From 0f1f990e105cb85e3bbfe3433db5106335dd0ae2 Mon Sep 17 00:00:00 2001 From: yoshikisd Date: Sat, 8 Nov 2025 23:33:05 +0000 Subject: [PATCH] Added a check for init_method in multigpu.setup --- src/cdtools/tools/multigpu/multigpu.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/src/cdtools/tools/multigpu/multigpu.py b/src/cdtools/tools/multigpu/multigpu.py index e176b1f..13ac4e6 100644 --- a/src/cdtools/tools/multigpu/multigpu.py +++ b/src/cdtools/tools/multigpu/multigpu.py @@ -227,7 +227,7 @@ def setup(rank: int = None, 'these parameters have been explicitly defined, or ', 'alternatively launch the multi-GPU job with torchrun.' ) - else: + elif init_method == 'env://': # Set up the environment variables os.environ['MASTER_ADDR'] = master_addr os.environ['MASTER_PORT'] = master_port @@ -240,13 +240,13 @@ def setup(rank: int = None, t.cuda.set_device(rank) if rank == 0: print('[INFO]: Initializing process group.') - + dist.init_process_group(rank=rank, world_size=world_size, backend=backend, init_method=init_method, timeout=datetime.timedelta(timeout)) - + if rank == 0: print('[INFO]: Process group initialized.')