xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 04:34:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Misc] Do not override NCCL_CUMEM_ENABLE if set explicitly (#19105)

Browse Source 
Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com>
This commit is contained in:
22quinn 2025-06-05 04:14:32 -07:00 committed by GitHub
parent 18093084be
commit 188a4590d8
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 12 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
vllm/env_override.py
View File

@ -4,17 +4,23 @@ import os
import torch
from vllm.logger import init_logger
logger = init_logger(__name__)
# set some common config/environment variables that should be set
# for all processes created by vllm and all processes
# that interact with vllm workers.
# they are executed whenever `import vllm` is called.
if not os.path.exists('/dev/nvidia-caps-imex-channels'):
# normally, we disable NCCL_CUMEM_ENABLE because it
# will cost 1~2 GiB GPU memory with cudagraph+allreduce,
# see https://github.com/NVIDIA/nccl/issues/1234
# for more details.
# However, NCCL requires NCCL_CUMEM_ENABLE to work with
if 'NCCL_CUMEM_ENABLE' in os.environ:
logger.warning(
"NCCL_CUMEM_ENABLE is set to %s, skipping override. "
"This may increase memory overhead with cudagraph+allreduce: "
"https://github.com/NVIDIA/nccl/issues/1234",
os.environ['NCCL_CUMEM_ENABLE'])
elif not os.path.exists('/dev/nvidia-caps-imex-channels'):
# NCCL requires NCCL_CUMEM_ENABLE to work with
# multi-node NVLink, typically on GB200-NVL72 systems.
# The ultimate way to detect multi-node NVLink is to use
# NVML APIs, which are too expensive to call here.