Skip to content

Handle special case of libuv for Windows #7064

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

Merged
loadams merged 5 commits into from
Feb 21, 2025
Merged

Handle special case of libuv for Windows #7064

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
5388e5d
Handle special case of libuv for Windows
loadams Feb 20, 2025
a6a49d4
Merge branch 'master' into loadams/libuv-windows
loadams Feb 20, 2025
cad0174
typo in varaible name
loadams Feb 20, 2025
5778af9
use_libuv is only added in torch 2.4+
loadams Feb 20, 2025
1ffdc44
missed not in if statement
loadams Feb 20, 2025
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
21 changes: 16 additions & 5 deletions deepspeed/comm/torch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -145,11 +145,22 @@ def has_reduce_scatter_tensor(self):

def init_process_group(self, backend, timeout, init_method, rank, world_size):
if not torch.distributed.is_initialized():
torch.distributed.init_process_group(backend,
timeout=timeout,
init_method=init_method,
rank=rank,
world_size=world_size)
if not required_torch_version(min_version=2.4):
# Windows torch builds do not come with lib_uv by default.
# More information here: https://pytorch.org/tutorials/intermediate/TCPStore_libuv_backend.html
use_libuv = False if os.name == "nt" else True
torch.distributed.init_process_group(backend,
timeout=timeout,
init_method=init_method,
rank=rank,
world_size=world_size,
use_libuv=use_libuv)
else:
torch.distributed.init_process_group(backend,
timeout=timeout,
init_method=init_method,
rank=rank,
world_size=world_size)
self.using_mpi = torch.distributed.get_backend() == 'mpi'

@disable_compiler_collective
Expand Down