biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[BugFix] Pop instead of del CUDA_VISIBLE_DEVICES (#12962)

Browse Source 
Signed-off-by: Hollow Man <hollowman@opensuse.org>
This commit is contained in:
ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟
2025-02-11 18:21:50 +02:00
committed by GitHub
parent 21f5d50fa5
commit 6c4dbe23eb
4 changed files with 9 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
tests/distributed/test_comm_ops.py
View File

@@ -22,7 +22,7 @@ def all_reduce_test_worker(tp_size: int, pp_size: int, rank: int,
# it is important to delete the CUDA_VISIBLE_DEVICES environment variable
# so that each worker can see all the GPUs
# they will be able to set the device to the correct GPU
del os.environ["CUDA_VISIBLE_DEVICES"]
os.environ.pop("CUDA_VISIBLE_DEVICES", None)
device = torch.device(f"cuda:{rank}")
torch.cuda.set_device(device)
init_test_distributed_environment(tp_size, pp_size, rank,
@@ -44,7 +44,7 @@ def all_gather_test_worker(tp_size: int, pp_size: int, rank: int,
# it is important to delete the CUDA_VISIBLE_DEVICES environment variable
# so that each worker can see all the GPUs
# they will be able to set the device to the correct GPU
del os.environ["CUDA_VISIBLE_DEVICES"]
os.environ.pop("CUDA_VISIBLE_DEVICES", None)
device = torch.device(f"cuda:{rank}")
torch.cuda.set_device(device)
init_test_distributed_environment(tp_size, pp_size, rank,
@@ -72,7 +72,7 @@ def broadcast_tensor_dict_test_worker(tp_size: int, pp_size: int, rank: int,
# it is important to delete the CUDA_VISIBLE_DEVICES environment variable
# so that each worker can see all the GPUs
# they will be able to set the device to the correct GPU
del os.environ["CUDA_VISIBLE_DEVICES"]
os.environ.pop("CUDA_VISIBLE_DEVICES", None)
device = torch.device(f"cuda:{rank}")
torch.cuda.set_device(device)
init_test_distributed_environment(tp_size, pp_size, rank,
@@ -108,7 +108,7 @@ def broadcast_tensor_dict_test_worker(tp_size: int, pp_size: int, rank: int,
@ray.remote(num_gpus=1, max_calls=1)
def send_recv_tensor_dict_test_worker(tp_size: int, pp_size: int, rank: int,
distributed_init_port: str):
del os.environ["CUDA_VISIBLE_DEVICES"]
os.environ.pop("CUDA_VISIBLE_DEVICES", None)
device = torch.device(f"cuda:{rank}")
torch.cuda.set_device(device)
init_test_distributed_environment(tp_size, pp_size, rank,
@@ -148,7 +148,7 @@ def send_recv_tensor_dict_test_worker(tp_size: int, pp_size: int, rank: int,
@ray.remote(num_gpus=1, max_calls=1)
def send_recv_test_worker(tp_size: int, pp_size: int, rank: int,
distributed_init_port: str):
del os.environ["CUDA_VISIBLE_DEVICES"]
os.environ.pop("CUDA_VISIBLE_DEVICES", None)
device = torch.device(f"cuda:{rank}")
torch.cuda.set_device(device)
init_test_distributed_environment(tp_size, pp_size, rank,