xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 06:55:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[misc] instruct pytorch to use nvml-based cuda check (#15951)

Browse Source 
Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
youkaichao 2025-04-03 01:02:58 +08:00 committed by GitHub
parent 8bd651b318
commit 1cab43c2d2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 25 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
vllm/__init__.py
View File

@ -4,9 +4,10 @@
# version library first. Such assumption is critical for some customization. # version library first. Such assumption is critical for some customization.
from .version import __version__, __version_tuple__ # isort:skip from .version import __version__, __version_tuple__ # isort:skip
import os # The environment variables override should be imported before any other
# modules to ensure that the environment variables are set before any
import torch # other modules are imported.
import vllm.env_override # isort:skip # noqa: F401
from vllm.engine.arg_utils import AsyncEngineArgs, EngineArgs from vllm.engine.arg_utils import AsyncEngineArgs, EngineArgs
from vllm.engine.async_llm_engine import AsyncLLMEngine from vllm.engine.async_llm_engine import AsyncLLMEngine
@ -23,19 +24,6 @@ from vllm.outputs import (ClassificationOutput, ClassificationRequestOutput,
from vllm.pooling_params import PoolingParams from vllm.pooling_params import PoolingParams
from vllm.sampling_params import SamplingParams from vllm.sampling_params import SamplingParams
# 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.
# see https://github.com/NVIDIA/nccl/issues/1234
os.environ['NCCL_CUMEM_ENABLE'] = '0'
# see https://github.com/vllm-project/vllm/issues/10480
os.environ['TORCHINDUCTOR_COMPILE_THREADS'] = '1'
# see https://github.com/vllm-project/vllm/issues/10619
torch._inductor.config.compile_threads = 1
__all__ = [ __all__ = [
"__version__", "__version__",
"__version_tuple__", "__version_tuple__",

21
vllm/env_override.py Normal file
View File

@ -0,0 +1,21 @@
# SPDX-License-Identifier: Apache-2.0
import os
import torch
# 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.
# see https://github.com/NVIDIA/nccl/issues/1234
os.environ['NCCL_CUMEM_ENABLE'] = '0'
# see https://github.com/vllm-project/vllm/pull/15951
# it avoids unintentional cuda initialization from torch.cuda.is_available()
os.environ['PYTORCH_NVML_BASED_CUDA_CHECK'] = '1'
# see https://github.com/vllm-project/vllm/issues/10480
os.environ['TORCHINDUCTOR_COMPILE_THREADS'] = '1'
# see https://github.com/vllm-project/vllm/issues/10619
torch._inductor.config.compile_threads = 1