Skip to content

[BugFix] Pop instead of del CUDA_VISIBLE_DEVICES #12962

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
merged 1 commit into from
Feb 11, 2025
Merged

[BugFix] Pop instead of del CUDA_VISIBLE_DEVICES #12962

Show file tree
Hide file tree
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
2 changes: 1 addition & 1 deletion examples/offline_inference/rlhf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,7 +92,7 @@ def __init__(self, *args, **kwargs):
# a hack to make the script work.
# stop ray from manipulating CUDA_VISIBLE_DEVICES
# at the top-level
del os.environ["CUDA_VISIBLE_DEVICES"]
os.environ.pop("CUDA_VISIBLE_DEVICES", None)
super().__init__(*args, **kwargs)


Expand Down
2 changes: 1 addition & 1 deletion examples/offline_inference/rlhf_colocate.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,7 +59,7 @@ def __init__(self, *args, bundle_indices: list, **kwargs):
# a hack to make the script work.
# stop ray from manipulating CUDA_VISIBLE_DEVICES
# at the top-level
del os.environ["CUDA_VISIBLE_DEVICES"]
os.environ.pop("CUDA_VISIBLE_DEVICES", None)
# every worker will use 0.4 GPU, so that we can schedule
# 2 instances on the same GPUs.
os.environ["VLLM_RAY_PER_WORKER_GPUS"] = "0.4"
Expand Down
10 changes: 5 additions & 5 deletions tests/distributed/test_comm_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -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,
Expand All @@ -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,
Expand Down Expand Up @@ -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,
Expand Down Expand Up @@ -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,
Expand Down Expand Up @@ -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,
Expand Down
4 changes: 2 additions & 2 deletions tests/distributed/test_custom_all_reduce.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,7 @@

@ray.remote(num_gpus=1, max_calls=1)
def graph_allreduce(tp_size, pp_size, rank, distributed_init_port):
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,
Expand Down Expand Up @@ -80,7 +80,7 @@ def graph_allreduce(tp_size, pp_size, rank, distributed_init_port):

@ray.remote(num_gpus=1, max_calls=1)
def eager_allreduce(tp_size, pp_size, rank, distributed_init_port):
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,
Expand Down