimeError, "all_reduce does not support"
):
dist.all_reduce(
_build_tensor(1, dtype=torch.cfloat), unsupported_op, group_id
@ -3004,7 +3004,7 @@ class DistributedTest:
)
def test_all_reduce_coalesced_max_complex_unsupported(self):
group, group_id, rank = self._init_global_test()
with self.assertRaisesRegex(ValueError, "all_reduce does not support"):
with self.assertRaisesRegex(RuntimeError, "all_reduce does not support"):
dist.all_reduce_coalesced(
[_build_tensor(1, dtype=torch.cfloat)], dist.ReduceOp.MAX, group_id
)
@ -8306,7 +8306,7 @@ class DistributedTest:
)
# Ensure errors are raised upon incorrect arguments.
with self.assertRaisesRegex(
ValueError,
RuntimeError,
"Expected argument scatter_object_output_list to be a list of size at least 1.",
):
dist.scatter_object_list([], scatter_list, src=src_rank)