Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[Core][Distributed] fix pynccl del error #4508

Merged
merged 1 commit into from
May 1, 2024
Merged

[Core][Distributed] fix pynccl del error #4508

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 4 additions & 8 deletions vllm/distributed/device_communicators/pynccl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -200,6 +200,10 @@ def from_torch(cls, op: ReduceOp) -> int:
ncclDataType_t, ctypes.c_void_p, ctypes.c_void_p
]

# be cautious! this is a collective call, it will block until all
# processes in the communicator have called this function.
# because Python object destruction can happen in random order,
# it is better not to call it at all.
# equivalent to c declaration:
# ncclResult_t ncclCommDestroy(ncclComm_t comm);
_c_ncclCommDestroy = nccl.ncclCommDestroy
Expand Down Expand Up @@ -278,11 +282,3 @@ def all_reduce(self,
ncclDataTypeEnum.from_torch(tensor.dtype),
ncclRedOpTypeEnum.from_torch(op), self.comm,
ctypes.c_void_p(stream.cuda_stream)))

def __del__(self):
# `dist` module might have been already destroyed
if hasattr(dist, 'destroy_process_group'):
dist.destroy_process_group()
# function might have been already destroyed
if _c_ncclCommDestroy is not None:
_c_ncclCommDestroy(self.comm)
Loading